Hugging Face's logo

AmirHussein
/
icefall-asr-mgb2-conformer_ctc-2022-27-06

Model card Files Community
AmirHussein commited on
Commit
647a6e0
•
1 Parent(s): d6e0834

.

Browse files
Files changed (11) hide show
  1. .gitattributes +1 -0
  2. decoding-results/log-attention-decoder/log-decode-2022-06-24-16-33-12 +52 -0
  3. decoding-results/log-attention-decoder/log-decode-2022-06-24-16-40-43 +0 -0
  4. decoding-results/log-attention-decoder/log-decode-2022-06-24-17-04-13 +1317 -0
  5. decoding-results/log-attention-decoder/log-decode-2022-06-24-17-22-16 +1428 -0
  6. decoding-results/log-attention-decoder/log-decode-2022-06-27-18-54-02 +6 -0
  7. decoding-results/log-attention-decoder/log-decode-2022-06-27-19-04-48 +1308 -0
  8. decoding-results/log-whole-lattice-rescoring/log-decode-2022-06-26-22-37-17 +202 -0
  9. decoding-results/log-whole-lattice-rescoring/log-decode-2022-06-26-23-11-51 +202 -0
  10. decoding-results/log-whole-lattice-rescoring/log-decode-2022-06-26-23-21-46 +0 -0
  11. decoding-results/log-whole-lattice-rescoring/log-decode-2022-06-27-18-46-45 +299 -0
.gitattributes CHANGED
@@ -38,3 +38,4 @@ data/lang_bpe_5000/*.pt filter=lfs diff=lfs merge=lfs -text
38
  data/lang_bpe_5000/*.vocab filter=lfs diff=lfs merge=lfs -text
39
  data/lang_bpe_5000/*.arpa filter=lfs diff=lfs merge=lfs -text
40
  data/lang_bpe_5000/*.txt filter=lfs diff=lfs merge=lfs -text
 
 
38
  data/lang_bpe_5000/*.vocab filter=lfs diff=lfs merge=lfs -text
39
  data/lang_bpe_5000/*.arpa filter=lfs diff=lfs merge=lfs -text
40
  data/lang_bpe_5000/*.txt filter=lfs diff=lfs merge=lfs -text
41
+ decoding-results/* filter=lfs diff=lfs merge=lfs -text
decoding-results/log-attention-decoder/log-decode-2022-06-24-16-33-12 ADDED
@@ -0,0 +1,52 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2022-06-24 16:33:12,810 INFO [decode.py:548] Decoding started
2
+ 2022-06-24 16:33:12,810 INFO [decode.py:549] {'subsampling_factor': 4, 'vgg_frontend': False, 'use_feat_batchnorm': True, 'feature_dim': 80, 'nhead': 8, 'attention_dim': 512, 'num_decoder_layers': 6, 'search_beam': 20, 'output_beam': 8, 'min_active_states': 30, 'max_active_states': 10000, 'use_double_scores': True, 'env_info': {'k2-version': '1.11', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '', 'k2-git-date': '', 'lhotse-version': '1.3.0.dev+git.a07121a.clean', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'test', 'icefall-git-sha1': '7e72d78-dirty', 'icefall-git-date': 'Sat May 28 19:13:53 2022', 'icefall-path': '/alt-arabic/speech/amir/k2/tmp/icefall', 'k2-path': '/home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/k2/__init__.py', 'lhotse-path': '/alt-arabic/speech/amir/k2/tmp/lhotse/lhotse/__init__.py', 'hostname': 'crimv3mgpu003', 'IP address': '10.141.0.1'}, 'epoch': 39, 'avg': 10, 'method': 'attention-decoder', 'num_paths': 1000, 'nbest_scale': 0.5, 'exp_dir': PosixPath('conformer_ctc/exp_5000_att0.8'), 'lang_dir': PosixPath('data/lang_bpe_5000'), 'lm_dir': PosixPath('data/lm'), 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 20, 'bucketing_sampler': True, 'num_buckets': 30, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'on_the_fly_feats': False, 'shuffle': False, 'drop_last': True, 'return_cuts': True, 'num_workers': 8, 'enable_spec_aug': False, 'spec_aug_time_warp_factor': 80, 'enable_musan': False}
3
+ 2022-06-24 16:33:13,088 INFO [lexicon.py:177] Loading pre-compiled data/lang_bpe_5000/Linv.pt
4
+ 2022-06-24 16:33:13,121 INFO [decode.py:559] device: cuda:0
5
+ 2022-06-24 16:33:17,106 INFO [decode.py:621] Loading pre-compiled G_4_gram.pt
6
+ 2022-06-24 16:33:17,873 INFO [decode.py:657] averaging ['conformer_ctc/exp_5000_att0.8/epoch-30.pt', 'conformer_ctc/exp_5000_att0.8/epoch-31.pt', 'conformer_ctc/exp_5000_att0.8/epoch-32.pt', 'conformer_ctc/exp_5000_att0.8/epoch-33.pt', 'conformer_ctc/exp_5000_att0.8/epoch-34.pt', 'conformer_ctc/exp_5000_att0.8/epoch-35.pt', 'conformer_ctc/exp_5000_att0.8/epoch-36.pt', 'conformer_ctc/exp_5000_att0.8/epoch-37.pt', 'conformer_ctc/exp_5000_att0.8/epoch-38.pt', 'conformer_ctc/exp_5000_att0.8/epoch-39.pt']
7
+ 2022-06-24 16:37:06,479 INFO [decode.py:664] Number of model parameters: 90786736
8
+ 2022-06-24 16:37:06,480 INFO [asr_datamodule.py:374] About to get test cuts
9
+ 2022-06-24 16:37:06,523 INFO [asr_datamodule.py:367] About to get dev cuts
10
+ 2022-06-24 16:37:11,701 INFO [decode.py:483] batch 0/?, cuts processed until now is 2
11
+ 2022-06-24 16:37:57,309 INFO [decode.py:733] Caught exception:
12
+ CUDA out of memory. Tried to allocate 452.00 MiB (GPU 0; 15.90 GiB total capacity; 13.98 GiB already allocated; 181.75 MiB free; 14.89 GiB reserved in total by PyTorch)
13
+ Exception raised from malloc at /opt/conda/conda-bld/pytorch_1616554788289/work/c10/cuda/CUDACachingAllocator.cpp:288 (most recent call first):
14
+ frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x42 (0x2aab1a77c2f2 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10.so)
15
+ frame #1: <unknown function> + 0x1bc21 (0x2aab1a518c21 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
16
+ frame #2: <unknown function> + 0x1c944 (0x2aab1a519944 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
17
+ frame #3: <unknown function> + 0x1cf63 (0x2aab1a519f63 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
18
+ frame #4: k2::PytorchCudaContext::Allocate(unsigned long, void**) + 0x5e (0x2aab381f0b0e in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/libk2context.so)
19
+ frame #5: k2::NewRegion(std::shared_ptr<k2::Context>, unsigned long) + 0x11e (0x2aab37f2099e in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/libk2context.so)
20
+ frame #6: k2::Array1<int>::Init(std::shared_ptr<k2::Context>, int, k2::Dtype) + 0xa9 (0x2aab37efba69 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/libk2context.so)
21
+ frame #7: <unknown function> + 0x1dfd1d (0x2aab38018d1d in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/libk2context.so)
22
+ frame #8: k2::Ragged<k2::Arc> k2::DeviceIntersector::FormatOutputTpl<k2::Hash::PackedAccessor>(k2::Array1<int>*, k2::Array1<int>*) + 0x786 (0x2aab38026836 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/libk2context.so)
23
+ frame #9: k2::IntersectDevice(k2::Ragged<k2::Arc>&, int, k2::Ragged<k2::Arc>&, int, k2::Array1<int> const&, k2::Array1<int>*, k2::Array1<int>*, bool) + 0x328 (0x2aab38019108 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/libk2context.so)
24
+ frame #10: <unknown function> + 0x70d49 (0x2aab33cabd49 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/_k2.cpython-38-x86_64-linux-gnu.so)
25
+ frame #11: <unknown function> + 0x240e3 (0x2aab33c5f0e3 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/_k2.cpython-38-x86_64-linux-gnu.so)
26
+ <omitting python frames>
27
+ frame #41: __libc_start_main + 0xf5 (0x2aaaab616555 in /lib64/libc.so.6)
28
+
29
+
30
+ 2022-06-24 16:37:57,310 INFO [decode.py:734] num_arcs before pruning: 1010153
31
+ 2022-06-24 16:37:57,310 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
32
+ 2022-06-24 16:37:57,329 INFO [decode.py:747] num_arcs after pruning: 2602
33
+ 2022-06-24 16:38:39,372 INFO [decode.py:733] Caught exception:
34
+ CUDA out of memory. Tried to allocate 1.85 GiB (GPU 0; 15.90 GiB total capacity; 12.31 GiB already allocated; 419.75 MiB free; 14.66 GiB reserved in total by PyTorch)
35
+ Exception raised from malloc at /opt/conda/conda-bld/pytorch_1616554788289/work/c10/cuda/CUDACachingAllocator.cpp:288 (most recent call first):
36
+ frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x42 (0x2aab1a77c2f2 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10.so)
37
+ frame #1: <unknown function> + 0x1bc21 (0x2aab1a518c21 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
38
+ frame #2: <unknown function> + 0x1c944 (0x2aab1a519944 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
39
+ frame #3: <unknown function> + 0x1cf63 (0x2aab1a519f63 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
40
+ frame #4: k2::PytorchCudaContext::Allocate(unsigned long, void**) + 0x5e (0x2aab381f0b0e in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/libk2context.so)
41
+ frame #5: k2::NewRegion(std::shared_ptr<k2::Context>, unsigned long) + 0x11e (0x2aab37f2099e in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/libk2context.so)
42
+ frame #6: k2::Ragged<k2::Arc> k2::DeviceIntersector::FormatOutputTpl<k2::Hash::PackedAccessor>(k2::Array1<int>*, k2::Array1<int>*) + 0x6ab (0x2aab3802675b in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/libk2context.so)
43
+ frame #7: k2::IntersectDevice(k2::Ragged<k2::Arc>&, int, k2::Ragged<k2::Arc>&, int, k2::Array1<int> const&, k2::Array1<int>*, k2::Array1<int>*, bool) + 0x328 (0x2aab38019108 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/libk2context.so)
44
+ frame #8: <unknown function> + 0x70d49 (0x2aab33cabd49 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/_k2.cpython-38-x86_64-linux-gnu.so)
45
+ frame #9: <unknown function> + 0x240e3 (0x2aab33c5f0e3 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/_k2.cpython-38-x86_64-linux-gnu.so)
46
+ <omitting python frames>
47
+ frame #39: __libc_start_main + 0xf5 (0x2aaaab616555 in /lib64/libc.so.6)
48
+
49
+
50
+ 2022-06-24 16:38:39,372 INFO [decode.py:734] num_arcs before pruning: 735461
51
+ 2022-06-24 16:38:39,372 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
52
+ 2022-06-24 16:38:39,387 INFO [decode.py:747] num_arcs after pruning: 3691
decoding-results/log-attention-decoder/log-decode-2022-06-24-16-40-43 ADDED
The diff for this file is too large to render. See raw diff
 
decoding-results/log-attention-decoder/log-decode-2022-06-24-17-04-13 ADDED
@@ -0,0 +1,1317 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2022-06-24 17:04:13,754 INFO [decode.py:548] Decoding started
2
+ 2022-06-24 17:04:13,754 INFO [decode.py:549] {'subsampling_factor': 4, 'vgg_frontend': False, 'use_feat_batchnorm': True, 'feature_dim': 80, 'nhead': 8, 'attention_dim': 512, 'num_decoder_layers': 6, 'search_beam': 20, 'output_beam': 8, 'min_active_states': 30, 'max_active_states': 10000, 'use_double_scores': True, 'env_info': {'k2-version': '1.11', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '', 'k2-git-date': '', 'lhotse-version': '1.3.0.dev+git.a07121a.clean', 'torch-cuda-available': False, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'test', 'icefall-git-sha1': '7e72d78-dirty', 'icefall-git-date': 'Sat May 28 19:13:53 2022', 'icefall-path': '/alt-arabic/speech/amir/k2/tmp/icefall', 'k2-path': '/home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/k2/__init__.py', 'lhotse-path': '/alt-arabic/speech/amir/k2/tmp/lhotse/lhotse/__init__.py', 'hostname': 'crimv3mgpu003', 'IP address': '10.141.0.1'}, 'epoch': 39, 'avg': 10, 'method': 'attention-decoder', 'num_paths': 1000, 'nbest_scale': 0.5, 'exp_dir': PosixPath('conformer_ctc/exp_5000_att0.8'), 'lang_dir': PosixPath('data/lang_bpe_5000'), 'lm_dir': PosixPath('data/lm'), 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 20, 'bucketing_sampler': True, 'num_buckets': 30, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'on_the_fly_feats': False, 'shuffle': False, 'drop_last': True, 'return_cuts': True, 'num_workers': 20, 'enable_spec_aug': False, 'spec_aug_time_warp_factor': 80, 'enable_musan': False}
3
+ 2022-06-24 17:04:14,031 INFO [lexicon.py:177] Loading pre-compiled data/lang_bpe_5000/Linv.pt
4
+ 2022-06-24 17:04:14,064 INFO [decode.py:559] device: cpu
5
+ 2022-06-24 17:04:19,929 INFO [decode.py:621] Loading pre-compiled G_4_gram.pt
6
+ 2022-06-24 17:04:23,746 INFO [decode.py:657] averaging ['conformer_ctc/exp_5000_att0.8/epoch-30.pt', 'conformer_ctc/exp_5000_att0.8/epoch-31.pt', 'conformer_ctc/exp_5000_att0.8/epoch-32.pt', 'conformer_ctc/exp_5000_att0.8/epoch-33.pt', 'conformer_ctc/exp_5000_att0.8/epoch-34.pt', 'conformer_ctc/exp_5000_att0.8/epoch-35.pt', 'conformer_ctc/exp_5000_att0.8/epoch-36.pt', 'conformer_ctc/exp_5000_att0.8/epoch-37.pt', 'conformer_ctc/exp_5000_att0.8/epoch-38.pt', 'conformer_ctc/exp_5000_att0.8/epoch-39.pt']
7
+ 2022-06-24 17:04:30,866 INFO [decode.py:664] Number of model parameters: 90786736
8
+ 2022-06-24 17:04:30,866 INFO [asr_datamodule.py:374] About to get test cuts
9
+ 2022-06-24 17:04:30,868 INFO [asr_datamodule.py:367] About to get dev cuts
10
+ 2022-06-24 17:04:41,864 INFO [decode.py:483] batch 0/?, cuts processed until now is 2
11
+ 2022-06-24 18:34:46,534 INFO [decode.py:483] batch 100/?, cuts processed until now is 250
12
+ 2022-06-24 18:47:04,383 INFO [decode.py:733] Caught exception:
13
+
14
+ Some bad things happened. Please read the above error messages and stack
15
+ trace. If you are using Python, the following command may be helpful:
16
+
17
+ gdb --args python /path/to/your/code.py
18
+
19
+ (You can use `gdb` to debug the code. Please consider compiling
20
+ a debug version of k2.).
21
+
22
+ If you are unable to fix it, please open an issue at:
23
+
24
+ https://github.com/k2-fsa/k2/issues/new
25
+
26
+
27
+ 2022-06-24 18:47:04,383 INFO [decode.py:734] num_arcs before pruning: 2310868
28
+ 2022-06-24 18:47:04,384 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
29
+ 2022-06-24 18:47:04,733 INFO [decode.py:747] num_arcs after pruning: 5012
30
+ 2022-06-24 20:10:31,760 INFO [decode.py:483] batch 200/?, cuts processed until now is 516
31
+ 2022-06-24 21:18:41,734 INFO [decode.py:733] Caught exception:
32
+
33
+ Some bad things happened. Please read the above error messages and stack
34
+ trace. If you are using Python, the following command may be helpful:
35
+
36
+ gdb --args python /path/to/your/code.py
37
+
38
+ (You can use `gdb` to debug the code. Please consider compiling
39
+ a debug version of k2.).
40
+
41
+ If you are unable to fix it, please open an issue at:
42
+
43
+ https://github.com/k2-fsa/k2/issues/new
44
+
45
+
46
+ 2022-06-24 21:18:41,735 INFO [decode.py:734] num_arcs before pruning: 1083317
47
+ 2022-06-24 21:18:41,736 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
48
+ 2022-06-24 21:18:41,887 INFO [decode.py:747] num_arcs after pruning: 12010
49
+ 2022-06-24 21:41:53,864 INFO [decode.py:483] batch 300/?, cuts processed until now is 791
50
+ 2022-06-24 23:01:45,578 INFO [decode.py:483] batch 400/?, cuts processed until now is 1047
51
+ 2022-06-25 00:48:28,171 INFO [decode.py:483] batch 500/?, cuts processed until now is 1298
52
+ 2022-06-25 01:51:39,834 INFO [decode.py:483] batch 600/?, cuts processed until now is 1572
53
+ 2022-06-25 02:49:30,229 INFO [decode.py:483] batch 700/?, cuts processed until now is 1843
54
+ 2022-06-25 03:53:42,536 INFO [decode.py:483] batch 800/?, cuts processed until now is 2097
55
+ 2022-06-25 05:16:00,106 INFO [decode.py:483] batch 900/?, cuts processed until now is 2366
56
+ 2022-06-25 06:57:14,063 INFO [decode.py:483] batch 1000/?, cuts processed until now is 2626
57
+ 2022-06-25 08:35:50,225 INFO [decode.py:483] batch 1100/?, cuts processed until now is 2887
58
+ 2022-06-25 10:09:36,309 INFO [decode.py:483] batch 1200/?, cuts processed until now is 3155
59
+ 2022-06-25 11:48:04,046 INFO [decode.py:483] batch 1300/?, cuts processed until now is 3410
60
+ 2022-06-25 13:12:06,293 INFO [decode.py:483] batch 1400/?, cuts processed until now is 3670
61
+ 2022-06-25 14:30:59,601 INFO [decode.py:483] batch 1500/?, cuts processed until now is 3933
62
+ 2022-06-25 15:45:48,988 INFO [decode.py:733] Caught exception:
63
+
64
+ Some bad things happened. Please read the above error messages and stack
65
+ trace. If you are using Python, the following command may be helpful:
66
+
67
+ gdb --args python /path/to/your/code.py
68
+
69
+ (You can use `gdb` to debug the code. Please consider compiling
70
+ a debug version of k2.).
71
+
72
+ If you are unable to fix it, please open an issue at:
73
+
74
+ https://github.com/k2-fsa/k2/issues/new
75
+
76
+
77
+ 2022-06-25 15:45:48,990 INFO [decode.py:734] num_arcs before pruning: 1453325
78
+ 2022-06-25 15:45:48,990 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
79
+ 2022-06-25 15:45:49,219 INFO [decode.py:747] num_arcs after pruning: 4879
80
+ 2022-06-25 16:01:22,610 INFO [decode.py:733] Caught exception:
81
+
82
+ Some bad things happened. Please read the above error messages and stack
83
+ trace. If you are using Python, the following command may be helpful:
84
+
85
+ gdb --args python /path/to/your/code.py
86
+
87
+ (You can use `gdb` to debug the code. Please consider compiling
88
+ a debug version of k2.).
89
+
90
+ If you are unable to fix it, please open an issue at:
91
+
92
+ https://github.com/k2-fsa/k2/issues/new
93
+
94
+
95
+ 2022-06-25 16:01:22,611 INFO [decode.py:734] num_arcs before pruning: 1195732
96
+ 2022-06-25 16:01:22,611 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
97
+ 2022-06-25 16:01:22,789 INFO [decode.py:747] num_arcs after pruning: 5734
98
+ 2022-06-25 16:06:11,843 INFO [decode.py:483] batch 1600/?, cuts processed until now is 4197
99
+ 2022-06-25 18:07:06,556 INFO [decode.py:483] batch 1700/?, cuts processed until now is 4440
100
+ 2022-06-25 19:32:17,442 INFO [decode.py:483] batch 1800/?, cuts processed until now is 4683
101
+ 2022-06-25 20:30:10,100 INFO [decode.py:483] batch 1900/?, cuts processed until now is 4911
102
+ 2022-06-25 21:27:10,127 INFO [decode.py:483] batch 2000/?, cuts processed until now is 5136
103
+ 2022-06-25 22:15:23,452 INFO [decode.py:483] batch 2100/?, cuts processed until now is 5362
104
+ 2022-06-25 22:20:43,949 INFO [decode.py:532]
105
+ For test, WER of different settings are:
106
+ ngram_lm_scale_0.01_attention_scale_0.5 14.97 best for test
107
+ ngram_lm_scale_0.05_attention_scale_0.5 14.98
108
+ ngram_lm_scale_0.01_attention_scale_0.3 14.99
109
+ ngram_lm_scale_0.01_attention_scale_0.6 15.0
110
+ ngram_lm_scale_0.01_attention_scale_0.7 15.02
111
+ ngram_lm_scale_0.05_attention_scale_0.3 15.02
112
+ ngram_lm_scale_0.05_attention_scale_0.6 15.02
113
+ ngram_lm_scale_0.08_attention_scale_0.5 15.02
114
+ ngram_lm_scale_0.1_attention_scale_0.5 15.04
115
+ ngram_lm_scale_0.08_attention_scale_0.6 15.05
116
+ ngram_lm_scale_0.05_attention_scale_0.7 15.06
117
+ ngram_lm_scale_0.08_attention_scale_0.3 15.07
118
+ ngram_lm_scale_0.1_attention_scale_0.3 15.07
119
+ ngram_lm_scale_0.01_attention_scale_0.9 15.08
120
+ ngram_lm_scale_0.08_attention_scale_0.7 15.08
121
+ ngram_lm_scale_0.1_attention_scale_0.6 15.08
122
+ ngram_lm_scale_0.01_attention_scale_0.1 15.09
123
+ ngram_lm_scale_0.05_attention_scale_0.9 15.09
124
+ ngram_lm_scale_0.01_attention_scale_1.0 15.1
125
+ ngram_lm_scale_0.05_attention_scale_0.1 15.1
126
+ ngram_lm_scale_0.1_attention_scale_0.7 15.1
127
+ ngram_lm_scale_0.01_attention_scale_0.08 15.11
128
+ ngram_lm_scale_0.05_attention_scale_1.0 15.11
129
+ ngram_lm_scale_0.08_attention_scale_0.9 15.11
130
+ ngram_lm_scale_0.01_attention_scale_1.1 15.12
131
+ ngram_lm_scale_0.05_attention_scale_0.08 15.13
132
+ ngram_lm_scale_0.05_attention_scale_1.1 15.13
133
+ ngram_lm_scale_0.08_attention_scale_1.0 15.13
134
+ ngram_lm_scale_0.1_attention_scale_0.9 15.13
135
+ ngram_lm_scale_0.01_attention_scale_0.05 15.14
136
+ ngram_lm_scale_0.1_attention_scale_1.0 15.14
137
+ ngram_lm_scale_0.01_attention_scale_1.2 15.15
138
+ ngram_lm_scale_0.08_attention_scale_0.08 15.15
139
+ ngram_lm_scale_0.08_attention_scale_0.1 15.15
140
+ ngram_lm_scale_0.01_attention_scale_1.3 15.16
141
+ ngram_lm_scale_0.08_attention_scale_1.1 15.16
142
+ ngram_lm_scale_0.1_attention_scale_0.1 15.16
143
+ ngram_lm_scale_0.05_attention_scale_0.05 15.17
144
+ ngram_lm_scale_0.05_attention_scale_1.2 15.17
145
+ ngram_lm_scale_0.1_attention_scale_0.08 15.17
146
+ ngram_lm_scale_0.1_attention_scale_1.1 15.17
147
+ ngram_lm_scale_0.05_attention_scale_1.3 15.18
148
+ ngram_lm_scale_0.08_attention_scale_0.05 15.19
149
+ ngram_lm_scale_0.08_attention_scale_1.2 15.19
150
+ ngram_lm_scale_0.1_attention_scale_1.2 15.2
151
+ ngram_lm_scale_0.01_attention_scale_0.01 15.21
152
+ ngram_lm_scale_0.01_attention_scale_1.5 15.21
153
+ ngram_lm_scale_0.08_attention_scale_1.3 15.21
154
+ ngram_lm_scale_0.1_attention_scale_0.05 15.21
155
+ ngram_lm_scale_0.05_attention_scale_1.5 15.22
156
+ ngram_lm_scale_0.1_attention_scale_1.3 15.22
157
+ ngram_lm_scale_0.01_attention_scale_1.7 15.23
158
+ ngram_lm_scale_0.05_attention_scale_1.7 15.25
159
+ ngram_lm_scale_0.08_attention_scale_1.5 15.25
160
+ ngram_lm_scale_0.01_attention_scale_1.9 15.26
161
+ ngram_lm_scale_0.05_attention_scale_0.01 15.26
162
+ ngram_lm_scale_0.1_attention_scale_1.5 15.26
163
+ ngram_lm_scale_0.01_attention_scale_2.0 15.27
164
+ ngram_lm_scale_0.08_attention_scale_0.01 15.27
165
+ ngram_lm_scale_0.3_attention_scale_0.7 15.27
166
+ ngram_lm_scale_0.3_attention_scale_0.9 15.27
167
+ ngram_lm_scale_0.1_attention_scale_0.01 15.28
168
+ ngram_lm_scale_0.3_attention_scale_1.0 15.28
169
+ ngram_lm_scale_0.08_attention_scale_1.7 15.29
170
+ ngram_lm_scale_0.3_attention_scale_0.6 15.29
171
+ ngram_lm_scale_0.05_attention_scale_1.9 15.3
172
+ ngram_lm_scale_0.1_attention_scale_1.7 15.3
173
+ ngram_lm_scale_0.3_attention_scale_0.5 15.3
174
+ ngram_lm_scale_0.01_attention_scale_2.1 15.31
175
+ ngram_lm_scale_0.01_attention_scale_2.2 15.31
176
+ ngram_lm_scale_0.05_attention_scale_2.0 15.31
177
+ ngram_lm_scale_0.08_attention_scale_1.9 15.31
178
+ ngram_lm_scale_0.3_attention_scale_1.1 15.31
179
+ ngram_lm_scale_0.01_attention_scale_2.3 15.32
180
+ ngram_lm_scale_0.05_attention_scale_2.1 15.32
181
+ ngram_lm_scale_0.05_attention_scale_2.2 15.33
182
+ ngram_lm_scale_0.05_attention_scale_2.3 15.33
183
+ ngram_lm_scale_0.08_attention_scale_2.0 15.33
184
+ ngram_lm_scale_0.1_attention_scale_1.9 15.33
185
+ ngram_lm_scale_0.1_attention_scale_2.0 15.33
186
+ ngram_lm_scale_0.01_attention_scale_2.5 15.34
187
+ ngram_lm_scale_0.08_attention_scale_2.1 15.34
188
+ ngram_lm_scale_0.1_attention_scale_2.1 15.34
189
+ ngram_lm_scale_0.3_attention_scale_1.2 15.34
190
+ ngram_lm_scale_0.08_attention_scale_2.2 15.35
191
+ ngram_lm_scale_0.1_attention_scale_2.2 15.35
192
+ ngram_lm_scale_0.05_attention_scale_2.5 15.36
193
+ ngram_lm_scale_0.08_attention_scale_2.3 15.36
194
+ ngram_lm_scale_0.1_attention_scale_2.3 15.36
195
+ ngram_lm_scale_0.3_attention_scale_1.3 15.36
196
+ ngram_lm_scale_0.08_attention_scale_2.5 15.37
197
+ ngram_lm_scale_0.3_attention_scale_0.3 15.38
198
+ ngram_lm_scale_0.01_attention_scale_3.0 15.39
199
+ ngram_lm_scale_0.1_attention_scale_2.5 15.39
200
+ ngram_lm_scale_0.05_attention_scale_3.0 15.41
201
+ ngram_lm_scale_0.3_attention_scale_1.5 15.42
202
+ ngram_lm_scale_0.08_attention_scale_3.0 15.45
203
+ ngram_lm_scale_0.1_attention_scale_3.0 15.46
204
+ ngram_lm_scale_0.3_attention_scale_1.7 15.47
205
+ ngram_lm_scale_0.3_attention_scale_1.9 15.48
206
+ ngram_lm_scale_0.3_attention_scale_2.1 15.48
207
+ ngram_lm_scale_0.01_attention_scale_4.0 15.49
208
+ ngram_lm_scale_0.3_attention_scale_2.0 15.49
209
+ ngram_lm_scale_0.05_attention_scale_4.0 15.5
210
+ ngram_lm_scale_0.3_attention_scale_2.2 15.5
211
+ ngram_lm_scale_0.08_attention_scale_4.0 15.52
212
+ ngram_lm_scale_0.3_attention_scale_2.3 15.52
213
+ ngram_lm_scale_0.1_attention_scale_4.0 15.53
214
+ ngram_lm_scale_0.3_attention_scale_0.1 15.55
215
+ ngram_lm_scale_0.3_attention_scale_2.5 15.55
216
+ ngram_lm_scale_0.01_attention_scale_5.0 15.56
217
+ ngram_lm_scale_0.05_attention_scale_5.0 15.57
218
+ ngram_lm_scale_0.08_attention_scale_5.0 15.59
219
+ ngram_lm_scale_0.1_attention_scale_5.0 15.59
220
+ ngram_lm_scale_0.3_attention_scale_0.08 15.59
221
+ ngram_lm_scale_0.3_attention_scale_3.0 15.6
222
+ ngram_lm_scale_0.5_attention_scale_1.5 15.62
223
+ ngram_lm_scale_0.5_attention_scale_1.1 15.63
224
+ ngram_lm_scale_0.5_attention_scale_1.3 15.63
225
+ ngram_lm_scale_0.3_attention_scale_4.0 15.64
226
+ ngram_lm_scale_0.5_attention_scale_1.2 15.64
227
+ ngram_lm_scale_0.5_attention_scale_0.9 15.65
228
+ ngram_lm_scale_0.5_attention_scale_1.0 15.65
229
+ ngram_lm_scale_0.5_attention_scale_1.7 15.66
230
+ ngram_lm_scale_0.3_attention_scale_5.0 15.67
231
+ ngram_lm_scale_0.5_attention_scale_1.9 15.67
232
+ ngram_lm_scale_0.5_attention_scale_2.0 15.67
233
+ ngram_lm_scale_0.5_attention_scale_2.1 15.68
234
+ ngram_lm_scale_0.5_attention_scale_2.2 15.68
235
+ ngram_lm_scale_0.5_attention_scale_2.3 15.68
236
+ ngram_lm_scale_0.3_attention_scale_0.05 15.69
237
+ ngram_lm_scale_0.5_attention_scale_2.5 15.69
238
+ ngram_lm_scale_0.5_attention_scale_0.7 15.71
239
+ ngram_lm_scale_0.5_attention_scale_3.0 15.71
240
+ ngram_lm_scale_0.5_attention_scale_4.0 15.76
241
+ ngram_lm_scale_0.5_attention_scale_0.6 15.77
242
+ ngram_lm_scale_0.5_attention_scale_5.0 15.79
243
+ ngram_lm_scale_0.6_attention_scale_2.0 15.81
244
+ ngram_lm_scale_0.3_attention_scale_0.01 15.82
245
+ ngram_lm_scale_0.6_attention_scale_1.9 15.82
246
+ ngram_lm_scale_0.6_attention_scale_2.1 15.82
247
+ ngram_lm_scale_0.6_attention_scale_2.2 15.82
248
+ ngram_lm_scale_0.6_attention_scale_2.3 15.82
249
+ ngram_lm_scale_0.6_attention_scale_2.5 15.82
250
+ ngram_lm_scale_0.6_attention_scale_3.0 15.82
251
+ ngram_lm_scale_0.6_attention_scale_1.5 15.83
252
+ ngram_lm_scale_0.5_attention_scale_0.5 15.84
253
+ ngram_lm_scale_0.6_attention_scale_1.7 15.84
254
+ ngram_lm_scale_0.6_attention_scale_1.1 15.85
255
+ ngram_lm_scale_0.6_attention_scale_4.0 15.85
256
+ ngram_lm_scale_0.6_attention_scale_1.2 15.86
257
+ ngram_lm_scale_0.6_attention_scale_1.3 15.86
258
+ ngram_lm_scale_0.6_attention_scale_5.0 15.86
259
+ ngram_lm_scale_0.6_attention_scale_1.0 15.88
260
+ ngram_lm_scale_0.6_attention_scale_0.9 15.92
261
+ ngram_lm_scale_0.7_attention_scale_4.0 15.92
262
+ ngram_lm_scale_0.7_attention_scale_5.0 15.92
263
+ ngram_lm_scale_0.7_attention_scale_2.5 15.96
264
+ ngram_lm_scale_0.7_attention_scale_3.0 15.96
265
+ ngram_lm_scale_0.7_attention_scale_2.3 15.99
266
+ ngram_lm_scale_0.7_attention_scale_2.2 16.0
267
+ ngram_lm_scale_0.7_attention_scale_2.1 16.01
268
+ ngram_lm_scale_0.5_attention_scale_0.3 16.02
269
+ ngram_lm_scale_0.7_attention_scale_2.0 16.02
270
+ ngram_lm_scale_0.7_attention_scale_1.9 16.03
271
+ ngram_lm_scale_0.7_attention_scale_1.7 16.04
272
+ ngram_lm_scale_0.7_attention_scale_1.5 16.06
273
+ ngram_lm_scale_0.6_attention_scale_0.7 16.07
274
+ ngram_lm_scale_0.7_attention_scale_1.3 16.1
275
+ ngram_lm_scale_0.9_attention_scale_5.0 16.11
276
+ ngram_lm_scale_0.6_attention_scale_0.6 16.13
277
+ ngram_lm_scale_0.7_attention_scale_1.2 16.14
278
+ ngram_lm_scale_0.9_attention_scale_4.0 16.16
279
+ ngram_lm_scale_0.7_attention_scale_1.1 16.18
280
+ ngram_lm_scale_1.0_attention_scale_5.0 16.19
281
+ ngram_lm_scale_0.7_attention_scale_1.0 16.21
282
+ ngram_lm_scale_0.6_attention_scale_0.5 16.24
283
+ ngram_lm_scale_0.9_attention_scale_3.0 16.26
284
+ ngram_lm_scale_1.0_attention_scale_4.0 16.26
285
+ ngram_lm_scale_1.1_attention_scale_5.0 16.28
286
+ ngram_lm_scale_0.7_attention_scale_0.9 16.29
287
+ ngram_lm_scale_0.9_attention_scale_2.5 16.32
288
+ ngram_lm_scale_0.9_attention_scale_2.3 16.37
289
+ ngram_lm_scale_0.9_attention_scale_2.1 16.39
290
+ ngram_lm_scale_0.9_attention_scale_2.2 16.39
291
+ ngram_lm_scale_1.0_attention_scale_3.0 16.39
292
+ ngram_lm_scale_1.1_attention_scale_4.0 16.39
293
+ ngram_lm_scale_1.2_attention_scale_5.0 16.41
294
+ ngram_lm_scale_0.9_attention_scale_2.0 16.42
295
+ ngram_lm_scale_0.9_attention_scale_1.9 16.45
296
+ ngram_lm_scale_1.0_attention_scale_2.5 16.47
297
+ ngram_lm_scale_0.7_attention_scale_0.7 16.5
298
+ ngram_lm_scale_1.3_attention_scale_5.0 16.5
299
+ ngram_lm_scale_1.1_attention_scale_3.0 16.53
300
+ ngram_lm_scale_1.2_attention_scale_4.0 16.53
301
+ ngram_lm_scale_1.0_attention_scale_2.3 16.55
302
+ ngram_lm_scale_0.9_attention_scale_1.7 16.56
303
+ ngram_lm_scale_1.0_attention_scale_2.2 16.59
304
+ ngram_lm_scale_0.7_attention_scale_0.6 16.61
305
+ ngram_lm_scale_1.0_attention_scale_2.1 16.63
306
+ ngram_lm_scale_0.5_attention_scale_0.1 16.66
307
+ ngram_lm_scale_0.9_attention_scale_1.5 16.67
308
+ ngram_lm_scale_0.6_attention_scale_0.3 16.68
309
+ ngram_lm_scale_1.0_attention_scale_2.0 16.68
310
+ ngram_lm_scale_1.3_attention_scale_4.0 16.69
311
+ ngram_lm_scale_1.0_attention_scale_1.9 16.73
312
+ ngram_lm_scale_1.1_attention_scale_2.5 16.73
313
+ ngram_lm_scale_0.5_attention_scale_0.08 16.76
314
+ ngram_lm_scale_1.5_attention_scale_5.0 16.78
315
+ ngram_lm_scale_1.2_attention_scale_3.0 16.79
316
+ ngram_lm_scale_0.7_attention_scale_0.5 16.84
317
+ ngram_lm_scale_0.9_attention_scale_1.3 16.84
318
+ ngram_lm_scale_1.1_attention_scale_2.3 16.84
319
+ ngram_lm_scale_1.0_attention_scale_1.7 16.86
320
+ ngram_lm_scale_1.1_attention_scale_2.2 16.89
321
+ ngram_lm_scale_0.5_attention_scale_0.05 16.93
322
+ ngram_lm_scale_0.9_attention_scale_1.2 16.94
323
+ ngram_lm_scale_1.1_attention_scale_2.1 16.98
324
+ ngram_lm_scale_1.3_attention_scale_3.0 17.02
325
+ ngram_lm_scale_1.1_attention_scale_2.0 17.03
326
+ ngram_lm_scale_1.2_attention_scale_2.5 17.03
327
+ ngram_lm_scale_1.5_attention_scale_4.0 17.03
328
+ ngram_lm_scale_1.7_attention_scale_5.0 17.03
329
+ ngram_lm_scale_0.9_attention_scale_1.1 17.05
330
+ ngram_lm_scale_1.0_attention_scale_1.5 17.05
331
+ ngram_lm_scale_1.1_attention_scale_1.9 17.11
332
+ ngram_lm_scale_1.2_attention_scale_2.3 17.16
333
+ ngram_lm_scale_0.9_attention_scale_1.0 17.2
334
+ ngram_lm_scale_0.5_attention_scale_0.01 17.21
335
+ ngram_lm_scale_1.2_attention_scale_2.2 17.22
336
+ ngram_lm_scale_1.0_attention_scale_1.3 17.29
337
+ ngram_lm_scale_1.2_attention_scale_2.1 17.29
338
+ ngram_lm_scale_1.1_attention_scale_1.7 17.3
339
+ ngram_lm_scale_1.3_attention_scale_2.5 17.3
340
+ ngram_lm_scale_0.9_attention_scale_0.9 17.36
341
+ ngram_lm_scale_1.9_attention_scale_5.0 17.38
342
+ ngram_lm_scale_1.2_attention_scale_2.0 17.41
343
+ ngram_lm_scale_1.0_attention_scale_1.2 17.42
344
+ ngram_lm_scale_1.7_attention_scale_4.0 17.46
345
+ ngram_lm_scale_1.3_attention_scale_2.3 17.5
346
+ ngram_lm_scale_1.1_attention_scale_1.5 17.52
347
+ ngram_lm_scale_1.2_attention_scale_1.9 17.52
348
+ ngram_lm_scale_0.7_attention_scale_0.3 17.55
349
+ ngram_lm_scale_1.0_attention_scale_1.1 17.55
350
+ ngram_lm_scale_1.3_attention_scale_2.2 17.58
351
+ ngram_lm_scale_2.0_attention_scale_5.0 17.59
352
+ ngram_lm_scale_1.5_attention_scale_3.0 17.61
353
+ ngram_lm_scale_0.6_attention_scale_0.1 17.66
354
+ ngram_lm_scale_1.3_attention_scale_2.1 17.71
355
+ ngram_lm_scale_1.0_attention_scale_1.0 17.75
356
+ ngram_lm_scale_1.2_attention_scale_1.7 17.76
357
+ ngram_lm_scale_1.1_attention_scale_1.3 17.82
358
+ ngram_lm_scale_0.6_attention_scale_0.08 17.84
359
+ ngram_lm_scale_2.1_attention_scale_5.0 17.84
360
+ ngram_lm_scale_1.3_attention_scale_2.0 17.85
361
+ ngram_lm_scale_0.9_attention_scale_0.7 17.88
362
+ ngram_lm_scale_1.9_attention_scale_4.0 17.98
363
+ ngram_lm_scale_1.3_attention_scale_1.9 17.99
364
+ ngram_lm_scale_1.1_attention_scale_1.2 18.05
365
+ ngram_lm_scale_1.5_attention_scale_2.5 18.05
366
+ ngram_lm_scale_2.2_attention_scale_5.0 18.06
367
+ ngram_lm_scale_1.0_attention_scale_0.9 18.08
368
+ ngram_lm_scale_1.2_attention_scale_1.5 18.08
369
+ ngram_lm_scale_0.6_attention_scale_0.05 18.1
370
+ ngram_lm_scale_1.7_attention_scale_3.0 18.27
371
+ ngram_lm_scale_1.3_attention_scale_1.7 18.29
372
+ ngram_lm_scale_2.0_attention_scale_4.0 18.29
373
+ ngram_lm_scale_2.3_attention_scale_5.0 18.29
374
+ ngram_lm_scale_0.9_attention_scale_0.6 18.3
375
+ ngram_lm_scale_1.5_attention_scale_2.3 18.33
376
+ ngram_lm_scale_1.1_attention_scale_1.1 18.36
377
+ ngram_lm_scale_0.6_attention_scale_0.01 18.45
378
+ ngram_lm_scale_1.5_attention_scale_2.2 18.51
379
+ ngram_lm_scale_1.2_attention_scale_1.3 18.57
380
+ ngram_lm_scale_2.1_attention_scale_4.0 18.58
381
+ ngram_lm_scale_1.5_attention_scale_2.1 18.67
382
+ ngram_lm_scale_1.1_attention_scale_1.0 18.71
383
+ ngram_lm_scale_2.5_attention_scale_5.0 18.74
384
+ ngram_lm_scale_0.9_attention_scale_0.5 18.8
385
+ ngram_lm_scale_1.3_attention_scale_1.5 18.82
386
+ ngram_lm_scale_2.2_attention_scale_4.0 18.86
387
+ ngram_lm_scale_1.5_attention_scale_2.0 18.9
388
+ ngram_lm_scale_1.0_attention_scale_0.7 18.92
389
+ ngram_lm_scale_1.2_attention_scale_1.2 18.92
390
+ ngram_lm_scale_0.7_attention_scale_0.1 18.95
391
+ ngram_lm_scale_1.7_attention_scale_2.5 19.01
392
+ ngram_lm_scale_1.9_attention_scale_3.0 19.09
393
+ ngram_lm_scale_1.1_attention_scale_0.9 19.1
394
+ ngram_lm_scale_0.7_attention_scale_0.08 19.17
395
+ ngram_lm_scale_1.5_attention_scale_1.9 19.2
396
+ ngram_lm_scale_2.3_attention_scale_4.0 19.21
397
+ ngram_lm_scale_1.2_attention_scale_1.1 19.3
398
+ ngram_lm_scale_1.7_attention_scale_2.3 19.46
399
+ ngram_lm_scale_1.0_attention_scale_0.6 19.47
400
+ ngram_lm_scale_0.7_attention_scale_0.05 19.5
401
+ ngram_lm_scale_1.3_attention_scale_1.3 19.5
402
+ ngram_lm_scale_2.0_attention_scale_3.0 19.61
403
+ ngram_lm_scale_1.2_attention_scale_1.0 19.67
404
+ ngram_lm_scale_1.7_attention_scale_2.2 19.76
405
+ ngram_lm_scale_1.5_attention_scale_1.7 19.78
406
+ ngram_lm_scale_1.3_attention_scale_1.2 19.83
407
+ ngram_lm_scale_1.7_attention_scale_2.1 19.96
408
+ ngram_lm_scale_0.7_attention_scale_0.01 20.02
409
+ ngram_lm_scale_1.0_attention_scale_0.5 20.1
410
+ ngram_lm_scale_2.5_attention_scale_4.0 20.1
411
+ ngram_lm_scale_1.1_attention_scale_0.7 20.15
412
+ ngram_lm_scale_1.2_attention_scale_0.9 20.19
413
+ ngram_lm_scale_2.1_attention_scale_3.0 20.19
414
+ ngram_lm_scale_1.9_attention_scale_2.5 20.2
415
+ ngram_lm_scale_1.3_attention_scale_1.1 20.28
416
+ ngram_lm_scale_1.7_attention_scale_2.0 20.28
417
+ ngram_lm_scale_0.9_attention_scale_0.3 20.31
418
+ ngram_lm_scale_3.0_attention_scale_5.0 20.41
419
+ ngram_lm_scale_1.5_attention_scale_1.5 20.49
420
+ ngram_lm_scale_1.7_attention_scale_1.9 20.63
421
+ ngram_lm_scale_2.2_attention_scale_3.0 20.67
422
+ ngram_lm_scale_1.9_attention_scale_2.3 20.78
423
+ ngram_lm_scale_1.1_attention_scale_0.6 20.79
424
+ ngram_lm_scale_2.0_attention_scale_2.5 20.81
425
+ ngram_lm_scale_1.3_attention_scale_1.0 20.83
426
+ ngram_lm_scale_1.9_attention_scale_2.2 21.05
427
+ ngram_lm_scale_2.3_attention_scale_3.0 21.18
428
+ ngram_lm_scale_1.7_attention_scale_1.7 21.32
429
+ ngram_lm_scale_1.9_attention_scale_2.1 21.36
430
+ ngram_lm_scale_1.5_attention_scale_1.3 21.39
431
+ ngram_lm_scale_2.0_attention_scale_2.3 21.39
432
+ ngram_lm_scale_2.1_attention_scale_2.5 21.44
433
+ ngram_lm_scale_1.3_attention_scale_0.9 21.53
434
+ ngram_lm_scale_1.2_attention_scale_0.7 21.54
435
+ ngram_lm_scale_1.1_attention_scale_0.5 21.67
436
+ ngram_lm_scale_1.9_attention_scale_2.0 21.78
437
+ ngram_lm_scale_2.0_attention_scale_2.2 21.79
438
+ ngram_lm_scale_1.5_attention_scale_1.2 21.93
439
+ ngram_lm_scale_1.0_attention_scale_0.3 22.04
440
+ ngram_lm_scale_2.2_attention_scale_2.5 22.15
441
+ ngram_lm_scale_2.1_attention_scale_2.3 22.18
442
+ ngram_lm_scale_2.0_attention_scale_2.1 22.2
443
+ ngram_lm_scale_3.0_attention_scale_4.0 22.22
444
+ ngram_lm_scale_1.9_attention_scale_1.9 22.23
445
+ ngram_lm_scale_1.7_attention_scale_1.5 22.37
446
+ ngram_lm_scale_1.2_attention_scale_0.6 22.38
447
+ ngram_lm_scale_2.5_attention_scale_3.0 22.4
448
+ ngram_lm_scale_2.1_attention_scale_2.2 22.54
449
+ ngram_lm_scale_1.5_attention_scale_1.1 22.56
450
+ ngram_lm_scale_2.0_attention_scale_2.0 22.61
451
+ ngram_lm_scale_0.9_attention_scale_0.1 22.7
452
+ ngram_lm_scale_2.3_attention_scale_2.5 22.84
453
+ ngram_lm_scale_2.2_attention_scale_2.3 22.92
454
+ ngram_lm_scale_0.9_attention_scale_0.08 23.02
455
+ ngram_lm_scale_1.3_attention_scale_0.7 23.02
456
+ ngram_lm_scale_2.1_attention_scale_2.1 23.03
457
+ ngram_lm_scale_2.0_attention_scale_1.9 23.16
458
+ ngram_lm_scale_1.9_attention_scale_1.7 23.31
459
+ ngram_lm_scale_1.5_attention_scale_1.0 23.32
460
+ ngram_lm_scale_1.2_attention_scale_0.5 23.4
461
+ ngram_lm_scale_2.2_attention_scale_2.2 23.45
462
+ ngram_lm_scale_2.1_attention_scale_2.0 23.55
463
+ ngram_lm_scale_0.9_attention_scale_0.05 23.6
464
+ ngram_lm_scale_1.7_attention_scale_1.3 23.65
465
+ ngram_lm_scale_2.3_attention_scale_2.3 23.79
466
+ ngram_lm_scale_2.2_attention_scale_2.1 23.94
467
+ ngram_lm_scale_1.1_attention_scale_0.3 23.95
468
+ ngram_lm_scale_1.3_attention_scale_0.6 24.04
469
+ ngram_lm_scale_1.5_attention_scale_0.9 24.07
470
+ ngram_lm_scale_2.1_attention_scale_1.9 24.07
471
+ ngram_lm_scale_2.0_attention_scale_1.7 24.25
472
+ ngram_lm_scale_2.3_attention_scale_2.2 24.27
473
+ ngram_lm_scale_4.0_attention_scale_5.0 24.3
474
+ ngram_lm_scale_1.7_attention_scale_1.2 24.31
475
+ ngram_lm_scale_0.9_attention_scale_0.01 24.33
476
+ ngram_lm_scale_2.5_attention_scale_2.5 24.42
477
+ ngram_lm_scale_2.2_attention_scale_2.0 24.45
478
+ ngram_lm_scale_1.9_attention_scale_1.5 24.52
479
+ ngram_lm_scale_2.3_attention_scale_2.1 24.75
480
+ ngram_lm_scale_2.2_attention_scale_1.9 24.97
481
+ ngram_lm_scale_1.0_attention_scale_0.1 25.02
482
+ ngram_lm_scale_1.7_attention_scale_1.1 25.03
483
+ ngram_lm_scale_1.3_attention_scale_0.5 25.08
484
+ ngram_lm_scale_2.1_attention_scale_1.7 25.16
485
+ ngram_lm_scale_2.3_attention_scale_2.0 25.26
486
+ ngram_lm_scale_2.5_attention_scale_2.3 25.32
487
+ ngram_lm_scale_1.0_attention_scale_0.08 25.39
488
+ ngram_lm_scale_2.0_attention_scale_1.5 25.4
489
+ ngram_lm_scale_3.0_attention_scale_3.0 25.55
490
+ ngram_lm_scale_2.5_attention_scale_2.2 25.72
491
+ ngram_lm_scale_2.3_attention_scale_1.9 25.73
492
+ ngram_lm_scale_1.9_attention_scale_1.3 25.78
493
+ ngram_lm_scale_1.7_attention_scale_1.0 25.86
494
+ ngram_lm_scale_1.0_attention_scale_0.05 25.93
495
+ ngram_lm_scale_1.2_attention_scale_0.3 25.96
496
+ ngram_lm_scale_1.5_attention_scale_0.7 25.99
497
+ ngram_lm_scale_2.2_attention_scale_1.7 26.03
498
+ ngram_lm_scale_2.5_attention_scale_2.1 26.23
499
+ ngram_lm_scale_2.1_attention_scale_1.5 26.41
500
+ ngram_lm_scale_1.9_attention_scale_1.2 26.54
501
+ ngram_lm_scale_1.7_attention_scale_0.9 26.72
502
+ ngram_lm_scale_1.0_attention_scale_0.01 26.74
503
+ ngram_lm_scale_2.5_attention_scale_2.0 26.85
504
+ ngram_lm_scale_2.0_attention_scale_1.3 26.88
505
+ ngram_lm_scale_2.3_attention_scale_1.7 26.98
506
+ ngram_lm_scale_4.0_attention_scale_4.0 27.04
507
+ ngram_lm_scale_1.5_attention_scale_0.6 27.1
508
+ ngram_lm_scale_1.1_attention_scale_0.1 27.23
509
+ ngram_lm_scale_1.9_attention_scale_1.1 27.38
510
+ ngram_lm_scale_2.5_attention_scale_1.9 27.39
511
+ ngram_lm_scale_2.2_attention_scale_1.5 27.4
512
+ ngram_lm_scale_1.1_attention_scale_0.08 27.59
513
+ ngram_lm_scale_2.0_attention_scale_1.2 27.67
514
+ ngram_lm_scale_3.0_attention_scale_2.5 27.73
515
+ ngram_lm_scale_1.3_attention_scale_0.3 27.76
516
+ ngram_lm_scale_2.1_attention_scale_1.3 27.87
517
+ ngram_lm_scale_5.0_attention_scale_5.0 28.03
518
+ ngram_lm_scale_1.1_attention_scale_0.05 28.14
519
+ ngram_lm_scale_2.3_attention_scale_1.5 28.24
520
+ ngram_lm_scale_1.9_attention_scale_1.0 28.26
521
+ ngram_lm_scale_1.5_attention_scale_0.5 28.4
522
+ ngram_lm_scale_2.0_attention_scale_1.1 28.49
523
+ ngram_lm_scale_2.5_attention_scale_1.7 28.55
524
+ ngram_lm_scale_2.1_attention_scale_1.2 28.65
525
+ ngram_lm_scale_3.0_attention_scale_2.3 28.67
526
+ ngram_lm_scale_2.2_attention_scale_1.3 28.82
527
+ ngram_lm_scale_1.7_attention_scale_0.7 28.84
528
+ ngram_lm_scale_1.1_attention_scale_0.01 28.96
529
+ ngram_lm_scale_3.0_attention_scale_2.2 29.13
530
+ ngram_lm_scale_1.9_attention_scale_0.9 29.16
531
+ ngram_lm_scale_1.2_attention_scale_0.1 29.24
532
+ ngram_lm_scale_2.0_attention_scale_1.0 29.32
533
+ ngram_lm_scale_2.1_attention_scale_1.1 29.49
534
+ ngram_lm_scale_2.2_attention_scale_1.2 29.6
535
+ ngram_lm_scale_1.2_attention_scale_0.08 29.62
536
+ ngram_lm_scale_3.0_attention_scale_2.1 29.65
537
+ ngram_lm_scale_2.3_attention_scale_1.3 29.72
538
+ ngram_lm_scale_2.5_attention_scale_1.5 29.86
539
+ ngram_lm_scale_1.7_attention_scale_0.6 29.92
540
+ ngram_lm_scale_1.2_attention_scale_0.05 30.17
541
+ ngram_lm_scale_3.0_attention_scale_2.0 30.19
542
+ ngram_lm_scale_2.0_attention_scale_0.9 30.24
543
+ ngram_lm_scale_2.1_attention_scale_1.0 30.32
544
+ ngram_lm_scale_2.2_attention_scale_1.1 30.38
545
+ ngram_lm_scale_2.3_attention_scale_1.2 30.41
546
+ ngram_lm_scale_4.0_attention_scale_3.0 30.56
547
+ ngram_lm_scale_3.0_attention_scale_1.9 30.66
548
+ ngram_lm_scale_5.0_attention_scale_4.0 30.83
549
+ ngram_lm_scale_1.2_attention_scale_0.01 30.85
550
+ ngram_lm_scale_1.3_attention_scale_0.1 30.98
551
+ ngram_lm_scale_1.5_attention_scale_0.3 31.09
552
+ ngram_lm_scale_1.7_attention_scale_0.5 31.12
553
+ ngram_lm_scale_2.1_attention_scale_0.9 31.13
554
+ ngram_lm_scale_2.5_attention_scale_1.3 31.13
555
+ ngram_lm_scale_2.2_attention_scale_1.0 31.15
556
+ ngram_lm_scale_2.3_attention_scale_1.1 31.15
557
+ ngram_lm_scale_1.9_attention_scale_0.7 31.19
558
+ ngram_lm_scale_1.3_attention_scale_0.08 31.3
559
+ ngram_lm_scale_3.0_attention_scale_1.7 31.66
560
+ ngram_lm_scale_1.3_attention_scale_0.05 31.79
561
+ ngram_lm_scale_2.5_attention_scale_1.2 31.82
562
+ ngram_lm_scale_2.3_attention_scale_1.0 31.9
563
+ ngram_lm_scale_2.2_attention_scale_0.9 31.98
564
+ ngram_lm_scale_2.0_attention_scale_0.7 32.08
565
+ ngram_lm_scale_1.9_attention_scale_0.6 32.14
566
+ ngram_lm_scale_4.0_attention_scale_2.5 32.25
567
+ ngram_lm_scale_1.3_attention_scale_0.01 32.33
568
+ ngram_lm_scale_2.5_attention_scale_1.1 32.52
569
+ ngram_lm_scale_2.3_attention_scale_0.9 32.66
570
+ ngram_lm_scale_3.0_attention_scale_1.5 32.7
571
+ ngram_lm_scale_2.1_attention_scale_0.7 32.8
572
+ ngram_lm_scale_2.0_attention_scale_0.6 32.94
573
+ ngram_lm_scale_4.0_attention_scale_2.3 32.98
574
+ ngram_lm_scale_1.9_attention_scale_0.5 33.05
575
+ ngram_lm_scale_2.5_attention_scale_1.0 33.07
576
+ ngram_lm_scale_1.7_attention_scale_0.3 33.24
577
+ ngram_lm_scale_4.0_attention_scale_2.2 33.29
578
+ ngram_lm_scale_5.0_attention_scale_3.0 33.37
579
+ ngram_lm_scale_2.2_attention_scale_0.7 33.43
580
+ ngram_lm_scale_1.5_attention_scale_0.1 33.5
581
+ ngram_lm_scale_2.1_attention_scale_0.6 33.58
582
+ ngram_lm_scale_1.5_attention_scale_0.08 33.7
583
+ ngram_lm_scale_2.0_attention_scale_0.5 33.71
584
+ ngram_lm_scale_4.0_attention_scale_2.1 33.71
585
+ ngram_lm_scale_2.5_attention_scale_0.9 33.72
586
+ ngram_lm_scale_3.0_attention_scale_1.3 33.72
587
+ ngram_lm_scale_1.5_attention_scale_0.05 34.0
588
+ ngram_lm_scale_2.3_attention_scale_0.7 34.03
589
+ ngram_lm_scale_4.0_attention_scale_2.0 34.05
590
+ ngram_lm_scale_2.2_attention_scale_0.6 34.13
591
+ ngram_lm_scale_3.0_attention_scale_1.2 34.23
592
+ ngram_lm_scale_2.1_attention_scale_0.5 34.25
593
+ ngram_lm_scale_1.5_attention_scale_0.01 34.35
594
+ ngram_lm_scale_4.0_attention_scale_1.9 34.39
595
+ ngram_lm_scale_1.9_attention_scale_0.3 34.53
596
+ ngram_lm_scale_2.3_attention_scale_0.6 34.62
597
+ ngram_lm_scale_3.0_attention_scale_1.1 34.66
598
+ ngram_lm_scale_5.0_attention_scale_2.5 34.72
599
+ ngram_lm_scale_2.2_attention_scale_0.5 34.74
600
+ ngram_lm_scale_2.5_attention_scale_0.7 34.87
601
+ ngram_lm_scale_1.7_attention_scale_0.1 34.96
602
+ ngram_lm_scale_4.0_attention_scale_1.7 35.0
603
+ ngram_lm_scale_2.0_attention_scale_0.3 35.07
604
+ ngram_lm_scale_3.0_attention_scale_1.0 35.1
605
+ ngram_lm_scale_1.7_attention_scale_0.08 35.14
606
+ ngram_lm_scale_2.3_attention_scale_0.5 35.18
607
+ ngram_lm_scale_5.0_attention_scale_2.3 35.18
608
+ ngram_lm_scale_1.7_attention_scale_0.05 35.37
609
+ ngram_lm_scale_2.5_attention_scale_0.6 35.37
610
+ ngram_lm_scale_5.0_attention_scale_2.2 35.38
611
+ ngram_lm_scale_2.1_attention_scale_0.3 35.47
612
+ ngram_lm_scale_3.0_attention_scale_0.9 35.47
613
+ ngram_lm_scale_4.0_attention_scale_1.5 35.55
614
+ ngram_lm_scale_5.0_attention_scale_2.1 35.59
615
+ ngram_lm_scale_1.7_attention_scale_0.01 35.67
616
+ ngram_lm_scale_2.5_attention_scale_0.5 35.77
617
+ ngram_lm_scale_5.0_attention_scale_2.0 35.81
618
+ ngram_lm_scale_2.2_attention_scale_0.3 35.86
619
+ ngram_lm_scale_1.9_attention_scale_0.1 35.98
620
+ ngram_lm_scale_5.0_attention_scale_1.9 36.06
621
+ ngram_lm_scale_1.9_attention_scale_0.08 36.1
622
+ ngram_lm_scale_4.0_attention_scale_1.3 36.1
623
+ ngram_lm_scale_3.0_attention_scale_0.7 36.17
624
+ ngram_lm_scale_2.3_attention_scale_0.3 36.21
625
+ ngram_lm_scale_1.9_attention_scale_0.05 36.3
626
+ ngram_lm_scale_4.0_attention_scale_1.2 36.33
627
+ ngram_lm_scale_5.0_attention_scale_1.7 36.38
628
+ ngram_lm_scale_2.0_attention_scale_0.1 36.4
629
+ ngram_lm_scale_1.9_attention_scale_0.01 36.49
630
+ ngram_lm_scale_2.0_attention_scale_0.08 36.49
631
+ ngram_lm_scale_3.0_attention_scale_0.6 36.51
632
+ ngram_lm_scale_4.0_attention_scale_1.1 36.56
633
+ ngram_lm_scale_2.0_attention_scale_0.05 36.58
634
+ ngram_lm_scale_2.1_attention_scale_0.1 36.64
635
+ ngram_lm_scale_2.1_attention_scale_0.08 36.71
636
+ ngram_lm_scale_2.5_attention_scale_0.3 36.71
637
+ ngram_lm_scale_5.0_attention_scale_1.5 36.75
638
+ ngram_lm_scale_2.0_attention_scale_0.01 36.76
639
+ ngram_lm_scale_4.0_attention_scale_1.0 36.78
640
+ ngram_lm_scale_2.1_attention_scale_0.05 36.83
641
+ ngram_lm_scale_2.2_attention_scale_0.1 36.83
642
+ ngram_lm_scale_3.0_attention_scale_0.5 36.84
643
+ ngram_lm_scale_2.2_attention_scale_0.08 36.94
644
+ ngram_lm_scale_4.0_attention_scale_0.9 37.01
645
+ ngram_lm_scale_2.1_attention_scale_0.01 37.07
646
+ ngram_lm_scale_2.3_attention_scale_0.1 37.08
647
+ ngram_lm_scale_2.2_attention_scale_0.05 37.09
648
+ ngram_lm_scale_5.0_attention_scale_1.3 37.11
649
+ ngram_lm_scale_2.3_attention_scale_0.08 37.17
650
+ ngram_lm_scale_5.0_attention_scale_1.2 37.29
651
+ ngram_lm_scale_2.2_attention_scale_0.01 37.3
652
+ ngram_lm_scale_2.3_attention_scale_0.05 37.3
653
+ ngram_lm_scale_2.3_attention_scale_0.01 37.45
654
+ ngram_lm_scale_2.5_attention_scale_0.1 37.45
655
+ ngram_lm_scale_3.0_attention_scale_0.3 37.46
656
+ ngram_lm_scale_5.0_attention_scale_1.1 37.5
657
+ ngram_lm_scale_4.0_attention_scale_0.7 37.53
658
+ ngram_lm_scale_2.5_attention_scale_0.08 37.54
659
+ ngram_lm_scale_2.5_attention_scale_0.05 37.63
660
+ ngram_lm_scale_5.0_attention_scale_1.0 37.64
661
+ ngram_lm_scale_4.0_attention_scale_0.6 37.68
662
+ ngram_lm_scale_2.5_attention_scale_0.01 37.76
663
+ ngram_lm_scale_5.0_attention_scale_0.9 37.82
664
+ ngram_lm_scale_4.0_attention_scale_0.5 37.94
665
+ ngram_lm_scale_3.0_attention_scale_0.1 38.07
666
+ ngram_lm_scale_3.0_attention_scale_0.08 38.11
667
+ ngram_lm_scale_5.0_attention_scale_0.7 38.14
668
+ ngram_lm_scale_3.0_attention_scale_0.05 38.2
669
+ ngram_lm_scale_3.0_attention_scale_0.01 38.32
670
+ ngram_lm_scale_4.0_attention_scale_0.3 38.32
671
+ ngram_lm_scale_5.0_attention_scale_0.6 38.33
672
+ ngram_lm_scale_5.0_attention_scale_0.5 38.47
673
+ ngram_lm_scale_4.0_attention_scale_0.1 38.73
674
+ ngram_lm_scale_4.0_attention_scale_0.08 38.77
675
+ ngram_lm_scale_5.0_attention_scale_0.3 38.78
676
+ ngram_lm_scale_4.0_attention_scale_0.05 38.85
677
+ ngram_lm_scale_4.0_attention_scale_0.01 38.92
678
+ ngram_lm_scale_5.0_attention_scale_0.1 39.07
679
+ ngram_lm_scale_5.0_attention_scale_0.08 39.11
680
+ ngram_lm_scale_5.0_attention_scale_0.05 39.17
681
+ ngram_lm_scale_5.0_attention_scale_0.01 39.22
682
+
683
+ 2022-06-25 22:21:00,565 INFO [decode.py:483] batch 0/?, cuts processed until now is 2
684
+ 2022-06-25 23:27:30,947 INFO [decode.py:483] batch 100/?, cuts processed until now is 277
685
+ 2022-06-26 00:57:14,858 INFO [decode.py:483] batch 200/?, cuts processed until now is 570
686
+ 2022-06-26 02:12:55,830 INFO [decode.py:483] batch 300/?, cuts processed until now is 872
687
+ 2022-06-26 03:29:19,515 INFO [decode.py:483] batch 400/?, cuts processed until now is 1159
688
+ 2022-06-26 04:56:43,624 INFO [decode.py:483] batch 500/?, cuts processed until now is 1433
689
+ 2022-06-26 06:10:18,732 INFO [decode.py:483] batch 600/?, cuts processed until now is 1723
690
+ 2022-06-26 07:18:12,039 INFO [decode.py:483] batch 700/?, cuts processed until now is 2012
691
+ 2022-06-26 08:44:29,122 INFO [decode.py:483] batch 800/?, cuts processed until now is 2287
692
+ 2022-06-26 10:12:10,048 INFO [decode.py:483] batch 900/?, cuts processed until now is 2582
693
+ 2022-06-26 11:16:50,308 INFO [decode.py:483] batch 1000/?, cuts processed until now is 2870
694
+ 2022-06-26 12:28:04,120 INFO [decode.py:483] batch 1100/?, cuts processed until now is 3152
695
+ 2022-06-26 13:37:56,712 INFO [decode.py:483] batch 1200/?, cuts processed until now is 3458
696
+ 2022-06-26 15:01:44,066 INFO [decode.py:483] batch 1300/?, cuts processed until now is 3731
697
+ 2022-06-26 16:36:03,534 INFO [decode.py:483] batch 1400/?, cuts processed until now is 4012
698
+ 2022-06-26 18:51:27,441 INFO [decode.py:483] batch 1500/?, cuts processed until now is 4290
699
+ 2022-06-26 19:03:44,177 INFO [decode.py:733] Caught exception:
700
+
701
+ Some bad things happened. Please read the above error messages and stack
702
+ trace. If you are using Python, the following command may be helpful:
703
+
704
+ gdb --args python /path/to/your/code.py
705
+
706
+ (You can use `gdb` to debug the code. Please consider compiling
707
+ a debug version of k2.).
708
+
709
+ If you are unable to fix it, please open an issue at:
710
+
711
+ https://github.com/k2-fsa/k2/issues/new
712
+
713
+
714
+ 2022-06-26 19:03:44,177 INFO [decode.py:734] num_arcs before pruning: 2390934
715
+ 2022-06-26 19:03:44,178 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
716
+ 2022-06-26 19:03:44,562 INFO [decode.py:747] num_arcs after pruning: 5752
717
+ 2022-06-26 19:57:22,287 INFO [decode.py:733] Caught exception:
718
+
719
+ Some bad things happened. Please read the above error messages and stack
720
+ trace. If you are using Python, the following command may be helpful:
721
+
722
+ gdb --args python /path/to/your/code.py
723
+
724
+ (You can use `gdb` to debug the code. Please consider compiling
725
+ a debug version of k2.).
726
+
727
+ If you are unable to fix it, please open an issue at:
728
+
729
+ https://github.com/k2-fsa/k2/issues/new
730
+
731
+
732
+ 2022-06-26 19:57:22,288 INFO [decode.py:734] num_arcs before pruning: 846107
733
+ 2022-06-26 19:57:22,288 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
734
+ 2022-06-26 19:57:22,439 INFO [decode.py:747] num_arcs after pruning: 8745
735
+ 2022-06-26 21:23:00,214 INFO [decode.py:483] batch 1600/?, cuts processed until now is 4539
736
+ 2022-06-26 23:32:50,861 INFO [decode.py:483] batch 1700/?, cuts processed until now is 4765
737
+ 2022-06-27 01:24:13,288 INFO [decode.py:483] batch 1800/?, cuts processed until now is 4973
738
+ 2022-06-27 01:34:55,855 INFO [decode.py:532]
739
+ For dev, WER of different settings are:
740
+ ngram_lm_scale_0.01_attention_scale_0.5 15.78 best for dev
741
+ ngram_lm_scale_0.01_attention_scale_0.6 15.79
742
+ ngram_lm_scale_0.01_attention_scale_0.7 15.8
743
+ ngram_lm_scale_0.05_attention_scale_0.5 15.8
744
+ ngram_lm_scale_0.05_attention_scale_0.6 15.8
745
+ ngram_lm_scale_0.01_attention_scale_0.3 15.81
746
+ ngram_lm_scale_0.05_attention_scale_0.3 15.81
747
+ ngram_lm_scale_0.05_attention_scale_0.7 15.81
748
+ ngram_lm_scale_0.08_attention_scale_0.3 15.81
749
+ ngram_lm_scale_0.08_attention_scale_0.5 15.81
750
+ ngram_lm_scale_0.01_attention_scale_0.9 15.82
751
+ ngram_lm_scale_0.08_attention_scale_0.6 15.82
752
+ ngram_lm_scale_0.08_attention_scale_0.7 15.83
753
+ ngram_lm_scale_0.1_attention_scale_0.3 15.83
754
+ ngram_lm_scale_0.1_attention_scale_0.5 15.83
755
+ ngram_lm_scale_0.1_attention_scale_0.6 15.83
756
+ ngram_lm_scale_0.01_attention_scale_1.0 15.84
757
+ ngram_lm_scale_0.05_attention_scale_0.9 15.84
758
+ ngram_lm_scale_0.08_attention_scale_0.9 15.86
759
+ ngram_lm_scale_0.1_attention_scale_0.7 15.86
760
+ ngram_lm_scale_0.05_attention_scale_1.0 15.87
761
+ ngram_lm_scale_0.01_attention_scale_1.1 15.88
762
+ ngram_lm_scale_0.1_attention_scale_0.9 15.88
763
+ ngram_lm_scale_0.01_attention_scale_1.2 15.9
764
+ ngram_lm_scale_0.01_attention_scale_1.3 15.91
765
+ ngram_lm_scale_0.05_attention_scale_1.1 15.91
766
+ ngram_lm_scale_0.08_attention_scale_1.0 15.91
767
+ ngram_lm_scale_0.1_attention_scale_1.0 15.91
768
+ ngram_lm_scale_0.01_attention_scale_0.1 15.92
769
+ ngram_lm_scale_0.05_attention_scale_1.2 15.93
770
+ ngram_lm_scale_0.05_attention_scale_0.1 15.94
771
+ ngram_lm_scale_0.05_attention_scale_1.3 15.94
772
+ ngram_lm_scale_0.08_attention_scale_1.1 15.94
773
+ ngram_lm_scale_0.08_attention_scale_1.2 15.94
774
+ ngram_lm_scale_0.01_attention_scale_1.5 15.95
775
+ ngram_lm_scale_0.08_attention_scale_0.1 15.95
776
+ ngram_lm_scale_0.1_attention_scale_1.1 15.95
777
+ ngram_lm_scale_0.01_attention_scale_0.08 15.96
778
+ ngram_lm_scale_0.05_attention_scale_0.08 15.97
779
+ ngram_lm_scale_0.08_attention_scale_0.08 15.97
780
+ ngram_lm_scale_0.1_attention_scale_0.1 15.97
781
+ ngram_lm_scale_0.1_attention_scale_1.2 15.97
782
+ ngram_lm_scale_0.01_attention_scale_1.7 15.98
783
+ ngram_lm_scale_0.05_attention_scale_1.5 15.98
784
+ ngram_lm_scale_0.1_attention_scale_1.3 15.98
785
+ ngram_lm_scale_0.01_attention_scale_0.05 15.99
786
+ ngram_lm_scale_0.08_attention_scale_1.3 15.99
787
+ ngram_lm_scale_0.1_attention_scale_0.08 15.99
788
+ ngram_lm_scale_0.05_attention_scale_0.05 16.01
789
+ ngram_lm_scale_0.08_attention_scale_1.5 16.01
790
+ ngram_lm_scale_0.05_attention_scale_1.7 16.02
791
+ ngram_lm_scale_0.08_attention_scale_0.05 16.03
792
+ ngram_lm_scale_0.1_attention_scale_0.05 16.03
793
+ ngram_lm_scale_0.1_attention_scale_1.5 16.04
794
+ ngram_lm_scale_0.01_attention_scale_0.01 16.05
795
+ ngram_lm_scale_0.01_attention_scale_1.9 16.05
796
+ ngram_lm_scale_0.01_attention_scale_2.0 16.07
797
+ ngram_lm_scale_0.08_attention_scale_1.7 16.07
798
+ ngram_lm_scale_0.05_attention_scale_0.01 16.08
799
+ ngram_lm_scale_0.05_attention_scale_1.9 16.08
800
+ ngram_lm_scale_0.1_attention_scale_1.7 16.09
801
+ ngram_lm_scale_0.08_attention_scale_0.01 16.1
802
+ ngram_lm_scale_0.08_attention_scale_1.9 16.1
803
+ ngram_lm_scale_0.05_attention_scale_2.0 16.11
804
+ ngram_lm_scale_0.01_attention_scale_2.1 16.12
805
+ ngram_lm_scale_0.1_attention_scale_1.9 16.12
806
+ ngram_lm_scale_0.05_attention_scale_2.1 16.13
807
+ ngram_lm_scale_0.08_attention_scale_2.0 16.13
808
+ ngram_lm_scale_0.1_attention_scale_0.01 16.13
809
+ ngram_lm_scale_0.01_attention_scale_2.2 16.14
810
+ ngram_lm_scale_0.08_attention_scale_2.1 16.14
811
+ ngram_lm_scale_0.01_attention_scale_2.3 16.15
812
+ ngram_lm_scale_0.05_attention_scale_2.2 16.15
813
+ ngram_lm_scale_0.1_attention_scale_2.0 16.15
814
+ ngram_lm_scale_0.3_attention_scale_0.9 16.16
815
+ ngram_lm_scale_0.01_attention_scale_2.5 16.17
816
+ ngram_lm_scale_0.05_attention_scale_2.3 16.17
817
+ ngram_lm_scale_0.08_attention_scale_2.2 16.18
818
+ ngram_lm_scale_0.1_attention_scale_2.1 16.18
819
+ ngram_lm_scale_0.3_attention_scale_0.5 16.18
820
+ ngram_lm_scale_0.3_attention_scale_0.6 16.18
821
+ ngram_lm_scale_0.3_attention_scale_0.7 16.18
822
+ ngram_lm_scale_0.3_attention_scale_1.0 16.18
823
+ ngram_lm_scale_0.3_attention_scale_1.1 16.18
824
+ ngram_lm_scale_0.05_attention_scale_2.5 16.19
825
+ ngram_lm_scale_0.1_attention_scale_2.2 16.19
826
+ ngram_lm_scale_0.08_attention_scale_2.3 16.2
827
+ ngram_lm_scale_0.1_attention_scale_2.3 16.21
828
+ ngram_lm_scale_0.08_attention_scale_2.5 16.22
829
+ ngram_lm_scale_0.3_attention_scale_0.3 16.22
830
+ ngram_lm_scale_0.3_attention_scale_1.2 16.22
831
+ ngram_lm_scale_0.3_attention_scale_1.3 16.22
832
+ ngram_lm_scale_0.01_attention_scale_3.0 16.23
833
+ ngram_lm_scale_0.1_attention_scale_2.5 16.23
834
+ ngram_lm_scale_0.3_attention_scale_1.5 16.24
835
+ ngram_lm_scale_0.05_attention_scale_3.0 16.27
836
+ ngram_lm_scale_0.3_attention_scale_1.7 16.29
837
+ ngram_lm_scale_0.08_attention_scale_3.0 16.3
838
+ ngram_lm_scale_0.1_attention_scale_3.0 16.31
839
+ ngram_lm_scale_0.3_attention_scale_1.9 16.31
840
+ ngram_lm_scale_0.3_attention_scale_2.0 16.34
841
+ ngram_lm_scale_0.3_attention_scale_2.1 16.36
842
+ ngram_lm_scale_0.3_attention_scale_2.2 16.38
843
+ ngram_lm_scale_0.3_attention_scale_2.3 16.38
844
+ ngram_lm_scale_0.01_attention_scale_4.0 16.39
845
+ ngram_lm_scale_0.05_attention_scale_4.0 16.4
846
+ ngram_lm_scale_0.3_attention_scale_0.1 16.41
847
+ ngram_lm_scale_0.3_attention_scale_2.5 16.42
848
+ ngram_lm_scale_0.08_attention_scale_4.0 16.45
849
+ ngram_lm_scale_0.3_attention_scale_0.08 16.45
850
+ ngram_lm_scale_0.1_attention_scale_4.0 16.47
851
+ ngram_lm_scale_0.01_attention_scale_5.0 16.52
852
+ ngram_lm_scale_0.3_attention_scale_3.0 16.53
853
+ ngram_lm_scale_0.05_attention_scale_5.0 16.54
854
+ ngram_lm_scale_0.3_attention_scale_0.05 16.54
855
+ ngram_lm_scale_0.08_attention_scale_5.0 16.57
856
+ ngram_lm_scale_0.1_attention_scale_5.0 16.57
857
+ ngram_lm_scale_0.5_attention_scale_0.9 16.58
858
+ ngram_lm_scale_0.5_attention_scale_1.2 16.58
859
+ ngram_lm_scale_0.5_attention_scale_1.1 16.59
860
+ ngram_lm_scale_0.5_attention_scale_1.3 16.6
861
+ ngram_lm_scale_0.3_attention_scale_4.0 16.61
862
+ ngram_lm_scale_0.5_attention_scale_1.0 16.61
863
+ ngram_lm_scale_0.5_attention_scale_1.5 16.64
864
+ ngram_lm_scale_0.3_attention_scale_5.0 16.66
865
+ ngram_lm_scale_0.5_attention_scale_0.7 16.66
866
+ ngram_lm_scale_0.5_attention_scale_1.9 16.66
867
+ ngram_lm_scale_0.5_attention_scale_2.0 16.66
868
+ ngram_lm_scale_0.5_attention_scale_2.1 16.66
869
+ ngram_lm_scale_0.3_attention_scale_0.01 16.67
870
+ ngram_lm_scale_0.5_attention_scale_1.7 16.67
871
+ ngram_lm_scale_0.5_attention_scale_2.2 16.68
872
+ ngram_lm_scale_0.5_attention_scale_2.3 16.69
873
+ ngram_lm_scale_0.5_attention_scale_2.5 16.69
874
+ ngram_lm_scale_0.5_attention_scale_0.6 16.73
875
+ ngram_lm_scale_0.5_attention_scale_3.0 16.73
876
+ ngram_lm_scale_0.5_attention_scale_0.5 16.79
877
+ ngram_lm_scale_0.5_attention_scale_4.0 16.8
878
+ ngram_lm_scale_0.5_attention_scale_5.0 16.83
879
+ ngram_lm_scale_0.6_attention_scale_3.0 16.83
880
+ ngram_lm_scale_0.6_attention_scale_2.1 16.85
881
+ ngram_lm_scale_0.6_attention_scale_2.2 16.85
882
+ ngram_lm_scale_0.6_attention_scale_2.3 16.85
883
+ ngram_lm_scale_0.6_attention_scale_2.5 16.86
884
+ ngram_lm_scale_0.6_attention_scale_2.0 16.87
885
+ ngram_lm_scale_0.6_attention_scale_1.7 16.89
886
+ ngram_lm_scale_0.6_attention_scale_1.9 16.89
887
+ ngram_lm_scale_0.6_attention_scale_4.0 16.9
888
+ ngram_lm_scale_0.6_attention_scale_1.5 16.91
889
+ ngram_lm_scale_0.6_attention_scale_5.0 16.93
890
+ ngram_lm_scale_0.6_attention_scale_1.2 16.94
891
+ ngram_lm_scale_0.6_attention_scale_1.1 16.95
892
+ ngram_lm_scale_0.6_attention_scale_1.3 16.95
893
+ ngram_lm_scale_0.6_attention_scale_1.0 16.96
894
+ ngram_lm_scale_0.6_attention_scale_0.9 16.97
895
+ ngram_lm_scale_0.5_attention_scale_0.3 17.0
896
+ ngram_lm_scale_0.7_attention_scale_4.0 17.01
897
+ ngram_lm_scale_0.7_attention_scale_5.0 17.01
898
+ ngram_lm_scale_0.7_attention_scale_3.0 17.02
899
+ ngram_lm_scale_0.7_attention_scale_2.5 17.06
900
+ ngram_lm_scale_0.7_attention_scale_2.3 17.09
901
+ ngram_lm_scale_0.7_attention_scale_2.1 17.1
902
+ ngram_lm_scale_0.7_attention_scale_2.2 17.1
903
+ ngram_lm_scale_0.6_attention_scale_0.7 17.12
904
+ ngram_lm_scale_0.7_attention_scale_2.0 17.13
905
+ ngram_lm_scale_0.7_attention_scale_1.9 17.14
906
+ ngram_lm_scale_0.7_attention_scale_1.7 17.16
907
+ ngram_lm_scale_0.7_attention_scale_1.5 17.18
908
+ ngram_lm_scale_0.7_attention_scale_1.3 17.21
909
+ ngram_lm_scale_0.6_attention_scale_0.6 17.24
910
+ ngram_lm_scale_0.9_attention_scale_5.0 17.27
911
+ ngram_lm_scale_0.7_attention_scale_1.2 17.28
912
+ ngram_lm_scale_0.7_attention_scale_1.1 17.32
913
+ ngram_lm_scale_0.6_attention_scale_0.5 17.34
914
+ ngram_lm_scale_0.9_attention_scale_4.0 17.34
915
+ ngram_lm_scale_0.7_attention_scale_1.0 17.39
916
+ ngram_lm_scale_1.0_attention_scale_5.0 17.39
917
+ ngram_lm_scale_0.9_attention_scale_3.0 17.47
918
+ ngram_lm_scale_0.7_attention_scale_0.9 17.48
919
+ ngram_lm_scale_1.0_attention_scale_4.0 17.48
920
+ ngram_lm_scale_1.1_attention_scale_5.0 17.52
921
+ ngram_lm_scale_0.9_attention_scale_2.5 17.56
922
+ ngram_lm_scale_0.9_attention_scale_2.3 17.6
923
+ ngram_lm_scale_0.9_attention_scale_2.2 17.63
924
+ ngram_lm_scale_1.1_attention_scale_4.0 17.64
925
+ ngram_lm_scale_1.2_attention_scale_5.0 17.64
926
+ ngram_lm_scale_0.5_attention_scale_0.1 17.66
927
+ ngram_lm_scale_0.9_attention_scale_2.1 17.66
928
+ ngram_lm_scale_1.0_attention_scale_3.0 17.66
929
+ ngram_lm_scale_0.9_attention_scale_2.0 17.67
930
+ ngram_lm_scale_0.7_attention_scale_0.7 17.68
931
+ ngram_lm_scale_0.9_attention_scale_1.9 17.74
932
+ ngram_lm_scale_1.3_attention_scale_5.0 17.76
933
+ ngram_lm_scale_1.0_attention_scale_2.5 17.77
934
+ ngram_lm_scale_0.6_attention_scale_0.3 17.78
935
+ ngram_lm_scale_1.2_attention_scale_4.0 17.78
936
+ ngram_lm_scale_0.5_attention_scale_0.08 17.81
937
+ ngram_lm_scale_0.7_attention_scale_0.6 17.83
938
+ ngram_lm_scale_0.9_attention_scale_1.7 17.84
939
+ ngram_lm_scale_1.0_attention_scale_2.3 17.85
940
+ ngram_lm_scale_1.1_attention_scale_3.0 17.87
941
+ ngram_lm_scale_1.0_attention_scale_2.2 17.9
942
+ ngram_lm_scale_0.9_attention_scale_1.5 17.95
943
+ ngram_lm_scale_1.0_attention_scale_2.1 17.96
944
+ ngram_lm_scale_0.5_attention_scale_0.05 17.97
945
+ ngram_lm_scale_1.3_attention_scale_4.0 17.98
946
+ ngram_lm_scale_1.0_attention_scale_2.0 18.01
947
+ ngram_lm_scale_1.1_attention_scale_2.5 18.06
948
+ ngram_lm_scale_1.5_attention_scale_5.0 18.08
949
+ ngram_lm_scale_0.7_attention_scale_0.5 18.09
950
+ ngram_lm_scale_1.0_attention_scale_1.9 18.09
951
+ ngram_lm_scale_1.2_attention_scale_3.0 18.1
952
+ ngram_lm_scale_0.9_attention_scale_1.3 18.16
953
+ ngram_lm_scale_1.1_attention_scale_2.3 18.19
954
+ ngram_lm_scale_1.0_attention_scale_1.7 18.24
955
+ ngram_lm_scale_1.1_attention_scale_2.2 18.24
956
+ ngram_lm_scale_0.5_attention_scale_0.01 18.28
957
+ ngram_lm_scale_0.9_attention_scale_1.2 18.29
958
+ ngram_lm_scale_1.1_attention_scale_2.1 18.33
959
+ ngram_lm_scale_1.3_attention_scale_3.0 18.33
960
+ ngram_lm_scale_1.5_attention_scale_4.0 18.36
961
+ ngram_lm_scale_1.7_attention_scale_5.0 18.38
962
+ ngram_lm_scale_1.1_attention_scale_2.0 18.39
963
+ ngram_lm_scale_1.2_attention_scale_2.5 18.39
964
+ ngram_lm_scale_1.0_attention_scale_1.5 18.44
965
+ ngram_lm_scale_0.9_attention_scale_1.1 18.45
966
+ ngram_lm_scale_1.1_attention_scale_1.9 18.5
967
+ ngram_lm_scale_1.2_attention_scale_2.3 18.53
968
+ ngram_lm_scale_0.9_attention_scale_1.0 18.59
969
+ ngram_lm_scale_1.2_attention_scale_2.2 18.63
970
+ ngram_lm_scale_1.0_attention_scale_1.3 18.73
971
+ ngram_lm_scale_1.1_attention_scale_1.7 18.74
972
+ ngram_lm_scale_1.2_attention_scale_2.1 18.74
973
+ ngram_lm_scale_1.3_attention_scale_2.5 18.74
974
+ ngram_lm_scale_0.9_attention_scale_0.9 18.77
975
+ ngram_lm_scale_0.6_attention_scale_0.1 18.79
976
+ ngram_lm_scale_0.7_attention_scale_0.3 18.79
977
+ ngram_lm_scale_1.2_attention_scale_2.0 18.88
978
+ ngram_lm_scale_1.9_attention_scale_5.0 18.88
979
+ ngram_lm_scale_1.0_attention_scale_1.2 18.92
980
+ ngram_lm_scale_1.7_attention_scale_4.0 18.94
981
+ ngram_lm_scale_0.6_attention_scale_0.08 19.0
982
+ ngram_lm_scale_1.3_attention_scale_2.3 19.0
983
+ ngram_lm_scale_1.2_attention_scale_1.9 19.03
984
+ ngram_lm_scale_1.1_attention_scale_1.5 19.06
985
+ ngram_lm_scale_2.0_attention_scale_5.0 19.06
986
+ ngram_lm_scale_1.5_attention_scale_3.0 19.07
987
+ ngram_lm_scale_1.0_attention_scale_1.1 19.09
988
+ ngram_lm_scale_1.3_attention_scale_2.2 19.19
989
+ ngram_lm_scale_0.6_attention_scale_0.05 19.21
990
+ ngram_lm_scale_2.1_attention_scale_5.0 19.23
991
+ ngram_lm_scale_0.9_attention_scale_0.7 19.29
992
+ ngram_lm_scale_1.3_attention_scale_2.1 19.31
993
+ ngram_lm_scale_1.2_attention_scale_1.7 19.34
994
+ ngram_lm_scale_1.0_attention_scale_1.0 19.4
995
+ ngram_lm_scale_1.1_attention_scale_1.3 19.4
996
+ ngram_lm_scale_1.9_attention_scale_4.0 19.42
997
+ ngram_lm_scale_1.3_attention_scale_2.0 19.43
998
+ ngram_lm_scale_2.2_attention_scale_5.0 19.45
999
+ ngram_lm_scale_1.3_attention_scale_1.9 19.53
1000
+ ngram_lm_scale_1.1_attention_scale_1.2 19.59
1001
+ ngram_lm_scale_1.0_attention_scale_0.9 19.61
1002
+ ngram_lm_scale_1.2_attention_scale_1.5 19.61
1003
+ ngram_lm_scale_1.5_attention_scale_2.5 19.63
1004
+ ngram_lm_scale_0.6_attention_scale_0.01 19.65
1005
+ ngram_lm_scale_0.9_attention_scale_0.6 19.71
1006
+ ngram_lm_scale_2.3_attention_scale_5.0 19.73
1007
+ ngram_lm_scale_2.0_attention_scale_4.0 19.75
1008
+ ngram_lm_scale_1.7_attention_scale_3.0 19.81
1009
+ ngram_lm_scale_1.3_attention_scale_1.7 19.89
1010
+ ngram_lm_scale_1.1_attention_scale_1.1 19.9
1011
+ ngram_lm_scale_1.5_attention_scale_2.3 19.96
1012
+ ngram_lm_scale_1.5_attention_scale_2.2 20.13
1013
+ ngram_lm_scale_2.1_attention_scale_4.0 20.14
1014
+ ngram_lm_scale_1.2_attention_scale_1.3 20.15
1015
+ ngram_lm_scale_0.9_attention_scale_0.5 20.18
1016
+ ngram_lm_scale_1.1_attention_scale_1.0 20.23
1017
+ ngram_lm_scale_0.7_attention_scale_0.1 20.25
1018
+ ngram_lm_scale_1.5_attention_scale_2.1 20.29
1019
+ ngram_lm_scale_2.5_attention_scale_5.0 20.37
1020
+ ngram_lm_scale_1.3_attention_scale_1.5 20.38
1021
+ ngram_lm_scale_1.0_attention_scale_0.7 20.4
1022
+ ngram_lm_scale_1.2_attention_scale_1.2 20.46
1023
+ ngram_lm_scale_0.7_attention_scale_0.08 20.49
1024
+ ngram_lm_scale_1.5_attention_scale_2.0 20.5
1025
+ ngram_lm_scale_2.2_attention_scale_4.0 20.5
1026
+ ngram_lm_scale_1.1_attention_scale_0.9 20.59
1027
+ ngram_lm_scale_1.7_attention_scale_2.5 20.65
1028
+ ngram_lm_scale_1.2_attention_scale_1.1 20.76
1029
+ ngram_lm_scale_1.9_attention_scale_3.0 20.77
1030
+ ngram_lm_scale_1.5_attention_scale_1.9 20.78
1031
+ ngram_lm_scale_0.7_attention_scale_0.05 20.84
1032
+ ngram_lm_scale_2.3_attention_scale_4.0 20.9
1033
+ ngram_lm_scale_1.0_attention_scale_0.6 20.91
1034
+ ngram_lm_scale_1.3_attention_scale_1.3 21.02
1035
+ ngram_lm_scale_1.7_attention_scale_2.3 21.12
1036
+ ngram_lm_scale_1.2_attention_scale_1.0 21.21
1037
+ ngram_lm_scale_2.0_attention_scale_3.0 21.25
1038
+ ngram_lm_scale_1.5_attention_scale_1.7 21.35
1039
+ ngram_lm_scale_1.7_attention_scale_2.2 21.37
1040
+ ngram_lm_scale_1.3_attention_scale_1.2 21.4
1041
+ ngram_lm_scale_0.7_attention_scale_0.01 21.42
1042
+ ngram_lm_scale_1.0_attention_scale_0.5 21.62
1043
+ ngram_lm_scale_1.7_attention_scale_2.1 21.66
1044
+ ngram_lm_scale_2.5_attention_scale_4.0 21.67
1045
+ ngram_lm_scale_1.1_attention_scale_0.7 21.68
1046
+ ngram_lm_scale_0.9_attention_scale_0.3 21.71
1047
+ ngram_lm_scale_2.1_attention_scale_3.0 21.79
1048
+ ngram_lm_scale_1.2_attention_scale_0.9 21.8
1049
+ ngram_lm_scale_1.9_attention_scale_2.5 21.87
1050
+ ngram_lm_scale_1.3_attention_scale_1.1 21.91
1051
+ ngram_lm_scale_1.7_attention_scale_2.0 21.94
1052
+ ngram_lm_scale_3.0_attention_scale_5.0 21.99
1053
+ ngram_lm_scale_1.5_attention_scale_1.5 22.12
1054
+ ngram_lm_scale_2.2_attention_scale_3.0 22.28
1055
+ ngram_lm_scale_1.7_attention_scale_1.9 22.31
1056
+ ngram_lm_scale_1.1_attention_scale_0.6 22.38
1057
+ ngram_lm_scale_1.9_attention_scale_2.3 22.41
1058
+ ngram_lm_scale_2.0_attention_scale_2.5 22.45
1059
+ ngram_lm_scale_1.3_attention_scale_1.0 22.46
1060
+ ngram_lm_scale_1.9_attention_scale_2.2 22.75
1061
+ ngram_lm_scale_2.3_attention_scale_3.0 22.87
1062
+ ngram_lm_scale_1.3_attention_scale_0.9 23.06
1063
+ ngram_lm_scale_1.5_attention_scale_1.3 23.07
1064
+ ngram_lm_scale_1.2_attention_scale_0.7 23.08
1065
+ ngram_lm_scale_1.7_attention_scale_1.7 23.13
1066
+ ngram_lm_scale_1.9_attention_scale_2.1 23.14
1067
+ ngram_lm_scale_2.0_attention_scale_2.3 23.16
1068
+ ngram_lm_scale_2.1_attention_scale_2.5 23.17
1069
+ ngram_lm_scale_1.1_attention_scale_0.5 23.27
1070
+ ngram_lm_scale_1.9_attention_scale_2.0 23.5
1071
+ ngram_lm_scale_2.0_attention_scale_2.2 23.51
1072
+ ngram_lm_scale_1.0_attention_scale_0.3 23.6
1073
+ ngram_lm_scale_1.5_attention_scale_1.2 23.6
1074
+ ngram_lm_scale_2.2_attention_scale_2.5 23.9
1075
+ ngram_lm_scale_2.1_attention_scale_2.3 23.91
1076
+ ngram_lm_scale_2.0_attention_scale_2.1 23.94
1077
+ ngram_lm_scale_1.9_attention_scale_1.9 23.99
1078
+ ngram_lm_scale_1.2_attention_scale_0.6 24.02
1079
+ ngram_lm_scale_1.7_attention_scale_1.5 24.06
1080
+ ngram_lm_scale_3.0_attention_scale_4.0 24.11
1081
+ ngram_lm_scale_2.5_attention_scale_3.0 24.24
1082
+ ngram_lm_scale_1.5_attention_scale_1.1 24.25
1083
+ ngram_lm_scale_0.9_attention_scale_0.1 24.27
1084
+ ngram_lm_scale_2.1_attention_scale_2.2 24.36
1085
+ ngram_lm_scale_2.0_attention_scale_2.0 24.42
1086
+ ngram_lm_scale_0.9_attention_scale_0.08 24.57
1087
+ ngram_lm_scale_1.3_attention_scale_0.7 24.62
1088
+ ngram_lm_scale_2.3_attention_scale_2.5 24.62
1089
+ ngram_lm_scale_2.2_attention_scale_2.3 24.72
1090
+ ngram_lm_scale_2.1_attention_scale_2.1 24.79
1091
+ ngram_lm_scale_2.0_attention_scale_1.9 24.87
1092
+ ngram_lm_scale_1.5_attention_scale_1.0 24.99
1093
+ ngram_lm_scale_1.2_attention_scale_0.5 25.0
1094
+ ngram_lm_scale_1.9_attention_scale_1.7 25.05
1095
+ ngram_lm_scale_2.2_attention_scale_2.2 25.15
1096
+ ngram_lm_scale_0.9_attention_scale_0.05 25.18
1097
+ ngram_lm_scale_2.1_attention_scale_2.0 25.28
1098
+ ngram_lm_scale_1.7_attention_scale_1.3 25.29
1099
+ ngram_lm_scale_2.3_attention_scale_2.3 25.47
1100
+ ngram_lm_scale_1.1_attention_scale_0.3 25.55
1101
+ ngram_lm_scale_1.3_attention_scale_0.6 25.6
1102
+ ngram_lm_scale_2.2_attention_scale_2.1 25.61
1103
+ ngram_lm_scale_1.5_attention_scale_0.9 25.73
1104
+ ngram_lm_scale_2.1_attention_scale_1.9 25.79
1105
+ ngram_lm_scale_2.3_attention_scale_2.2 25.92
1106
+ ngram_lm_scale_0.9_attention_scale_0.01 25.95
1107
+ ngram_lm_scale_4.0_attention_scale_5.0 25.96
1108
+ ngram_lm_scale_2.0_attention_scale_1.7 25.97
1109
+ ngram_lm_scale_1.7_attention_scale_1.2 26.03
1110
+ ngram_lm_scale_2.5_attention_scale_2.5 26.03
1111
+ ngram_lm_scale_2.2_attention_scale_2.0 26.1
1112
+ ngram_lm_scale_1.9_attention_scale_1.5 26.22
1113
+ ngram_lm_scale_2.3_attention_scale_2.1 26.44
1114
+ ngram_lm_scale_1.0_attention_scale_0.1 26.63
1115
+ ngram_lm_scale_1.3_attention_scale_0.5 26.68
1116
+ ngram_lm_scale_2.2_attention_scale_1.9 26.69
1117
+ ngram_lm_scale_1.7_attention_scale_1.1 26.71
1118
+ ngram_lm_scale_2.1_attention_scale_1.7 26.89
1119
+ ngram_lm_scale_2.3_attention_scale_2.0 26.96
1120
+ ngram_lm_scale_2.5_attention_scale_2.3 26.98
1121
+ ngram_lm_scale_1.0_attention_scale_0.08 26.99
1122
+ ngram_lm_scale_2.0_attention_scale_1.5 27.15
1123
+ ngram_lm_scale_3.0_attention_scale_3.0 27.22
1124
+ ngram_lm_scale_1.9_attention_scale_1.3 27.44
1125
+ ngram_lm_scale_1.7_attention_scale_1.0 27.46
1126
+ ngram_lm_scale_2.5_attention_scale_2.2 27.47
1127
+ ngram_lm_scale_2.3_attention_scale_1.9 27.49
1128
+ ngram_lm_scale_1.0_attention_scale_0.05 27.54
1129
+ ngram_lm_scale_1.2_attention_scale_0.3 27.57
1130
+ ngram_lm_scale_1.5_attention_scale_0.7 27.67
1131
+ ngram_lm_scale_2.2_attention_scale_1.7 27.75
1132
+ ngram_lm_scale_2.5_attention_scale_2.1 27.96
1133
+ ngram_lm_scale_2.1_attention_scale_1.5 28.08
1134
+ ngram_lm_scale_1.9_attention_scale_1.2 28.17
1135
+ ngram_lm_scale_1.0_attention_scale_0.01 28.32
1136
+ ngram_lm_scale_1.7_attention_scale_0.9 28.38
1137
+ ngram_lm_scale_2.5_attention_scale_2.0 28.4
1138
+ ngram_lm_scale_2.0_attention_scale_1.3 28.5
1139
+ ngram_lm_scale_2.3_attention_scale_1.7 28.57
1140
+ ngram_lm_scale_4.0_attention_scale_4.0 28.74
1141
+ ngram_lm_scale_1.5_attention_scale_0.6 28.79
1142
+ ngram_lm_scale_1.1_attention_scale_0.1 28.81
1143
+ ngram_lm_scale_1.9_attention_scale_1.1 28.98
1144
+ ngram_lm_scale_2.2_attention_scale_1.5 28.99
1145
+ ngram_lm_scale_2.5_attention_scale_1.9 29.0
1146
+ ngram_lm_scale_1.1_attention_scale_0.08 29.2
1147
+ ngram_lm_scale_2.0_attention_scale_1.2 29.21
1148
+ ngram_lm_scale_3.0_attention_scale_2.5 29.36
1149
+ ngram_lm_scale_2.1_attention_scale_1.3 29.42
1150
+ ngram_lm_scale_1.3_attention_scale_0.3 29.49
1151
+ ngram_lm_scale_5.0_attention_scale_5.0 29.66
1152
+ ngram_lm_scale_2.3_attention_scale_1.5 29.78
1153
+ ngram_lm_scale_1.1_attention_scale_0.05 29.79
1154
+ ngram_lm_scale_1.9_attention_scale_1.0 29.8
1155
+ ngram_lm_scale_2.0_attention_scale_1.1 30.02
1156
+ ngram_lm_scale_1.5_attention_scale_0.5 30.04
1157
+ ngram_lm_scale_2.5_attention_scale_1.7 30.09
1158
+ ngram_lm_scale_3.0_attention_scale_2.3 30.14
1159
+ ngram_lm_scale_2.1_attention_scale_1.2 30.19
1160
+ ngram_lm_scale_2.2_attention_scale_1.3 30.32
1161
+ ngram_lm_scale_1.7_attention_scale_0.7 30.36
1162
+ ngram_lm_scale_1.1_attention_scale_0.01 30.53
1163
+ ngram_lm_scale_3.0_attention_scale_2.2 30.68
1164
+ ngram_lm_scale_1.9_attention_scale_0.9 30.76
1165
+ ngram_lm_scale_1.2_attention_scale_0.1 30.81
1166
+ ngram_lm_scale_2.0_attention_scale_1.0 30.87
1167
+ ngram_lm_scale_2.1_attention_scale_1.1 31.02
1168
+ ngram_lm_scale_2.2_attention_scale_1.2 31.09
1169
+ ngram_lm_scale_3.0_attention_scale_2.1 31.12
1170
+ ngram_lm_scale_2.3_attention_scale_1.3 31.18
1171
+ ngram_lm_scale_1.2_attention_scale_0.08 31.2
1172
+ ngram_lm_scale_2.5_attention_scale_1.5 31.35
1173
+ ngram_lm_scale_1.7_attention_scale_0.6 31.49
1174
+ ngram_lm_scale_3.0_attention_scale_2.0 31.62
1175
+ ngram_lm_scale_2.0_attention_scale_0.9 31.75
1176
+ ngram_lm_scale_1.2_attention_scale_0.05 31.77
1177
+ ngram_lm_scale_2.1_attention_scale_1.0 31.78
1178
+ ngram_lm_scale_2.2_attention_scale_1.1 31.87
1179
+ ngram_lm_scale_2.3_attention_scale_1.2 31.91
1180
+ ngram_lm_scale_4.0_attention_scale_3.0 32.04
1181
+ ngram_lm_scale_3.0_attention_scale_1.9 32.1
1182
+ ngram_lm_scale_5.0_attention_scale_4.0 32.25
1183
+ ngram_lm_scale_1.2_attention_scale_0.01 32.41
1184
+ ngram_lm_scale_1.5_attention_scale_0.3 32.5
1185
+ ngram_lm_scale_1.3_attention_scale_0.1 32.52
1186
+ ngram_lm_scale_1.7_attention_scale_0.5 32.55
1187
+ ngram_lm_scale_2.2_attention_scale_1.0 32.59
1188
+ ngram_lm_scale_2.3_attention_scale_1.1 32.59
1189
+ ngram_lm_scale_1.9_attention_scale_0.7 32.6
1190
+ ngram_lm_scale_2.1_attention_scale_0.9 32.61
1191
+ ngram_lm_scale_2.5_attention_scale_1.3 32.63
1192
+ ngram_lm_scale_1.3_attention_scale_0.08 32.75
1193
+ ngram_lm_scale_3.0_attention_scale_1.7 33.19
1194
+ ngram_lm_scale_2.5_attention_scale_1.2 33.22
1195
+ ngram_lm_scale_1.3_attention_scale_0.05 33.24
1196
+ ngram_lm_scale_2.3_attention_scale_1.0 33.26
1197
+ ngram_lm_scale_2.2_attention_scale_0.9 33.29
1198
+ ngram_lm_scale_2.0_attention_scale_0.7 33.42
1199
+ ngram_lm_scale_1.9_attention_scale_0.6 33.48
1200
+ ngram_lm_scale_4.0_attention_scale_2.5 33.7
1201
+ ngram_lm_scale_1.3_attention_scale_0.01 33.83
1202
+ ngram_lm_scale_2.5_attention_scale_1.1 33.86
1203
+ ngram_lm_scale_2.3_attention_scale_0.9 33.99
1204
+ ngram_lm_scale_3.0_attention_scale_1.5 34.09
1205
+ ngram_lm_scale_2.1_attention_scale_0.7 34.14
1206
+ ngram_lm_scale_2.0_attention_scale_0.6 34.27
1207
+ ngram_lm_scale_1.9_attention_scale_0.5 34.36
1208
+ ngram_lm_scale_4.0_attention_scale_2.3 34.42
1209
+ ngram_lm_scale_2.5_attention_scale_1.0 34.49
1210
+ ngram_lm_scale_1.7_attention_scale_0.3 34.54
1211
+ ngram_lm_scale_4.0_attention_scale_2.2 34.77
1212
+ ngram_lm_scale_2.2_attention_scale_0.7 34.78
1213
+ ngram_lm_scale_5.0_attention_scale_3.0 34.81
1214
+ ngram_lm_scale_1.5_attention_scale_0.1 34.85
1215
+ ngram_lm_scale_2.1_attention_scale_0.6 34.88
1216
+ ngram_lm_scale_2.0_attention_scale_0.5 35.01
1217
+ ngram_lm_scale_1.5_attention_scale_0.08 35.03
1218
+ ngram_lm_scale_2.5_attention_scale_0.9 35.03
1219
+ ngram_lm_scale_3.0_attention_scale_1.3 35.07
1220
+ ngram_lm_scale_4.0_attention_scale_2.1 35.08
1221
+ ngram_lm_scale_2.3_attention_scale_0.7 35.29
1222
+ ngram_lm_scale_1.5_attention_scale_0.05 35.31
1223
+ ngram_lm_scale_4.0_attention_scale_2.0 35.35
1224
+ ngram_lm_scale_2.2_attention_scale_0.6 35.4
1225
+ ngram_lm_scale_3.0_attention_scale_1.2 35.48
1226
+ ngram_lm_scale_2.1_attention_scale_0.5 35.53
1227
+ ngram_lm_scale_4.0_attention_scale_1.9 35.66
1228
+ ngram_lm_scale_1.5_attention_scale_0.01 35.74
1229
+ ngram_lm_scale_2.3_attention_scale_0.6 35.79
1230
+ ngram_lm_scale_1.9_attention_scale_0.3 35.84
1231
+ ngram_lm_scale_3.0_attention_scale_1.1 35.86
1232
+ ngram_lm_scale_2.2_attention_scale_0.5 35.95
1233
+ ngram_lm_scale_5.0_attention_scale_2.5 35.98
1234
+ ngram_lm_scale_2.5_attention_scale_0.7 36.0
1235
+ ngram_lm_scale_1.7_attention_scale_0.1 36.21
1236
+ ngram_lm_scale_4.0_attention_scale_1.7 36.23
1237
+ ngram_lm_scale_3.0_attention_scale_1.0 36.26
1238
+ ngram_lm_scale_2.0_attention_scale_0.3 36.33
1239
+ ngram_lm_scale_2.3_attention_scale_0.5 36.33
1240
+ ngram_lm_scale_1.7_attention_scale_0.08 36.39
1241
+ ngram_lm_scale_5.0_attention_scale_2.3 36.42
1242
+ ngram_lm_scale_2.5_attention_scale_0.6 36.49
1243
+ ngram_lm_scale_5.0_attention_scale_2.2 36.59
1244
+ ngram_lm_scale_1.7_attention_scale_0.05 36.61
1245
+ ngram_lm_scale_3.0_attention_scale_0.9 36.61
1246
+ ngram_lm_scale_2.1_attention_scale_0.3 36.73
1247
+ ngram_lm_scale_4.0_attention_scale_1.5 36.75
1248
+ ngram_lm_scale_5.0_attention_scale_2.1 36.84
1249
+ ngram_lm_scale_1.7_attention_scale_0.01 36.86
1250
+ ngram_lm_scale_2.5_attention_scale_0.5 37.0
1251
+ ngram_lm_scale_2.2_attention_scale_0.3 37.05
1252
+ ngram_lm_scale_5.0_attention_scale_2.0 37.05
1253
+ ngram_lm_scale_1.9_attention_scale_0.1 37.17
1254
+ ngram_lm_scale_5.0_attention_scale_1.9 37.27
1255
+ ngram_lm_scale_1.9_attention_scale_0.08 37.3
1256
+ ngram_lm_scale_4.0_attention_scale_1.3 37.32
1257
+ ngram_lm_scale_2.3_attention_scale_0.3 37.39
1258
+ ngram_lm_scale_3.0_attention_scale_0.7 37.39
1259
+ ngram_lm_scale_1.9_attention_scale_0.05 37.48
1260
+ ngram_lm_scale_2.0_attention_scale_0.1 37.52
1261
+ ngram_lm_scale_4.0_attention_scale_1.2 37.56
1262
+ ngram_lm_scale_2.0_attention_scale_0.08 37.63
1263
+ ngram_lm_scale_5.0_attention_scale_1.7 37.66
1264
+ ngram_lm_scale_1.9_attention_scale_0.01 37.72
1265
+ ngram_lm_scale_2.1_attention_scale_0.1 37.78
1266
+ ngram_lm_scale_3.0_attention_scale_0.6 37.79
1267
+ ngram_lm_scale_2.0_attention_scale_0.05 37.8
1268
+ ngram_lm_scale_4.0_attention_scale_1.1 37.82
1269
+ ngram_lm_scale_2.1_attention_scale_0.08 37.87
1270
+ ngram_lm_scale_2.5_attention_scale_0.3 37.91
1271
+ ngram_lm_scale_2.0_attention_scale_0.01 37.99
1272
+ ngram_lm_scale_2.1_attention_scale_0.05 38.03
1273
+ ngram_lm_scale_5.0_attention_scale_1.5 38.04
1274
+ ngram_lm_scale_2.2_attention_scale_0.1 38.05
1275
+ ngram_lm_scale_4.0_attention_scale_1.0 38.11
1276
+ ngram_lm_scale_3.0_attention_scale_0.5 38.13
1277
+ ngram_lm_scale_2.2_attention_scale_0.08 38.16
1278
+ ngram_lm_scale_2.1_attention_scale_0.01 38.23
1279
+ ngram_lm_scale_2.3_attention_scale_0.1 38.27
1280
+ ngram_lm_scale_2.2_attention_scale_0.05 38.28
1281
+ ngram_lm_scale_4.0_attention_scale_0.9 38.33
1282
+ ngram_lm_scale_2.3_attention_scale_0.08 38.35
1283
+ ngram_lm_scale_5.0_attention_scale_1.3 38.41
1284
+ ngram_lm_scale_2.2_attention_scale_0.01 38.45
1285
+ ngram_lm_scale_2.3_attention_scale_0.05 38.47
1286
+ ngram_lm_scale_5.0_attention_scale_1.2 38.58
1287
+ ngram_lm_scale_2.5_attention_scale_0.1 38.61
1288
+ ngram_lm_scale_2.3_attention_scale_0.01 38.65
1289
+ ngram_lm_scale_2.5_attention_scale_0.08 38.68
1290
+ ngram_lm_scale_3.0_attention_scale_0.3 38.72
1291
+ ngram_lm_scale_5.0_attention_scale_1.1 38.73
1292
+ ngram_lm_scale_2.5_attention_scale_0.05 38.77
1293
+ ngram_lm_scale_4.0_attention_scale_0.7 38.77
1294
+ ngram_lm_scale_5.0_attention_scale_1.0 38.9
1295
+ ngram_lm_scale_2.5_attention_scale_0.01 38.93
1296
+ ngram_lm_scale_4.0_attention_scale_0.6 38.94
1297
+ ngram_lm_scale_5.0_attention_scale_0.9 39.04
1298
+ ngram_lm_scale_4.0_attention_scale_0.5 39.11
1299
+ ngram_lm_scale_3.0_attention_scale_0.1 39.23
1300
+ ngram_lm_scale_3.0_attention_scale_0.08 39.3
1301
+ ngram_lm_scale_5.0_attention_scale_0.7 39.33
1302
+ ngram_lm_scale_3.0_attention_scale_0.05 39.38
1303
+ ngram_lm_scale_5.0_attention_scale_0.6 39.44
1304
+ ngram_lm_scale_3.0_attention_scale_0.01 39.48
1305
+ ngram_lm_scale_4.0_attention_scale_0.3 39.49
1306
+ ngram_lm_scale_5.0_attention_scale_0.5 39.59
1307
+ ngram_lm_scale_4.0_attention_scale_0.1 39.91
1308
+ ngram_lm_scale_4.0_attention_scale_0.08 39.94
1309
+ ngram_lm_scale_5.0_attention_scale_0.3 39.95
1310
+ ngram_lm_scale_4.0_attention_scale_0.05 39.98
1311
+ ngram_lm_scale_4.0_attention_scale_0.01 40.02
1312
+ ngram_lm_scale_5.0_attention_scale_0.1 40.22
1313
+ ngram_lm_scale_5.0_attention_scale_0.08 40.25
1314
+ ngram_lm_scale_5.0_attention_scale_0.05 40.3
1315
+ ngram_lm_scale_5.0_attention_scale_0.01 40.35
1316
+
1317
+ 2022-06-27 01:34:55,856 INFO [decode.py:695] Done!
decoding-results/log-attention-decoder/log-decode-2022-06-24-17-22-16 ADDED
@@ -0,0 +1,1428 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2022-06-24 17:22:16,163 INFO [decode.py:548] Decoding started
2
+ 2022-06-24 17:22:16,164 INFO [decode.py:549] {'subsampling_factor': 4, 'vgg_frontend': False, 'use_feat_batchnorm': True, 'feature_dim': 80, 'nhead': 8, 'attention_dim': 512, 'num_decoder_layers': 6, 'search_beam': 20, 'output_beam': 8, 'min_active_states': 30, 'max_active_states': 10000, 'use_double_scores': True, 'env_info': {'k2-version': '1.11', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '', 'k2-git-date': '', 'lhotse-version': '1.3.0.dev+git.a07121a.clean', 'torch-cuda-available': False, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'test', 'icefall-git-sha1': '7e72d78-dirty', 'icefall-git-date': 'Sat May 28 19:13:53 2022', 'icefall-path': '/alt-arabic/speech/amir/k2/tmp/icefall', 'k2-path': '/home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/k2/__init__.py', 'lhotse-path': '/alt-arabic/speech/amir/k2/tmp/lhotse/lhotse/__init__.py', 'hostname': 'crimv3srv026', 'IP address': '10.141.0.21'}, 'epoch': 39, 'avg': 10, 'method': 'attention-decoder', 'num_paths': 1000, 'nbest_scale': 0.5, 'exp_dir': PosixPath('conformer_ctc/exp_5000_att0.8'), 'lang_dir': PosixPath('data/lang_bpe_5000'), 'lm_dir': PosixPath('data/lm'), 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 100, 'bucketing_sampler': True, 'num_buckets': 30, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'on_the_fly_feats': False, 'shuffle': False, 'drop_last': True, 'return_cuts': True, 'num_workers': 20, 'enable_spec_aug': False, 'spec_aug_time_warp_factor': 80, 'enable_musan': False}
3
+ 2022-06-24 17:22:16,482 INFO [lexicon.py:177] Loading pre-compiled data/lang_bpe_5000/Linv.pt
4
+ 2022-06-24 17:22:16,882 INFO [decode.py:559] device: cpu
5
+ 2022-06-24 17:22:40,122 INFO [decode.py:621] Loading pre-compiled G_4_gram.pt
6
+ 2022-06-24 17:22:46,616 INFO [decode.py:657] averaging ['conformer_ctc/exp_5000_att0.8/epoch-30.pt', 'conformer_ctc/exp_5000_att0.8/epoch-31.pt', 'conformer_ctc/exp_5000_att0.8/epoch-32.pt', 'conformer_ctc/exp_5000_att0.8/epoch-33.pt', 'conformer_ctc/exp_5000_att0.8/epoch-34.pt', 'conformer_ctc/exp_5000_att0.8/epoch-35.pt', 'conformer_ctc/exp_5000_att0.8/epoch-36.pt', 'conformer_ctc/exp_5000_att0.8/epoch-37.pt', 'conformer_ctc/exp_5000_att0.8/epoch-38.pt', 'conformer_ctc/exp_5000_att0.8/epoch-39.pt']
7
+ 2022-06-24 17:26:25,489 INFO [decode.py:664] Number of model parameters: 90786736
8
+ 2022-06-24 17:26:25,489 INFO [asr_datamodule.py:374] About to get test cuts
9
+ 2022-06-24 17:26:25,524 INFO [asr_datamodule.py:367] About to get dev cuts
10
+ 2022-06-24 17:29:23,798 INFO [decode.py:483] batch 0/?, cuts processed until now is 13
11
+ 2022-06-24 18:44:36,966 INFO [decode.py:733] Caught exception:
12
+
13
+ Some bad things happened. Please read the above error messages and stack
14
+ trace. If you are using Python, the following command may be helpful:
15
+
16
+ gdb --args python /path/to/your/code.py
17
+
18
+ (You can use `gdb` to debug the code. Please consider compiling
19
+ a debug version of k2.).
20
+
21
+ If you are unable to fix it, please open an issue at:
22
+
23
+ https://github.com/k2-fsa/k2/issues/new
24
+
25
+
26
+ 2022-06-24 18:44:36,969 INFO [decode.py:734] num_arcs before pruning: 3793830
27
+ 2022-06-24 18:44:36,969 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
28
+ 2022-06-24 18:44:37,490 INFO [decode.py:747] num_arcs after pruning: 15363
29
+ 2022-06-24 18:49:44,859 INFO [decode.py:733] Caught exception:
30
+
31
+ Some bad things happened. Please read the above error messages and stack
32
+ trace. If you are using Python, the following command may be helpful:
33
+
34
+ gdb --args python /path/to/your/code.py
35
+
36
+ (You can use `gdb` to debug the code. Please consider compiling
37
+ a debug version of k2.).
38
+
39
+ If you are unable to fix it, please open an issue at:
40
+
41
+ https://github.com/k2-fsa/k2/issues/new
42
+
43
+
44
+ 2022-06-24 18:49:44,860 INFO [decode.py:734] num_arcs before pruning: 7793289
45
+ 2022-06-24 18:49:44,860 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
46
+ 2022-06-24 18:49:45,947 INFO [decode.py:747] num_arcs after pruning: 21157
47
+ 2022-06-24 19:57:06,281 INFO [decode.py:733] Caught exception:
48
+
49
+ Some bad things happened. Please read the above error messages and stack
50
+ trace. If you are using Python, the following command may be helpful:
51
+
52
+ gdb --args python /path/to/your/code.py
53
+
54
+ (You can use `gdb` to debug the code. Please consider compiling
55
+ a debug version of k2.).
56
+
57
+ If you are unable to fix it, please open an issue at:
58
+
59
+ https://github.com/k2-fsa/k2/issues/new
60
+
61
+
62
+ 2022-06-24 19:57:06,283 INFO [decode.py:734] num_arcs before pruning: 4231973
63
+ 2022-06-24 19:57:06,284 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
64
+ 2022-06-24 19:57:06,903 INFO [decode.py:747] num_arcs after pruning: 24242
65
+ 2022-06-25 00:20:28,938 INFO [decode.py:733] Caught exception:
66
+
67
+ Some bad things happened. Please read the above error messages and stack
68
+ trace. If you are using Python, the following command may be helpful:
69
+
70
+ gdb --args python /path/to/your/code.py
71
+
72
+ (You can use `gdb` to debug the code. Please consider compiling
73
+ a debug version of k2.).
74
+
75
+ If you are unable to fix it, please open an issue at:
76
+
77
+ https://github.com/k2-fsa/k2/issues/new
78
+
79
+
80
+ 2022-06-25 00:20:28,940 INFO [decode.py:734] num_arcs before pruning: 6579893
81
+ 2022-06-25 00:20:28,940 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
82
+ 2022-06-25 00:20:29,808 INFO [decode.py:747] num_arcs after pruning: 22371
83
+ 2022-06-25 00:28:47,575 INFO [decode.py:483] batch 100/?, cuts processed until now is 1483
84
+ 2022-06-25 06:00:11,280 INFO [decode.py:733] Caught exception:
85
+
86
+ Some bad things happened. Please read the above error messages and stack
87
+ trace. If you are using Python, the following command may be helpful:
88
+
89
+ gdb --args python /path/to/your/code.py
90
+
91
+ (You can use `gdb` to debug the code. Please consider compiling
92
+ a debug version of k2.).
93
+
94
+ If you are unable to fix it, please open an issue at:
95
+
96
+ https://github.com/k2-fsa/k2/issues/new
97
+
98
+
99
+ 2022-06-25 06:00:11,282 INFO [decode.py:734] num_arcs before pruning: 7814892
100
+ 2022-06-25 06:00:11,282 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
101
+ 2022-06-25 06:00:12,279 INFO [decode.py:747] num_arcs after pruning: 23294
102
+ 2022-06-25 07:10:42,039 INFO [decode.py:483] batch 200/?, cuts processed until now is 3045
103
+ 2022-06-25 08:39:07,227 INFO [decode.py:733] Caught exception:
104
+
105
+ Some bad things happened. Please read the above error messages and stack
106
+ trace. If you are using Python, the following command may be helpful:
107
+
108
+ gdb --args python /path/to/your/code.py
109
+
110
+ (You can use `gdb` to debug the code. Please consider compiling
111
+ a debug version of k2.).
112
+
113
+ If you are unable to fix it, please open an issue at:
114
+
115
+ https://github.com/k2-fsa/k2/issues/new
116
+
117
+
118
+ 2022-06-25 08:39:07,231 INFO [decode.py:734] num_arcs before pruning: 5093346
119
+ 2022-06-25 08:39:07,231 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
120
+ 2022-06-25 08:39:08,033 INFO [decode.py:747] num_arcs after pruning: 21764
121
+ 2022-06-25 09:11:54,803 INFO [decode.py:733] Caught exception:
122
+
123
+ Some bad things happened. Please read the above error messages and stack
124
+ trace. If you are using Python, the following command may be helpful:
125
+
126
+ gdb --args python /path/to/your/code.py
127
+
128
+ (You can use `gdb` to debug the code. Please consider compiling
129
+ a debug version of k2.).
130
+
131
+ If you are unable to fix it, please open an issue at:
132
+
133
+ https://github.com/k2-fsa/k2/issues/new
134
+
135
+
136
+ 2022-06-25 09:11:54,804 INFO [decode.py:734] num_arcs before pruning: 10821654
137
+ 2022-06-25 09:11:54,804 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
138
+ 2022-06-25 09:11:56,386 INFO [decode.py:747] num_arcs after pruning: 28715
139
+ 2022-06-25 14:37:16,871 INFO [decode.py:483] batch 300/?, cuts processed until now is 4598
140
+ 2022-06-25 14:58:22,045 INFO [decode.py:733] Caught exception:
141
+
142
+ Some bad things happened. Please read the above error messages and stack
143
+ trace. If you are using Python, the following command may be helpful:
144
+
145
+ gdb --args python /path/to/your/code.py
146
+
147
+ (You can use `gdb` to debug the code. Please consider compiling
148
+ a debug version of k2.).
149
+
150
+ If you are unable to fix it, please open an issue at:
151
+
152
+ https://github.com/k2-fsa/k2/issues/new
153
+
154
+
155
+ 2022-06-25 14:58:22,046 INFO [decode.py:734] num_arcs before pruning: 5552108
156
+ 2022-06-25 14:58:22,046 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
157
+ 2022-06-25 14:58:22,815 INFO [decode.py:747] num_arcs after pruning: 21203
158
+ 2022-06-25 17:28:28,538 INFO [decode.py:532]
159
+ For test, WER of different settings are:
160
+ ngram_lm_scale_0.01_attention_scale_0.5 15.07 best for test
161
+ ngram_lm_scale_0.05_attention_scale_0.5 15.07
162
+ ngram_lm_scale_0.01_attention_scale_0.3 15.08
163
+ ngram_lm_scale_0.01_attention_scale_0.7 15.08
164
+ ngram_lm_scale_0.08_attention_scale_0.5 15.08
165
+ ngram_lm_scale_0.01_attention_scale_0.6 15.09
166
+ ngram_lm_scale_0.05_attention_scale_0.6 15.09
167
+ ngram_lm_scale_0.05_attention_scale_0.3 15.1
168
+ ngram_lm_scale_0.01_attention_scale_0.9 15.11
169
+ ngram_lm_scale_0.05_attention_scale_0.7 15.11
170
+ ngram_lm_scale_0.08_attention_scale_0.3 15.11
171
+ ngram_lm_scale_0.08_attention_scale_0.6 15.11
172
+ ngram_lm_scale_0.08_attention_scale_0.7 15.11
173
+ ngram_lm_scale_0.1_attention_scale_0.5 15.11
174
+ ngram_lm_scale_0.1_attention_scale_0.3 15.12
175
+ ngram_lm_scale_0.1_attention_scale_0.6 15.12
176
+ ngram_lm_scale_0.01_attention_scale_1.0 15.13
177
+ ngram_lm_scale_0.05_attention_scale_0.9 15.13
178
+ ngram_lm_scale_0.1_attention_scale_0.7 15.13
179
+ ngram_lm_scale_0.01_attention_scale_0.1 15.14
180
+ ngram_lm_scale_0.01_attention_scale_1.1 15.14
181
+ ngram_lm_scale_0.05_attention_scale_1.0 15.14
182
+ ngram_lm_scale_0.08_attention_scale_0.9 15.14
183
+ ngram_lm_scale_0.1_attention_scale_0.9 15.14
184
+ ngram_lm_scale_0.01_attention_scale_1.2 15.16
185
+ ngram_lm_scale_0.05_attention_scale_0.1 15.16
186
+ ngram_lm_scale_0.08_attention_scale_1.0 15.16
187
+ ngram_lm_scale_0.05_attention_scale_1.1 15.17
188
+ ngram_lm_scale_0.01_attention_scale_0.08 15.18
189
+ ngram_lm_scale_0.01_attention_scale_1.3 15.18
190
+ ngram_lm_scale_0.05_attention_scale_0.08 15.18
191
+ ngram_lm_scale_0.08_attention_scale_0.08 15.18
192
+ ngram_lm_scale_0.08_attention_scale_0.1 15.18
193
+ ngram_lm_scale_0.1_attention_scale_1.0 15.18
194
+ ngram_lm_scale_0.05_attention_scale_1.2 15.19
195
+ ngram_lm_scale_0.1_attention_scale_0.1 15.19
196
+ ngram_lm_scale_0.08_attention_scale_1.1 15.2
197
+ ngram_lm_scale_0.05_attention_scale_1.3 15.21
198
+ ngram_lm_scale_0.1_attention_scale_0.08 15.21
199
+ ngram_lm_scale_0.1_attention_scale_1.1 15.21
200
+ ngram_lm_scale_0.01_attention_scale_0.05 15.22
201
+ ngram_lm_scale_0.05_attention_scale_0.05 15.22
202
+ ngram_lm_scale_0.08_attention_scale_1.2 15.22
203
+ ngram_lm_scale_0.01_attention_scale_1.5 15.23
204
+ ngram_lm_scale_0.08_attention_scale_0.05 15.24
205
+ ngram_lm_scale_0.1_attention_scale_1.2 15.24
206
+ ngram_lm_scale_0.05_attention_scale_1.5 15.25
207
+ ngram_lm_scale_0.08_attention_scale_1.3 15.25
208
+ ngram_lm_scale_0.1_attention_scale_0.05 15.25
209
+ ngram_lm_scale_0.01_attention_scale_1.7 15.26
210
+ ngram_lm_scale_0.1_attention_scale_1.3 15.26
211
+ ngram_lm_scale_0.01_attention_scale_0.01 15.27
212
+ ngram_lm_scale_0.05_attention_scale_1.7 15.28
213
+ ngram_lm_scale_0.08_attention_scale_1.5 15.28
214
+ ngram_lm_scale_0.1_attention_scale_1.5 15.28
215
+ ngram_lm_scale_0.01_attention_scale_1.9 15.29
216
+ ngram_lm_scale_0.01_attention_scale_2.0 15.29
217
+ ngram_lm_scale_0.01_attention_scale_2.1 15.29
218
+ ngram_lm_scale_0.05_attention_scale_1.9 15.29
219
+ ngram_lm_scale_0.08_attention_scale_0.01 15.29
220
+ ngram_lm_scale_0.08_attention_scale_1.7 15.29
221
+ ngram_lm_scale_0.1_attention_scale_0.01 15.29
222
+ ngram_lm_scale_0.1_attention_scale_1.7 15.29
223
+ ngram_lm_scale_0.3_attention_scale_0.6 15.29
224
+ ngram_lm_scale_0.3_attention_scale_0.7 15.29
225
+ ngram_lm_scale_0.01_attention_scale_2.2 15.3
226
+ ngram_lm_scale_0.05_attention_scale_0.01 15.3
227
+ ngram_lm_scale_0.01_attention_scale_2.3 15.31
228
+ ngram_lm_scale_0.05_attention_scale_2.0 15.31
229
+ ngram_lm_scale_0.05_attention_scale_2.1 15.31
230
+ ngram_lm_scale_0.05_attention_scale_2.2 15.31
231
+ ngram_lm_scale_0.08_attention_scale_1.9 15.31
232
+ ngram_lm_scale_0.08_attention_scale_2.0 15.31
233
+ ngram_lm_scale_0.3_attention_scale_0.5 15.31
234
+ ngram_lm_scale_0.3_attention_scale_0.9 15.31
235
+ ngram_lm_scale_0.1_attention_scale_1.9 15.32
236
+ ngram_lm_scale_0.3_attention_scale_1.0 15.32
237
+ ngram_lm_scale_0.01_attention_scale_2.5 15.33
238
+ ngram_lm_scale_0.05_attention_scale_2.3 15.33
239
+ ngram_lm_scale_0.05_attention_scale_2.5 15.33
240
+ ngram_lm_scale_0.08_attention_scale_2.1 15.33
241
+ ngram_lm_scale_0.1_attention_scale_2.0 15.33
242
+ ngram_lm_scale_0.1_attention_scale_2.2 15.33
243
+ ngram_lm_scale_0.3_attention_scale_1.1 15.33
244
+ ngram_lm_scale_0.08_attention_scale_2.2 15.34
245
+ ngram_lm_scale_0.08_attention_scale_2.3 15.34
246
+ ngram_lm_scale_0.1_attention_scale_2.1 15.34
247
+ ngram_lm_scale_0.1_attention_scale_2.3 15.35
248
+ ngram_lm_scale_0.3_attention_scale_1.2 15.35
249
+ ngram_lm_scale_0.01_attention_scale_3.0 15.36
250
+ ngram_lm_scale_0.08_attention_scale_2.5 15.36
251
+ ngram_lm_scale_0.1_attention_scale_2.5 15.37
252
+ ngram_lm_scale_0.3_attention_scale_1.3 15.37
253
+ ngram_lm_scale_0.05_attention_scale_3.0 15.38
254
+ ngram_lm_scale_0.3_attention_scale_0.3 15.38
255
+ ngram_lm_scale_0.08_attention_scale_3.0 15.39
256
+ ngram_lm_scale_0.1_attention_scale_3.0 15.4
257
+ ngram_lm_scale_0.3_attention_scale_1.5 15.41
258
+ ngram_lm_scale_0.01_attention_scale_4.0 15.43
259
+ ngram_lm_scale_0.3_attention_scale_1.7 15.43
260
+ ngram_lm_scale_0.05_attention_scale_4.0 15.45
261
+ ngram_lm_scale_0.08_attention_scale_4.0 15.45
262
+ ngram_lm_scale_0.3_attention_scale_1.9 15.45
263
+ ngram_lm_scale_0.3_attention_scale_2.0 15.45
264
+ ngram_lm_scale_0.3_attention_scale_2.1 15.45
265
+ ngram_lm_scale_0.1_attention_scale_4.0 15.46
266
+ ngram_lm_scale_0.3_attention_scale_2.2 15.46
267
+ ngram_lm_scale_0.3_attention_scale_2.3 15.47
268
+ ngram_lm_scale_0.01_attention_scale_5.0 15.48
269
+ ngram_lm_scale_0.05_attention_scale_5.0 15.49
270
+ ngram_lm_scale_0.3_attention_scale_2.5 15.49
271
+ ngram_lm_scale_0.08_attention_scale_5.0 15.51
272
+ ngram_lm_scale_0.1_attention_scale_5.0 15.51
273
+ ngram_lm_scale_0.3_attention_scale_3.0 15.52
274
+ ngram_lm_scale_0.3_attention_scale_0.1 15.55
275
+ ngram_lm_scale_0.3_attention_scale_4.0 15.57
276
+ ngram_lm_scale_0.3_attention_scale_0.08 15.59
277
+ ngram_lm_scale_0.3_attention_scale_5.0 15.61
278
+ ngram_lm_scale_0.5_attention_scale_1.1 15.64
279
+ ngram_lm_scale_0.5_attention_scale_1.2 15.65
280
+ ngram_lm_scale_0.5_attention_scale_1.5 15.65
281
+ ngram_lm_scale_0.5_attention_scale_1.9 15.65
282
+ ngram_lm_scale_0.5_attention_scale_2.0 15.65
283
+ ngram_lm_scale_0.5_attention_scale_2.1 15.65
284
+ ngram_lm_scale_0.5_attention_scale_2.5 15.65
285
+ ngram_lm_scale_0.5_attention_scale_0.9 15.66
286
+ ngram_lm_scale_0.5_attention_scale_1.0 15.66
287
+ ngram_lm_scale_0.5_attention_scale_1.3 15.66
288
+ ngram_lm_scale_0.5_attention_scale_1.7 15.66
289
+ ngram_lm_scale_0.5_attention_scale_2.2 15.66
290
+ ngram_lm_scale_0.5_attention_scale_2.3 15.66
291
+ ngram_lm_scale_0.3_attention_scale_0.05 15.67
292
+ ngram_lm_scale_0.5_attention_scale_3.0 15.69
293
+ ngram_lm_scale_0.5_attention_scale_0.7 15.72
294
+ ngram_lm_scale_0.5_attention_scale_0.6 15.74
295
+ ngram_lm_scale_0.5_attention_scale_4.0 15.74
296
+ ngram_lm_scale_0.5_attention_scale_5.0 15.76
297
+ ngram_lm_scale_0.3_attention_scale_0.01 15.78
298
+ ngram_lm_scale_0.6_attention_scale_2.0 15.79
299
+ ngram_lm_scale_0.6_attention_scale_2.2 15.79
300
+ ngram_lm_scale_0.5_attention_scale_0.5 15.8
301
+ ngram_lm_scale_0.6_attention_scale_1.9 15.8
302
+ ngram_lm_scale_0.6_attention_scale_2.1 15.8
303
+ ngram_lm_scale_0.6_attention_scale_2.3 15.8
304
+ ngram_lm_scale_0.6_attention_scale_2.5 15.8
305
+ ngram_lm_scale_0.6_attention_scale_1.7 15.81
306
+ ngram_lm_scale_0.6_attention_scale_1.5 15.82
307
+ ngram_lm_scale_0.6_attention_scale_3.0 15.82
308
+ ngram_lm_scale_0.6_attention_scale_5.0 15.82
309
+ ngram_lm_scale_0.6_attention_scale_1.2 15.83
310
+ ngram_lm_scale_0.6_attention_scale_1.3 15.84
311
+ ngram_lm_scale_0.6_attention_scale_4.0 15.84
312
+ ngram_lm_scale_0.6_attention_scale_1.0 15.85
313
+ ngram_lm_scale_0.6_attention_scale_1.1 15.85
314
+ ngram_lm_scale_0.6_attention_scale_0.9 15.87
315
+ ngram_lm_scale_0.7_attention_scale_5.0 15.9
316
+ ngram_lm_scale_0.7_attention_scale_2.5 15.91
317
+ ngram_lm_scale_0.7_attention_scale_4.0 15.91
318
+ ngram_lm_scale_0.7_attention_scale_3.0 15.92
319
+ ngram_lm_scale_0.7_attention_scale_2.2 15.93
320
+ ngram_lm_scale_0.7_attention_scale_2.3 15.93
321
+ ngram_lm_scale_0.7_attention_scale_2.0 15.96
322
+ ngram_lm_scale_0.7_attention_scale_2.1 15.97
323
+ ngram_lm_scale_0.6_attention_scale_0.7 15.98
324
+ ngram_lm_scale_0.7_attention_scale_1.7 15.98
325
+ ngram_lm_scale_0.7_attention_scale_1.9 15.98
326
+ ngram_lm_scale_0.5_attention_scale_0.3 16.0
327
+ ngram_lm_scale_0.7_attention_scale_1.5 16.01
328
+ ngram_lm_scale_0.9_attention_scale_5.0 16.05
329
+ ngram_lm_scale_0.7_attention_scale_1.3 16.06
330
+ ngram_lm_scale_0.6_attention_scale_0.6 16.07
331
+ ngram_lm_scale_0.7_attention_scale_1.2 16.08
332
+ ngram_lm_scale_0.9_attention_scale_4.0 16.1
333
+ ngram_lm_scale_0.7_attention_scale_1.1 16.12
334
+ ngram_lm_scale_0.7_attention_scale_1.0 16.15
335
+ ngram_lm_scale_1.0_attention_scale_5.0 16.17
336
+ ngram_lm_scale_0.9_attention_scale_3.0 16.2
337
+ ngram_lm_scale_0.7_attention_scale_0.9 16.22
338
+ ngram_lm_scale_1.1_attention_scale_5.0 16.22
339
+ ngram_lm_scale_1.0_attention_scale_4.0 16.23
340
+ ngram_lm_scale_0.6_attention_scale_0.5 16.24
341
+ ngram_lm_scale_0.9_attention_scale_2.5 16.28
342
+ ngram_lm_scale_0.9_attention_scale_2.2 16.3
343
+ ngram_lm_scale_0.9_attention_scale_2.3 16.31
344
+ ngram_lm_scale_0.9_attention_scale_2.1 16.32
345
+ ngram_lm_scale_0.9_attention_scale_2.0 16.34
346
+ ngram_lm_scale_1.0_attention_scale_3.0 16.35
347
+ ngram_lm_scale_1.1_attention_scale_4.0 16.35
348
+ ngram_lm_scale_1.2_attention_scale_5.0 16.35
349
+ ngram_lm_scale_0.9_attention_scale_1.9 16.4
350
+ ngram_lm_scale_0.7_attention_scale_0.7 16.43
351
+ ngram_lm_scale_1.0_attention_scale_2.5 16.44
352
+ ngram_lm_scale_0.9_attention_scale_1.7 16.46
353
+ ngram_lm_scale_1.2_attention_scale_4.0 16.47
354
+ ngram_lm_scale_1.3_attention_scale_5.0 16.48
355
+ ngram_lm_scale_1.0_attention_scale_2.3 16.5
356
+ ngram_lm_scale_1.1_attention_scale_3.0 16.51
357
+ ngram_lm_scale_1.0_attention_scale_2.2 16.53
358
+ ngram_lm_scale_1.0_attention_scale_2.1 16.55
359
+ ngram_lm_scale_0.7_attention_scale_0.6 16.59
360
+ ngram_lm_scale_0.9_attention_scale_1.5 16.59
361
+ ngram_lm_scale_1.0_attention_scale_2.0 16.6
362
+ ngram_lm_scale_1.3_attention_scale_4.0 16.61
363
+ ngram_lm_scale_0.5_attention_scale_0.1 16.64
364
+ ngram_lm_scale_1.1_attention_scale_2.5 16.66
365
+ ngram_lm_scale_0.6_attention_scale_0.3 16.67
366
+ ngram_lm_scale_1.0_attention_scale_1.9 16.67
367
+ ngram_lm_scale_1.2_attention_scale_3.0 16.7
368
+ ngram_lm_scale_1.5_attention_scale_5.0 16.7
369
+ ngram_lm_scale_0.5_attention_scale_0.08 16.75
370
+ ngram_lm_scale_0.9_attention_scale_1.3 16.75
371
+ ngram_lm_scale_1.1_attention_scale_2.3 16.77
372
+ ngram_lm_scale_0.7_attention_scale_0.5 16.81
373
+ ngram_lm_scale_1.0_attention_scale_1.7 16.82
374
+ ngram_lm_scale_1.1_attention_scale_2.2 16.85
375
+ ngram_lm_scale_0.9_attention_scale_1.2 16.87
376
+ ngram_lm_scale_1.1_attention_scale_2.1 16.9
377
+ ngram_lm_scale_0.5_attention_scale_0.05 16.91
378
+ ngram_lm_scale_1.3_attention_scale_3.0 16.96
379
+ ngram_lm_scale_1.2_attention_scale_2.5 16.97
380
+ ngram_lm_scale_1.5_attention_scale_4.0 16.98
381
+ ngram_lm_scale_1.1_attention_scale_2.0 17.0
382
+ ngram_lm_scale_1.7_attention_scale_5.0 17.01
383
+ ngram_lm_scale_0.9_attention_scale_1.1 17.05
384
+ ngram_lm_scale_1.0_attention_scale_1.5 17.05
385
+ ngram_lm_scale_1.1_attention_scale_1.9 17.08
386
+ ngram_lm_scale_1.2_attention_scale_2.3 17.12
387
+ ngram_lm_scale_0.5_attention_scale_0.01 17.16
388
+ ngram_lm_scale_1.2_attention_scale_2.2 17.2
389
+ ngram_lm_scale_0.9_attention_scale_1.0 17.21
390
+ ngram_lm_scale_1.1_attention_scale_1.7 17.27
391
+ ngram_lm_scale_1.2_attention_scale_2.1 17.27
392
+ ngram_lm_scale_1.3_attention_scale_2.5 17.27
393
+ ngram_lm_scale_1.0_attention_scale_1.3 17.33
394
+ ngram_lm_scale_0.9_attention_scale_0.9 17.36
395
+ ngram_lm_scale_1.9_attention_scale_5.0 17.36
396
+ ngram_lm_scale_1.2_attention_scale_2.0 17.37
397
+ ngram_lm_scale_1.7_attention_scale_4.0 17.41
398
+ ngram_lm_scale_1.3_attention_scale_2.3 17.46
399
+ ngram_lm_scale_1.0_attention_scale_1.2 17.48
400
+ ngram_lm_scale_1.2_attention_scale_1.9 17.49
401
+ ngram_lm_scale_1.5_attention_scale_3.0 17.55
402
+ ngram_lm_scale_1.1_attention_scale_1.5 17.56
403
+ ngram_lm_scale_2.0_attention_scale_5.0 17.56
404
+ ngram_lm_scale_0.7_attention_scale_0.3 17.57
405
+ ngram_lm_scale_1.3_attention_scale_2.2 17.58
406
+ ngram_lm_scale_1.0_attention_scale_1.1 17.59
407
+ ngram_lm_scale_0.6_attention_scale_0.1 17.62
408
+ ngram_lm_scale_1.3_attention_scale_2.1 17.71
409
+ ngram_lm_scale_2.1_attention_scale_5.0 17.78
410
+ ngram_lm_scale_0.6_attention_scale_0.08 17.79
411
+ ngram_lm_scale_1.2_attention_scale_1.7 17.8
412
+ ngram_lm_scale_1.0_attention_scale_1.0 17.84
413
+ ngram_lm_scale_1.1_attention_scale_1.3 17.84
414
+ ngram_lm_scale_1.3_attention_scale_2.0 17.84
415
+ ngram_lm_scale_0.9_attention_scale_0.7 17.92
416
+ ngram_lm_scale_1.9_attention_scale_4.0 17.93
417
+ ngram_lm_scale_1.3_attention_scale_1.9 17.98
418
+ ngram_lm_scale_2.2_attention_scale_5.0 17.99
419
+ ngram_lm_scale_0.6_attention_scale_0.05 18.07
420
+ ngram_lm_scale_1.1_attention_scale_1.2 18.07
421
+ ngram_lm_scale_1.5_attention_scale_2.5 18.1
422
+ ngram_lm_scale_1.2_attention_scale_1.5 18.12
423
+ ngram_lm_scale_1.0_attention_scale_0.9 18.17
424
+ ngram_lm_scale_2.3_attention_scale_5.0 18.24
425
+ ngram_lm_scale_2.0_attention_scale_4.0 18.25
426
+ ngram_lm_scale_0.9_attention_scale_0.6 18.26
427
+ ngram_lm_scale_1.7_attention_scale_3.0 18.26
428
+ ngram_lm_scale_1.3_attention_scale_1.7 18.35
429
+ ngram_lm_scale_1.5_attention_scale_2.3 18.37
430
+ ngram_lm_scale_1.1_attention_scale_1.1 18.38
431
+ ngram_lm_scale_0.6_attention_scale_0.01 18.41
432
+ ngram_lm_scale_1.5_attention_scale_2.2 18.52
433
+ ngram_lm_scale_2.1_attention_scale_4.0 18.52
434
+ ngram_lm_scale_1.2_attention_scale_1.3 18.59
435
+ ngram_lm_scale_1.5_attention_scale_2.1 18.66
436
+ ngram_lm_scale_1.1_attention_scale_1.0 18.72
437
+ ngram_lm_scale_2.5_attention_scale_5.0 18.72
438
+ ngram_lm_scale_0.9_attention_scale_0.5 18.73
439
+ ngram_lm_scale_2.2_attention_scale_4.0 18.8
440
+ ngram_lm_scale_1.3_attention_scale_1.5 18.81
441
+ ngram_lm_scale_1.5_attention_scale_2.0 18.82
442
+ ngram_lm_scale_1.2_attention_scale_1.2 18.9
443
+ ngram_lm_scale_1.7_attention_scale_2.5 18.9
444
+ ngram_lm_scale_0.7_attention_scale_0.1 18.92
445
+ ngram_lm_scale_1.0_attention_scale_0.7 18.94
446
+ ngram_lm_scale_1.9_attention_scale_3.0 19.04
447
+ ngram_lm_scale_1.1_attention_scale_0.9 19.06
448
+ ngram_lm_scale_1.5_attention_scale_1.9 19.07
449
+ ngram_lm_scale_0.7_attention_scale_0.08 19.14
450
+ ngram_lm_scale_2.3_attention_scale_4.0 19.23
451
+ ngram_lm_scale_1.2_attention_scale_1.1 19.27
452
+ ngram_lm_scale_1.7_attention_scale_2.3 19.38
453
+ ngram_lm_scale_1.3_attention_scale_1.3 19.43
454
+ ngram_lm_scale_1.0_attention_scale_0.6 19.44
455
+ ngram_lm_scale_0.7_attention_scale_0.05 19.5
456
+ ngram_lm_scale_2.0_attention_scale_3.0 19.57
457
+ ngram_lm_scale_1.2_attention_scale_1.0 19.63
458
+ ngram_lm_scale_1.7_attention_scale_2.2 19.67
459
+ ngram_lm_scale_1.5_attention_scale_1.7 19.71
460
+ ngram_lm_scale_1.3_attention_scale_1.2 19.85
461
+ ngram_lm_scale_1.7_attention_scale_2.1 19.96
462
+ ngram_lm_scale_2.5_attention_scale_4.0 19.98
463
+ ngram_lm_scale_0.7_attention_scale_0.01 20.01
464
+ ngram_lm_scale_2.1_attention_scale_3.0 20.02
465
+ ngram_lm_scale_1.0_attention_scale_0.5 20.06
466
+ ngram_lm_scale_1.1_attention_scale_0.7 20.1
467
+ ngram_lm_scale_1.9_attention_scale_2.5 20.13
468
+ ngram_lm_scale_1.2_attention_scale_0.9 20.16
469
+ ngram_lm_scale_0.9_attention_scale_0.3 20.25
470
+ ngram_lm_scale_1.3_attention_scale_1.1 20.25
471
+ ngram_lm_scale_1.7_attention_scale_2.0 20.26
472
+ ngram_lm_scale_3.0_attention_scale_5.0 20.31
473
+ ngram_lm_scale_1.5_attention_scale_1.5 20.44
474
+ ngram_lm_scale_1.7_attention_scale_1.9 20.58
475
+ ngram_lm_scale_2.2_attention_scale_3.0 20.6
476
+ ngram_lm_scale_1.9_attention_scale_2.3 20.74
477
+ ngram_lm_scale_1.1_attention_scale_0.6 20.78
478
+ ngram_lm_scale_2.0_attention_scale_2.5 20.79
479
+ ngram_lm_scale_1.3_attention_scale_1.0 20.84
480
+ ngram_lm_scale_1.9_attention_scale_2.2 21.1
481
+ ngram_lm_scale_2.3_attention_scale_3.0 21.19
482
+ ngram_lm_scale_1.5_attention_scale_1.3 21.36
483
+ ngram_lm_scale_1.7_attention_scale_1.7 21.39
484
+ ngram_lm_scale_1.9_attention_scale_2.1 21.44
485
+ ngram_lm_scale_2.0_attention_scale_2.3 21.44
486
+ ngram_lm_scale_1.3_attention_scale_0.9 21.48
487
+ ngram_lm_scale_1.2_attention_scale_0.7 21.49
488
+ ngram_lm_scale_2.1_attention_scale_2.5 21.49
489
+ ngram_lm_scale_1.1_attention_scale_0.5 21.63
490
+ ngram_lm_scale_2.0_attention_scale_2.2 21.79
491
+ ngram_lm_scale_1.9_attention_scale_2.0 21.81
492
+ ngram_lm_scale_1.0_attention_scale_0.3 21.92
493
+ ngram_lm_scale_1.5_attention_scale_1.2 21.94
494
+ ngram_lm_scale_2.2_attention_scale_2.5 22.14
495
+ ngram_lm_scale_2.1_attention_scale_2.3 22.17
496
+ ngram_lm_scale_2.0_attention_scale_2.1 22.2
497
+ ngram_lm_scale_3.0_attention_scale_4.0 22.21
498
+ ngram_lm_scale_1.9_attention_scale_1.9 22.23
499
+ ngram_lm_scale_1.2_attention_scale_0.6 22.37
500
+ ngram_lm_scale_1.7_attention_scale_1.5 22.37
501
+ ngram_lm_scale_2.5_attention_scale_3.0 22.41
502
+ ngram_lm_scale_1.5_attention_scale_1.1 22.59
503
+ ngram_lm_scale_0.9_attention_scale_0.1 22.6
504
+ ngram_lm_scale_2.1_attention_scale_2.2 22.62
505
+ ngram_lm_scale_2.0_attention_scale_2.0 22.69
506
+ ngram_lm_scale_2.3_attention_scale_2.5 22.91
507
+ ngram_lm_scale_0.9_attention_scale_0.08 22.92
508
+ ngram_lm_scale_1.3_attention_scale_0.7 22.98
509
+ ngram_lm_scale_2.2_attention_scale_2.3 22.98
510
+ ngram_lm_scale_2.1_attention_scale_2.1 23.1
511
+ ngram_lm_scale_2.0_attention_scale_1.9 23.21
512
+ ngram_lm_scale_1.9_attention_scale_1.7 23.32
513
+ ngram_lm_scale_1.2_attention_scale_0.5 23.38
514
+ ngram_lm_scale_1.5_attention_scale_1.0 23.38
515
+ ngram_lm_scale_2.2_attention_scale_2.2 23.43
516
+ ngram_lm_scale_0.9_attention_scale_0.05 23.51
517
+ ngram_lm_scale_2.1_attention_scale_2.0 23.56
518
+ ngram_lm_scale_1.7_attention_scale_1.3 23.71
519
+ ngram_lm_scale_2.3_attention_scale_2.3 23.81
520
+ ngram_lm_scale_1.1_attention_scale_0.3 23.9
521
+ ngram_lm_scale_2.2_attention_scale_2.1 23.95
522
+ ngram_lm_scale_1.3_attention_scale_0.6 24.01
523
+ ngram_lm_scale_2.1_attention_scale_1.9 24.11
524
+ ngram_lm_scale_1.5_attention_scale_0.9 24.18
525
+ ngram_lm_scale_2.3_attention_scale_2.2 24.27
526
+ ngram_lm_scale_0.9_attention_scale_0.01 24.31
527
+ ngram_lm_scale_2.0_attention_scale_1.7 24.32
528
+ ngram_lm_scale_1.7_attention_scale_1.2 24.35
529
+ ngram_lm_scale_4.0_attention_scale_5.0 24.35
530
+ ngram_lm_scale_2.5_attention_scale_2.5 24.46
531
+ ngram_lm_scale_2.2_attention_scale_2.0 24.47
532
+ ngram_lm_scale_1.9_attention_scale_1.5 24.57
533
+ ngram_lm_scale_2.3_attention_scale_2.1 24.84
534
+ ngram_lm_scale_1.0_attention_scale_0.1 24.91
535
+ ngram_lm_scale_2.2_attention_scale_1.9 25.0
536
+ ngram_lm_scale_1.3_attention_scale_0.5 25.02
537
+ ngram_lm_scale_1.7_attention_scale_1.1 25.12
538
+ ngram_lm_scale_2.1_attention_scale_1.7 25.2
539
+ ngram_lm_scale_1.0_attention_scale_0.08 25.3
540
+ ngram_lm_scale_2.3_attention_scale_2.0 25.32
541
+ ngram_lm_scale_2.5_attention_scale_2.3 25.38
542
+ ngram_lm_scale_2.0_attention_scale_1.5 25.49
543
+ ngram_lm_scale_3.0_attention_scale_3.0 25.62
544
+ ngram_lm_scale_2.5_attention_scale_2.2 25.8
545
+ ngram_lm_scale_2.3_attention_scale_1.9 25.81
546
+ ngram_lm_scale_1.0_attention_scale_0.05 25.82
547
+ ngram_lm_scale_1.9_attention_scale_1.3 25.88
548
+ ngram_lm_scale_1.7_attention_scale_1.0 25.92
549
+ ngram_lm_scale_1.2_attention_scale_0.3 25.93
550
+ ngram_lm_scale_1.5_attention_scale_0.7 26.0
551
+ ngram_lm_scale_2.2_attention_scale_1.7 26.12
552
+ ngram_lm_scale_2.5_attention_scale_2.1 26.33
553
+ ngram_lm_scale_2.1_attention_scale_1.5 26.49
554
+ ngram_lm_scale_1.9_attention_scale_1.2 26.64
555
+ ngram_lm_scale_1.0_attention_scale_0.01 26.66
556
+ ngram_lm_scale_1.7_attention_scale_0.9 26.77
557
+ ngram_lm_scale_2.5_attention_scale_2.0 26.91
558
+ ngram_lm_scale_2.0_attention_scale_1.3 26.93
559
+ ngram_lm_scale_2.3_attention_scale_1.7 27.04
560
+ ngram_lm_scale_1.5_attention_scale_0.6 27.13
561
+ ngram_lm_scale_4.0_attention_scale_4.0 27.2
562
+ ngram_lm_scale_1.1_attention_scale_0.1 27.21
563
+ ngram_lm_scale_2.2_attention_scale_1.5 27.45
564
+ ngram_lm_scale_2.5_attention_scale_1.9 27.46
565
+ ngram_lm_scale_1.9_attention_scale_1.1 27.47
566
+ ngram_lm_scale_1.1_attention_scale_0.08 27.58
567
+ ngram_lm_scale_1.3_attention_scale_0.3 27.75
568
+ ngram_lm_scale_2.0_attention_scale_1.2 27.76
569
+ ngram_lm_scale_3.0_attention_scale_2.5 27.82
570
+ ngram_lm_scale_2.1_attention_scale_1.3 27.95
571
+ ngram_lm_scale_1.1_attention_scale_0.05 28.1
572
+ ngram_lm_scale_5.0_attention_scale_5.0 28.15
573
+ ngram_lm_scale_2.3_attention_scale_1.5 28.33
574
+ ngram_lm_scale_1.5_attention_scale_0.5 28.37
575
+ ngram_lm_scale_1.9_attention_scale_1.0 28.4
576
+ ngram_lm_scale_2.0_attention_scale_1.1 28.59
577
+ ngram_lm_scale_2.5_attention_scale_1.7 28.61
578
+ ngram_lm_scale_3.0_attention_scale_2.3 28.67
579
+ ngram_lm_scale_2.1_attention_scale_1.2 28.76
580
+ ngram_lm_scale_1.1_attention_scale_0.01 28.87
581
+ ngram_lm_scale_2.2_attention_scale_1.3 28.88
582
+ ngram_lm_scale_1.7_attention_scale_0.7 28.94
583
+ ngram_lm_scale_1.2_attention_scale_0.1 29.15
584
+ ngram_lm_scale_3.0_attention_scale_2.2 29.17
585
+ ngram_lm_scale_1.9_attention_scale_0.9 29.29
586
+ ngram_lm_scale_2.0_attention_scale_1.0 29.41
587
+ ngram_lm_scale_1.2_attention_scale_0.08 29.49
588
+ ngram_lm_scale_2.1_attention_scale_1.1 29.55
589
+ ngram_lm_scale_2.2_attention_scale_1.2 29.65
590
+ ngram_lm_scale_3.0_attention_scale_2.1 29.67
591
+ ngram_lm_scale_2.3_attention_scale_1.3 29.75
592
+ ngram_lm_scale_2.5_attention_scale_1.5 29.93
593
+ ngram_lm_scale_1.7_attention_scale_0.6 29.94
594
+ ngram_lm_scale_1.2_attention_scale_0.05 30.03
595
+ ngram_lm_scale_3.0_attention_scale_2.0 30.2
596
+ ngram_lm_scale_2.0_attention_scale_0.9 30.31
597
+ ngram_lm_scale_2.1_attention_scale_1.0 30.35
598
+ ngram_lm_scale_2.2_attention_scale_1.1 30.38
599
+ ngram_lm_scale_2.3_attention_scale_1.2 30.44
600
+ ngram_lm_scale_4.0_attention_scale_3.0 30.53
601
+ ngram_lm_scale_3.0_attention_scale_1.9 30.65
602
+ ngram_lm_scale_5.0_attention_scale_4.0 30.71
603
+ ngram_lm_scale_1.2_attention_scale_0.01 30.77
604
+ ngram_lm_scale_1.3_attention_scale_0.1 30.86
605
+ ngram_lm_scale_1.5_attention_scale_0.3 31.01
606
+ ngram_lm_scale_1.7_attention_scale_0.5 31.1
607
+ ngram_lm_scale_2.1_attention_scale_0.9 31.15
608
+ ngram_lm_scale_1.3_attention_scale_0.08 31.19
609
+ ngram_lm_scale_2.2_attention_scale_1.0 31.19
610
+ ngram_lm_scale_2.5_attention_scale_1.3 31.2
611
+ ngram_lm_scale_2.3_attention_scale_1.1 31.21
612
+ ngram_lm_scale_1.9_attention_scale_0.7 31.23
613
+ ngram_lm_scale_1.3_attention_scale_0.05 31.68
614
+ ngram_lm_scale_3.0_attention_scale_1.7 31.69
615
+ ngram_lm_scale_2.5_attention_scale_1.2 31.84
616
+ ngram_lm_scale_2.3_attention_scale_1.0 31.88
617
+ ngram_lm_scale_2.2_attention_scale_0.9 31.91
618
+ ngram_lm_scale_2.0_attention_scale_0.7 32.0
619
+ ngram_lm_scale_1.9_attention_scale_0.6 32.05
620
+ ngram_lm_scale_1.3_attention_scale_0.01 32.25
621
+ ngram_lm_scale_4.0_attention_scale_2.5 32.3
622
+ ngram_lm_scale_2.5_attention_scale_1.1 32.43
623
+ ngram_lm_scale_2.3_attention_scale_0.9 32.57
624
+ ngram_lm_scale_3.0_attention_scale_1.5 32.7
625
+ ngram_lm_scale_2.1_attention_scale_0.7 32.73
626
+ ngram_lm_scale_2.0_attention_scale_0.6 32.8
627
+ ngram_lm_scale_1.9_attention_scale_0.5 32.9
628
+ ngram_lm_scale_4.0_attention_scale_2.3 33.0
629
+ ngram_lm_scale_2.5_attention_scale_1.0 33.08
630
+ ngram_lm_scale_1.7_attention_scale_0.3 33.11
631
+ ngram_lm_scale_4.0_attention_scale_2.2 33.33
632
+ ngram_lm_scale_2.2_attention_scale_0.7 33.34
633
+ ngram_lm_scale_1.5_attention_scale_0.1 33.35
634
+ ngram_lm_scale_5.0_attention_scale_3.0 33.43
635
+ ngram_lm_scale_2.1_attention_scale_0.6 33.46
636
+ ngram_lm_scale_1.5_attention_scale_0.08 33.58
637
+ ngram_lm_scale_2.0_attention_scale_0.5 33.62
638
+ ngram_lm_scale_2.5_attention_scale_0.9 33.66
639
+ ngram_lm_scale_3.0_attention_scale_1.3 33.71
640
+ ngram_lm_scale_4.0_attention_scale_2.1 33.71
641
+ ngram_lm_scale_2.3_attention_scale_0.7 33.86
642
+ ngram_lm_scale_1.5_attention_scale_0.05 33.91
643
+ ngram_lm_scale_2.2_attention_scale_0.6 33.98
644
+ ngram_lm_scale_4.0_attention_scale_2.0 34.03
645
+ ngram_lm_scale_3.0_attention_scale_1.2 34.1
646
+ ngram_lm_scale_2.1_attention_scale_0.5 34.15
647
+ ngram_lm_scale_1.5_attention_scale_0.01 34.29
648
+ ngram_lm_scale_4.0_attention_scale_1.9 34.35
649
+ ngram_lm_scale_1.9_attention_scale_0.3 34.45
650
+ ngram_lm_scale_2.3_attention_scale_0.6 34.47
651
+ ngram_lm_scale_3.0_attention_scale_1.1 34.53
652
+ ngram_lm_scale_2.2_attention_scale_0.5 34.6
653
+ ngram_lm_scale_2.5_attention_scale_0.7 34.69
654
+ ngram_lm_scale_5.0_attention_scale_2.5 34.7
655
+ ngram_lm_scale_1.7_attention_scale_0.1 34.92
656
+ ngram_lm_scale_3.0_attention_scale_1.0 34.92
657
+ ngram_lm_scale_2.0_attention_scale_0.3 34.94
658
+ ngram_lm_scale_4.0_attention_scale_1.7 34.96
659
+ ngram_lm_scale_2.3_attention_scale_0.5 34.98
660
+ ngram_lm_scale_1.7_attention_scale_0.08 35.07
661
+ ngram_lm_scale_5.0_attention_scale_2.3 35.14
662
+ ngram_lm_scale_2.5_attention_scale_0.6 35.18
663
+ ngram_lm_scale_1.7_attention_scale_0.05 35.27
664
+ ngram_lm_scale_3.0_attention_scale_0.9 35.35
665
+ ngram_lm_scale_5.0_attention_scale_2.2 35.36
666
+ ngram_lm_scale_2.1_attention_scale_0.3 35.39
667
+ ngram_lm_scale_4.0_attention_scale_1.5 35.48
668
+ ngram_lm_scale_5.0_attention_scale_2.1 35.53
669
+ ngram_lm_scale_1.7_attention_scale_0.01 35.6
670
+ ngram_lm_scale_2.5_attention_scale_0.5 35.66
671
+ ngram_lm_scale_2.2_attention_scale_0.3 35.74
672
+ ngram_lm_scale_5.0_attention_scale_2.0 35.74
673
+ ngram_lm_scale_1.9_attention_scale_0.1 35.84
674
+ ngram_lm_scale_1.9_attention_scale_0.08 35.95
675
+ ngram_lm_scale_5.0_attention_scale_1.9 35.97
676
+ ngram_lm_scale_4.0_attention_scale_1.3 36.01
677
+ ngram_lm_scale_3.0_attention_scale_0.7 36.05
678
+ ngram_lm_scale_2.3_attention_scale_0.3 36.08
679
+ ngram_lm_scale_1.9_attention_scale_0.05 36.14
680
+ ngram_lm_scale_2.0_attention_scale_0.1 36.17
681
+ ngram_lm_scale_4.0_attention_scale_1.2 36.24
682
+ ngram_lm_scale_2.0_attention_scale_0.08 36.27
683
+ ngram_lm_scale_5.0_attention_scale_1.7 36.31
684
+ ngram_lm_scale_1.9_attention_scale_0.01 36.34
685
+ ngram_lm_scale_2.0_attention_scale_0.05 36.42
686
+ ngram_lm_scale_3.0_attention_scale_0.6 36.43
687
+ ngram_lm_scale_2.1_attention_scale_0.1 36.46
688
+ ngram_lm_scale_4.0_attention_scale_1.1 36.48
689
+ ngram_lm_scale_2.5_attention_scale_0.3 36.55
690
+ ngram_lm_scale_2.1_attention_scale_0.08 36.57
691
+ ngram_lm_scale_2.0_attention_scale_0.01 36.63
692
+ ngram_lm_scale_2.1_attention_scale_0.05 36.68
693
+ ngram_lm_scale_2.2_attention_scale_0.1 36.72
694
+ ngram_lm_scale_5.0_attention_scale_1.5 36.74
695
+ ngram_lm_scale_4.0_attention_scale_1.0 36.75
696
+ ngram_lm_scale_3.0_attention_scale_0.5 36.76
697
+ ngram_lm_scale_2.2_attention_scale_0.08 36.79
698
+ ngram_lm_scale_2.1_attention_scale_0.01 36.9
699
+ ngram_lm_scale_2.2_attention_scale_0.05 36.96
700
+ ngram_lm_scale_2.3_attention_scale_0.1 36.96
701
+ ngram_lm_scale_4.0_attention_scale_0.9 36.97
702
+ ngram_lm_scale_2.3_attention_scale_0.08 37.03
703
+ ngram_lm_scale_5.0_attention_scale_1.3 37.07
704
+ ngram_lm_scale_2.2_attention_scale_0.01 37.11
705
+ ngram_lm_scale_2.3_attention_scale_0.05 37.12
706
+ ngram_lm_scale_5.0_attention_scale_1.2 37.24
707
+ ngram_lm_scale_2.3_attention_scale_0.01 37.26
708
+ ngram_lm_scale_2.5_attention_scale_0.1 37.29
709
+ ngram_lm_scale_3.0_attention_scale_0.3 37.33
710
+ ngram_lm_scale_2.5_attention_scale_0.08 37.35
711
+ ngram_lm_scale_4.0_attention_scale_0.7 37.39
712
+ ngram_lm_scale_5.0_attention_scale_1.1 37.4
713
+ ngram_lm_scale_2.5_attention_scale_0.05 37.42
714
+ ngram_lm_scale_2.5_attention_scale_0.01 37.55
715
+ ngram_lm_scale_4.0_attention_scale_0.6 37.58
716
+ ngram_lm_scale_5.0_attention_scale_1.0 37.58
717
+ ngram_lm_scale_5.0_attention_scale_0.9 37.76
718
+ ngram_lm_scale_4.0_attention_scale_0.5 37.78
719
+ ngram_lm_scale_3.0_attention_scale_0.1 37.9
720
+ ngram_lm_scale_3.0_attention_scale_0.08 37.94
721
+ ngram_lm_scale_3.0_attention_scale_0.05 38.03
722
+ ngram_lm_scale_5.0_attention_scale_0.7 38.05
723
+ ngram_lm_scale_3.0_attention_scale_0.01 38.13
724
+ ngram_lm_scale_5.0_attention_scale_0.6 38.2
725
+ ngram_lm_scale_4.0_attention_scale_0.3 38.21
726
+ ngram_lm_scale_5.0_attention_scale_0.5 38.37
727
+ ngram_lm_scale_4.0_attention_scale_0.1 38.57
728
+ ngram_lm_scale_4.0_attention_scale_0.08 38.61
729
+ ngram_lm_scale_5.0_attention_scale_0.3 38.65
730
+ ngram_lm_scale_4.0_attention_scale_0.05 38.67
731
+ ngram_lm_scale_4.0_attention_scale_0.01 38.74
732
+ ngram_lm_scale_5.0_attention_scale_0.1 38.93
733
+ ngram_lm_scale_5.0_attention_scale_0.08 38.97
734
+ ngram_lm_scale_5.0_attention_scale_0.05 39.01
735
+ ngram_lm_scale_5.0_attention_scale_0.01 39.06
736
+
737
+ 2022-06-25 17:31:30,127 INFO [decode.py:483] batch 0/?, cuts processed until now is 13
738
+ 2022-06-26 00:14:14,317 INFO [decode.py:483] batch 100/?, cuts processed until now is 1548
739
+ 2022-06-26 05:48:34,635 INFO [decode.py:733] Caught exception:
740
+
741
+ Some bad things happened. Please read the above error messages and stack
742
+ trace. If you are using Python, the following command may be helpful:
743
+
744
+ gdb --args python /path/to/your/code.py
745
+
746
+ (You can use `gdb` to debug the code. Please consider compiling
747
+ a debug version of k2.).
748
+
749
+ If you are unable to fix it, please open an issue at:
750
+
751
+ https://github.com/k2-fsa/k2/issues/new
752
+
753
+
754
+ 2022-06-26 05:48:34,636 INFO [decode.py:734] num_arcs before pruning: 6585058
755
+ 2022-06-26 05:48:34,636 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
756
+ 2022-06-26 05:48:35,670 INFO [decode.py:747] num_arcs after pruning: 26992
757
+ 2022-06-26 06:38:15,357 INFO [decode.py:483] batch 200/?, cuts processed until now is 3195
758
+ 2022-06-26 07:23:18,776 INFO [decode.py:733] Caught exception:
759
+
760
+ Some bad things happened. Please read the above error messages and stack
761
+ trace. If you are using Python, the following command may be helpful:
762
+
763
+ gdb --args python /path/to/your/code.py
764
+
765
+ (You can use `gdb` to debug the code. Please consider compiling
766
+ a debug version of k2.).
767
+
768
+ If you are unable to fix it, please open an issue at:
769
+
770
+ https://github.com/k2-fsa/k2/issues/new
771
+
772
+
773
+ 2022-06-26 07:23:18,777 INFO [decode.py:734] num_arcs before pruning: 5733129
774
+ 2022-06-26 07:23:18,778 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
775
+ 2022-06-26 07:23:19,605 INFO [decode.py:747] num_arcs after pruning: 28851
776
+ 2022-06-26 07:50:08,702 INFO [decode.py:733] Caught exception:
777
+
778
+ Some bad things happened. Please read the above error messages and stack
779
+ trace. If you are using Python, the following command may be helpful:
780
+
781
+ gdb --args python /path/to/your/code.py
782
+
783
+ (You can use `gdb` to debug the code. Please consider compiling
784
+ a debug version of k2.).
785
+
786
+ If you are unable to fix it, please open an issue at:
787
+
788
+ https://github.com/k2-fsa/k2/issues/new
789
+
790
+
791
+ 2022-06-26 07:50:08,703 INFO [decode.py:734] num_arcs before pruning: 6461523
792
+ 2022-06-26 07:50:08,703 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
793
+ 2022-06-26 07:50:09,674 INFO [decode.py:747] num_arcs after pruning: 22617
794
+ 2022-06-26 09:38:10,983 INFO [decode.py:733] Caught exception:
795
+
796
+ Some bad things happened. Please read the above error messages and stack
797
+ trace. If you are using Python, the following command may be helpful:
798
+
799
+ gdb --args python /path/to/your/code.py
800
+
801
+ (You can use `gdb` to debug the code. Please consider compiling
802
+ a debug version of k2.).
803
+
804
+ If you are unable to fix it, please open an issue at:
805
+
806
+ https://github.com/k2-fsa/k2/issues/new
807
+
808
+
809
+ 2022-06-26 09:38:10,984 INFO [decode.py:734] num_arcs before pruning: 5956089
810
+ 2022-06-26 09:38:10,984 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
811
+ 2022-06-26 09:38:11,799 INFO [decode.py:747] num_arcs after pruning: 20569
812
+ 2022-06-26 11:43:09,309 INFO [decode.py:733] Caught exception:
813
+
814
+ Some bad things happened. Please read the above error messages and stack
815
+ trace. If you are using Python, the following command may be helpful:
816
+
817
+ gdb --args python /path/to/your/code.py
818
+
819
+ (You can use `gdb` to debug the code. Please consider compiling
820
+ a debug version of k2.).
821
+
822
+ If you are unable to fix it, please open an issue at:
823
+
824
+ https://github.com/k2-fsa/k2/issues/new
825
+
826
+
827
+ 2022-06-26 11:43:09,311 INFO [decode.py:734] num_arcs before pruning: 4422748
828
+ 2022-06-26 11:43:09,311 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
829
+ 2022-06-26 11:43:09,913 INFO [decode.py:747] num_arcs after pruning: 16921
830
+ 2022-06-26 11:59:44,567 INFO [decode.py:733] Caught exception:
831
+
832
+ Some bad things happened. Please read the above error messages and stack
833
+ trace. If you are using Python, the following command may be helpful:
834
+
835
+ gdb --args python /path/to/your/code.py
836
+
837
+ (You can use `gdb` to debug the code. Please consider compiling
838
+ a debug version of k2.).
839
+
840
+ If you are unable to fix it, please open an issue at:
841
+
842
+ https://github.com/k2-fsa/k2/issues/new
843
+
844
+
845
+ 2022-06-26 11:59:44,568 INFO [decode.py:734] num_arcs before pruning: 5991196
846
+ 2022-06-26 11:59:44,568 INFO [decode.py:737] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
847
+ 2022-06-26 11:59:45,380 INFO [decode.py:747] num_arcs after pruning: 35987
848
+ 2022-06-26 14:20:44,143 INFO [decode.py:483] batch 300/?, cuts processed until now is 4803
849
+ 2022-06-26 15:43:53,294 INFO [decode.py:532]
850
+ For dev, WER of different settings are:
851
+ ngram_lm_scale_0.01_attention_scale_0.5 15.78 best for dev
852
+ ngram_lm_scale_0.01_attention_scale_0.6 15.78
853
+ ngram_lm_scale_0.01_attention_scale_0.7 15.78
854
+ ngram_lm_scale_0.05_attention_scale_0.6 15.79
855
+ ngram_lm_scale_0.01_attention_scale_0.9 15.8
856
+ ngram_lm_scale_0.05_attention_scale_0.5 15.81
857
+ ngram_lm_scale_0.05_attention_scale_0.7 15.81
858
+ ngram_lm_scale_0.08_attention_scale_0.7 15.81
859
+ ngram_lm_scale_0.01_attention_scale_0.3 15.82
860
+ ngram_lm_scale_0.01_attention_scale_1.0 15.82
861
+ ngram_lm_scale_0.05_attention_scale_0.9 15.82
862
+ ngram_lm_scale_0.08_attention_scale_0.5 15.82
863
+ ngram_lm_scale_0.08_attention_scale_0.6 15.82
864
+ ngram_lm_scale_0.08_attention_scale_0.9 15.82
865
+ ngram_lm_scale_0.1_attention_scale_0.6 15.82
866
+ ngram_lm_scale_0.1_attention_scale_0.7 15.83
867
+ ngram_lm_scale_0.05_attention_scale_1.0 15.84
868
+ ngram_lm_scale_0.1_attention_scale_0.5 15.84
869
+ ngram_lm_scale_0.01_attention_scale_1.1 15.85
870
+ ngram_lm_scale_0.05_attention_scale_0.3 15.85
871
+ ngram_lm_scale_0.05_attention_scale_1.1 15.85
872
+ ngram_lm_scale_0.08_attention_scale_0.3 15.85
873
+ ngram_lm_scale_0.1_attention_scale_0.3 15.85
874
+ ngram_lm_scale_0.1_attention_scale_0.9 15.86
875
+ ngram_lm_scale_0.1_attention_scale_1.0 15.86
876
+ ngram_lm_scale_0.01_attention_scale_1.2 15.87
877
+ ngram_lm_scale_0.08_attention_scale_1.0 15.87
878
+ ngram_lm_scale_0.01_attention_scale_1.3 15.89
879
+ ngram_lm_scale_0.08_attention_scale_1.1 15.89
880
+ ngram_lm_scale_0.01_attention_scale_0.1 15.9
881
+ ngram_lm_scale_0.05_attention_scale_1.2 15.9
882
+ ngram_lm_scale_0.1_attention_scale_1.1 15.9
883
+ ngram_lm_scale_0.05_attention_scale_1.3 15.91
884
+ ngram_lm_scale_0.08_attention_scale_1.2 15.91
885
+ ngram_lm_scale_0.01_attention_scale_0.08 15.92
886
+ ngram_lm_scale_0.01_attention_scale_1.5 15.92
887
+ ngram_lm_scale_0.05_attention_scale_0.1 15.93
888
+ ngram_lm_scale_0.1_attention_scale_1.2 15.93
889
+ ngram_lm_scale_0.05_attention_scale_0.08 15.94
890
+ ngram_lm_scale_0.08_attention_scale_0.1 15.94
891
+ ngram_lm_scale_0.08_attention_scale_1.3 15.94
892
+ ngram_lm_scale_0.08_attention_scale_0.08 15.96
893
+ ngram_lm_scale_0.1_attention_scale_0.1 15.96
894
+ ngram_lm_scale_0.1_attention_scale_1.3 15.96
895
+ ngram_lm_scale_0.01_attention_scale_1.7 15.97
896
+ ngram_lm_scale_0.05_attention_scale_1.5 15.97
897
+ ngram_lm_scale_0.01_attention_scale_0.05 15.99
898
+ ngram_lm_scale_0.05_attention_scale_0.05 15.99
899
+ ngram_lm_scale_0.05_attention_scale_1.7 15.99
900
+ ngram_lm_scale_0.08_attention_scale_1.5 15.99
901
+ ngram_lm_scale_0.1_attention_scale_0.08 15.99
902
+ ngram_lm_scale_0.01_attention_scale_0.01 16.0
903
+ ngram_lm_scale_0.01_attention_scale_1.9 16.0
904
+ ngram_lm_scale_0.08_attention_scale_0.05 16.0
905
+ ngram_lm_scale_0.01_attention_scale_2.0 16.01
906
+ ngram_lm_scale_0.1_attention_scale_1.5 16.01
907
+ ngram_lm_scale_0.1_attention_scale_0.05 16.02
908
+ ngram_lm_scale_0.05_attention_scale_1.9 16.03
909
+ ngram_lm_scale_0.08_attention_scale_1.7 16.03
910
+ ngram_lm_scale_0.08_attention_scale_1.9 16.03
911
+ ngram_lm_scale_0.01_attention_scale_2.1 16.04
912
+ ngram_lm_scale_0.05_attention_scale_2.0 16.04
913
+ ngram_lm_scale_0.1_attention_scale_1.7 16.04
914
+ ngram_lm_scale_0.01_attention_scale_2.2 16.05
915
+ ngram_lm_scale_0.05_attention_scale_0.01 16.05
916
+ ngram_lm_scale_0.05_attention_scale_2.1 16.05
917
+ ngram_lm_scale_0.1_attention_scale_1.9 16.05
918
+ ngram_lm_scale_0.08_attention_scale_2.0 16.06
919
+ ngram_lm_scale_0.01_attention_scale_2.3 16.07
920
+ ngram_lm_scale_0.1_attention_scale_2.0 16.07
921
+ ngram_lm_scale_0.3_attention_scale_0.9 16.07
922
+ ngram_lm_scale_0.05_attention_scale_2.2 16.08
923
+ ngram_lm_scale_0.08_attention_scale_0.01 16.08
924
+ ngram_lm_scale_0.08_attention_scale_2.1 16.09
925
+ ngram_lm_scale_0.08_attention_scale_2.2 16.09
926
+ ngram_lm_scale_0.1_attention_scale_2.1 16.09
927
+ ngram_lm_scale_0.1_attention_scale_2.2 16.09
928
+ ngram_lm_scale_0.3_attention_scale_1.0 16.09
929
+ ngram_lm_scale_0.01_attention_scale_2.5 16.1
930
+ ngram_lm_scale_0.05_attention_scale_2.3 16.1
931
+ ngram_lm_scale_0.08_attention_scale_2.3 16.1
932
+ ngram_lm_scale_0.3_attention_scale_1.1 16.1
933
+ ngram_lm_scale_0.3_attention_scale_0.7 16.11
934
+ ngram_lm_scale_0.05_attention_scale_2.5 16.12
935
+ ngram_lm_scale_0.1_attention_scale_0.01 16.12
936
+ ngram_lm_scale_0.1_attention_scale_2.3 16.12
937
+ ngram_lm_scale_0.3_attention_scale_0.6 16.12
938
+ ngram_lm_scale_0.3_attention_scale_1.2 16.12
939
+ ngram_lm_scale_0.3_attention_scale_1.3 16.13
940
+ ngram_lm_scale_0.08_attention_scale_2.5 16.14
941
+ ngram_lm_scale_0.1_attention_scale_2.5 16.14
942
+ ngram_lm_scale_0.3_attention_scale_1.5 16.14
943
+ ngram_lm_scale_0.3_attention_scale_0.5 16.16
944
+ ngram_lm_scale_0.01_attention_scale_3.0 16.18
945
+ ngram_lm_scale_0.3_attention_scale_1.7 16.2
946
+ ngram_lm_scale_0.3_attention_scale_0.3 16.21
947
+ ngram_lm_scale_0.05_attention_scale_3.0 16.22
948
+ ngram_lm_scale_0.08_attention_scale_3.0 16.24
949
+ ngram_lm_scale_0.3_attention_scale_1.9 16.25
950
+ ngram_lm_scale_0.1_attention_scale_3.0 16.27
951
+ ngram_lm_scale_0.3_attention_scale_2.0 16.28
952
+ ngram_lm_scale_0.3_attention_scale_2.1 16.3
953
+ ngram_lm_scale_0.3_attention_scale_2.2 16.31
954
+ ngram_lm_scale_0.3_attention_scale_2.3 16.32
955
+ ngram_lm_scale_0.01_attention_scale_4.0 16.34
956
+ ngram_lm_scale_0.05_attention_scale_4.0 16.35
957
+ ngram_lm_scale_0.3_attention_scale_2.5 16.35
958
+ ngram_lm_scale_0.08_attention_scale_4.0 16.37
959
+ ngram_lm_scale_0.1_attention_scale_4.0 16.39
960
+ ngram_lm_scale_0.3_attention_scale_0.1 16.4
961
+ ngram_lm_scale_0.3_attention_scale_3.0 16.41
962
+ ngram_lm_scale_0.3_attention_scale_0.08 16.42
963
+ ngram_lm_scale_0.01_attention_scale_5.0 16.43
964
+ ngram_lm_scale_0.05_attention_scale_5.0 16.45
965
+ ngram_lm_scale_0.08_attention_scale_5.0 16.47
966
+ ngram_lm_scale_0.1_attention_scale_5.0 16.49
967
+ ngram_lm_scale_0.3_attention_scale_0.05 16.49
968
+ ngram_lm_scale_0.5_attention_scale_1.2 16.53
969
+ ngram_lm_scale_0.3_attention_scale_4.0 16.54
970
+ ngram_lm_scale_0.5_attention_scale_0.9 16.54
971
+ ngram_lm_scale_0.5_attention_scale_1.3 16.54
972
+ ngram_lm_scale_0.5_attention_scale_1.0 16.55
973
+ ngram_lm_scale_0.5_attention_scale_1.1 16.55
974
+ ngram_lm_scale_0.5_attention_scale_1.9 16.55
975
+ ngram_lm_scale_0.5_attention_scale_1.7 16.56
976
+ ngram_lm_scale_0.5_attention_scale_2.0 16.56
977
+ ngram_lm_scale_0.5_attention_scale_1.5 16.57
978
+ ngram_lm_scale_0.5_attention_scale_2.1 16.58
979
+ ngram_lm_scale_0.5_attention_scale_2.2 16.58
980
+ ngram_lm_scale_0.3_attention_scale_0.01 16.59
981
+ ngram_lm_scale_0.5_attention_scale_2.3 16.6
982
+ ngram_lm_scale_0.5_attention_scale_2.5 16.6
983
+ ngram_lm_scale_0.3_attention_scale_5.0 16.61
984
+ ngram_lm_scale_0.5_attention_scale_0.7 16.62
985
+ ngram_lm_scale_0.5_attention_scale_3.0 16.65
986
+ ngram_lm_scale_0.5_attention_scale_0.6 16.69
987
+ ngram_lm_scale_0.5_attention_scale_4.0 16.72
988
+ ngram_lm_scale_0.5_attention_scale_0.5 16.74
989
+ ngram_lm_scale_0.5_attention_scale_5.0 16.75
990
+ ngram_lm_scale_0.6_attention_scale_2.1 16.79
991
+ ngram_lm_scale_0.6_attention_scale_2.2 16.79
992
+ ngram_lm_scale_0.6_attention_scale_2.3 16.79
993
+ ngram_lm_scale_0.6_attention_scale_3.0 16.79
994
+ ngram_lm_scale_0.6_attention_scale_2.0 16.81
995
+ ngram_lm_scale_0.6_attention_scale_2.5 16.81
996
+ ngram_lm_scale_0.6_attention_scale_1.9 16.82
997
+ ngram_lm_scale_0.6_attention_scale_1.7 16.83
998
+ ngram_lm_scale_0.6_attention_scale_4.0 16.83
999
+ ngram_lm_scale_0.6_attention_scale_1.5 16.85
1000
+ ngram_lm_scale_0.6_attention_scale_1.3 16.87
1001
+ ngram_lm_scale_0.6_attention_scale_1.2 16.88
1002
+ ngram_lm_scale_0.6_attention_scale_1.1 16.89
1003
+ ngram_lm_scale_0.6_attention_scale_5.0 16.89
1004
+ ngram_lm_scale_0.6_attention_scale_1.0 16.94
1005
+ ngram_lm_scale_0.7_attention_scale_4.0 16.95
1006
+ ngram_lm_scale_0.6_attention_scale_0.9 16.96
1007
+ ngram_lm_scale_0.7_attention_scale_3.0 16.97
1008
+ ngram_lm_scale_0.7_attention_scale_5.0 16.97
1009
+ ngram_lm_scale_0.5_attention_scale_0.3 16.98
1010
+ ngram_lm_scale_0.7_attention_scale_2.5 17.0
1011
+ ngram_lm_scale_0.7_attention_scale_2.1 17.02
1012
+ ngram_lm_scale_0.7_attention_scale_2.2 17.02
1013
+ ngram_lm_scale_0.7_attention_scale_2.3 17.02
1014
+ ngram_lm_scale_0.7_attention_scale_2.0 17.04
1015
+ ngram_lm_scale_0.7_attention_scale_1.9 17.06
1016
+ ngram_lm_scale_0.7_attention_scale_1.7 17.09
1017
+ ngram_lm_scale_0.6_attention_scale_0.7 17.1
1018
+ ngram_lm_scale_0.7_attention_scale_1.5 17.15
1019
+ ngram_lm_scale_0.9_attention_scale_5.0 17.18
1020
+ ngram_lm_scale_0.7_attention_scale_1.3 17.2
1021
+ ngram_lm_scale_0.6_attention_scale_0.6 17.23
1022
+ ngram_lm_scale_0.7_attention_scale_1.2 17.24
1023
+ ngram_lm_scale_0.9_attention_scale_4.0 17.24
1024
+ ngram_lm_scale_0.7_attention_scale_1.1 17.26
1025
+ ngram_lm_scale_1.0_attention_scale_5.0 17.28
1026
+ ngram_lm_scale_0.6_attention_scale_0.5 17.31
1027
+ ngram_lm_scale_0.7_attention_scale_1.0 17.33
1028
+ ngram_lm_scale_0.9_attention_scale_3.0 17.35
1029
+ ngram_lm_scale_1.0_attention_scale_4.0 17.38
1030
+ ngram_lm_scale_1.1_attention_scale_5.0 17.43
1031
+ ngram_lm_scale_0.7_attention_scale_0.9 17.45
1032
+ ngram_lm_scale_0.9_attention_scale_2.5 17.47
1033
+ ngram_lm_scale_0.9_attention_scale_2.3 17.53
1034
+ ngram_lm_scale_1.1_attention_scale_4.0 17.55
1035
+ ngram_lm_scale_0.9_attention_scale_2.2 17.57
1036
+ ngram_lm_scale_1.2_attention_scale_5.0 17.57
1037
+ ngram_lm_scale_1.0_attention_scale_3.0 17.59
1038
+ ngram_lm_scale_0.9_attention_scale_2.1 17.61
1039
+ ngram_lm_scale_0.9_attention_scale_2.0 17.64
1040
+ ngram_lm_scale_0.5_attention_scale_0.1 17.65
1041
+ ngram_lm_scale_0.7_attention_scale_0.7 17.67
1042
+ ngram_lm_scale_0.9_attention_scale_1.9 17.69
1043
+ ngram_lm_scale_1.3_attention_scale_5.0 17.69
1044
+ ngram_lm_scale_1.2_attention_scale_4.0 17.73
1045
+ ngram_lm_scale_0.5_attention_scale_0.08 17.76
1046
+ ngram_lm_scale_1.0_attention_scale_2.5 17.76
1047
+ ngram_lm_scale_0.6_attention_scale_0.3 17.78
1048
+ ngram_lm_scale_0.7_attention_scale_0.6 17.8
1049
+ ngram_lm_scale_1.0_attention_scale_2.3 17.81
1050
+ ngram_lm_scale_1.1_attention_scale_3.0 17.82
1051
+ ngram_lm_scale_0.9_attention_scale_1.7 17.83
1052
+ ngram_lm_scale_1.0_attention_scale_2.2 17.87
1053
+ ngram_lm_scale_1.3_attention_scale_4.0 17.9
1054
+ ngram_lm_scale_1.0_attention_scale_2.1 17.94
1055
+ ngram_lm_scale_0.5_attention_scale_0.05 17.95
1056
+ ngram_lm_scale_0.9_attention_scale_1.5 17.96
1057
+ ngram_lm_scale_1.1_attention_scale_2.5 17.97
1058
+ ngram_lm_scale_1.5_attention_scale_5.0 17.99
1059
+ ngram_lm_scale_1.2_attention_scale_3.0 18.0
1060
+ ngram_lm_scale_1.0_attention_scale_2.0 18.01
1061
+ ngram_lm_scale_1.0_attention_scale_1.9 18.04
1062
+ ngram_lm_scale_0.7_attention_scale_0.5 18.09
1063
+ ngram_lm_scale_1.1_attention_scale_2.3 18.12
1064
+ ngram_lm_scale_0.9_attention_scale_1.3 18.14
1065
+ ngram_lm_scale_1.1_attention_scale_2.2 18.21
1066
+ ngram_lm_scale_1.0_attention_scale_1.7 18.23
1067
+ ngram_lm_scale_0.9_attention_scale_1.2 18.25
1068
+ ngram_lm_scale_0.5_attention_scale_0.01 18.27
1069
+ ngram_lm_scale_1.1_attention_scale_2.1 18.27
1070
+ ngram_lm_scale_1.3_attention_scale_3.0 18.32
1071
+ ngram_lm_scale_1.5_attention_scale_4.0 18.35
1072
+ ngram_lm_scale_1.7_attention_scale_5.0 18.36
1073
+ ngram_lm_scale_1.1_attention_scale_2.0 18.37
1074
+ ngram_lm_scale_1.2_attention_scale_2.5 18.37
1075
+ ngram_lm_scale_0.9_attention_scale_1.1 18.38
1076
+ ngram_lm_scale_1.0_attention_scale_1.5 18.44
1077
+ ngram_lm_scale_1.1_attention_scale_1.9 18.47
1078
+ ngram_lm_scale_1.2_attention_scale_2.3 18.5
1079
+ ngram_lm_scale_0.9_attention_scale_1.0 18.57
1080
+ ngram_lm_scale_1.2_attention_scale_2.2 18.59
1081
+ ngram_lm_scale_1.1_attention_scale_1.7 18.69
1082
+ ngram_lm_scale_0.9_attention_scale_0.9 18.7
1083
+ ngram_lm_scale_1.2_attention_scale_2.1 18.7
1084
+ ngram_lm_scale_1.0_attention_scale_1.3 18.71
1085
+ ngram_lm_scale_1.3_attention_scale_2.5 18.71
1086
+ ngram_lm_scale_0.7_attention_scale_0.3 18.76
1087
+ ngram_lm_scale_0.6_attention_scale_0.1 18.78
1088
+ ngram_lm_scale_1.9_attention_scale_5.0 18.78
1089
+ ngram_lm_scale_1.2_attention_scale_2.0 18.81
1090
+ ngram_lm_scale_1.7_attention_scale_4.0 18.83
1091
+ ngram_lm_scale_1.0_attention_scale_1.2 18.88
1092
+ ngram_lm_scale_1.3_attention_scale_2.3 18.91
1093
+ ngram_lm_scale_1.2_attention_scale_1.9 18.95
1094
+ ngram_lm_scale_1.5_attention_scale_3.0 18.95
1095
+ ngram_lm_scale_0.6_attention_scale_0.08 18.96
1096
+ ngram_lm_scale_2.0_attention_scale_5.0 18.96
1097
+ ngram_lm_scale_1.1_attention_scale_1.5 18.97
1098
+ ngram_lm_scale_1.0_attention_scale_1.1 19.04
1099
+ ngram_lm_scale_1.3_attention_scale_2.2 19.08
1100
+ ngram_lm_scale_1.3_attention_scale_2.1 19.17
1101
+ ngram_lm_scale_2.1_attention_scale_5.0 19.18
1102
+ ngram_lm_scale_0.6_attention_scale_0.05 19.21
1103
+ ngram_lm_scale_0.9_attention_scale_0.7 19.27
1104
+ ngram_lm_scale_1.2_attention_scale_1.7 19.28
1105
+ ngram_lm_scale_1.3_attention_scale_2.0 19.34
1106
+ ngram_lm_scale_1.9_attention_scale_4.0 19.36
1107
+ ngram_lm_scale_1.0_attention_scale_1.0 19.4
1108
+ ngram_lm_scale_2.2_attention_scale_5.0 19.41
1109
+ ngram_lm_scale_1.1_attention_scale_1.3 19.45
1110
+ ngram_lm_scale_1.3_attention_scale_1.9 19.53
1111
+ ngram_lm_scale_1.5_attention_scale_2.5 19.54
1112
+ ngram_lm_scale_1.0_attention_scale_0.9 19.63
1113
+ ngram_lm_scale_0.6_attention_scale_0.01 19.64
1114
+ ngram_lm_scale_0.9_attention_scale_0.6 19.66
1115
+ ngram_lm_scale_1.1_attention_scale_1.2 19.67
1116
+ ngram_lm_scale_1.2_attention_scale_1.5 19.67
1117
+ ngram_lm_scale_2.3_attention_scale_5.0 19.68
1118
+ ngram_lm_scale_2.0_attention_scale_4.0 19.72
1119
+ ngram_lm_scale_1.7_attention_scale_3.0 19.77
1120
+ ngram_lm_scale_1.5_attention_scale_2.3 19.91
1121
+ ngram_lm_scale_1.1_attention_scale_1.1 19.94
1122
+ ngram_lm_scale_1.3_attention_scale_1.7 19.94
1123
+ ngram_lm_scale_2.1_attention_scale_4.0 20.03
1124
+ ngram_lm_scale_1.5_attention_scale_2.2 20.1
1125
+ ngram_lm_scale_1.2_attention_scale_1.3 20.19
1126
+ ngram_lm_scale_0.9_attention_scale_0.5 20.2
1127
+ ngram_lm_scale_0.7_attention_scale_0.1 20.26
1128
+ ngram_lm_scale_1.1_attention_scale_1.0 20.26
1129
+ ngram_lm_scale_2.5_attention_scale_5.0 20.27
1130
+ ngram_lm_scale_1.5_attention_scale_2.1 20.3
1131
+ ngram_lm_scale_2.2_attention_scale_4.0 20.38
1132
+ ngram_lm_scale_1.3_attention_scale_1.5 20.4
1133
+ ngram_lm_scale_1.0_attention_scale_0.7 20.47
1134
+ ngram_lm_scale_1.2_attention_scale_1.2 20.48
1135
+ ngram_lm_scale_1.5_attention_scale_2.0 20.51
1136
+ ngram_lm_scale_0.7_attention_scale_0.08 20.53
1137
+ ngram_lm_scale_1.7_attention_scale_2.5 20.58
1138
+ ngram_lm_scale_1.1_attention_scale_0.9 20.61
1139
+ ngram_lm_scale_1.9_attention_scale_3.0 20.67
1140
+ ngram_lm_scale_1.5_attention_scale_1.9 20.76
1141
+ ngram_lm_scale_1.2_attention_scale_1.1 20.8
1142
+ ngram_lm_scale_0.7_attention_scale_0.05 20.83
1143
+ ngram_lm_scale_2.3_attention_scale_4.0 20.84
1144
+ ngram_lm_scale_1.0_attention_scale_0.6 20.96
1145
+ ngram_lm_scale_1.7_attention_scale_2.3 21.03
1146
+ ngram_lm_scale_1.3_attention_scale_1.3 21.05
1147
+ ngram_lm_scale_2.0_attention_scale_3.0 21.15
1148
+ ngram_lm_scale_1.5_attention_scale_1.7 21.25
1149
+ ngram_lm_scale_1.2_attention_scale_1.0 21.26
1150
+ ngram_lm_scale_1.7_attention_scale_2.2 21.27
1151
+ ngram_lm_scale_0.7_attention_scale_0.01 21.36
1152
+ ngram_lm_scale_1.3_attention_scale_1.2 21.39
1153
+ ngram_lm_scale_1.7_attention_scale_2.1 21.55
1154
+ ngram_lm_scale_2.5_attention_scale_4.0 21.61
1155
+ ngram_lm_scale_1.0_attention_scale_0.5 21.62
1156
+ ngram_lm_scale_2.1_attention_scale_3.0 21.68
1157
+ ngram_lm_scale_1.1_attention_scale_0.7 21.69
1158
+ ngram_lm_scale_0.9_attention_scale_0.3 21.71
1159
+ ngram_lm_scale_1.2_attention_scale_0.9 21.75
1160
+ ngram_lm_scale_1.9_attention_scale_2.5 21.79
1161
+ ngram_lm_scale_1.7_attention_scale_2.0 21.83
1162
+ ngram_lm_scale_1.3_attention_scale_1.1 21.84
1163
+ ngram_lm_scale_3.0_attention_scale_5.0 21.94
1164
+ ngram_lm_scale_1.5_attention_scale_1.5 22.0
1165
+ ngram_lm_scale_1.7_attention_scale_1.9 22.2
1166
+ ngram_lm_scale_2.2_attention_scale_3.0 22.2
1167
+ ngram_lm_scale_1.9_attention_scale_2.3 22.28
1168
+ ngram_lm_scale_1.3_attention_scale_1.0 22.33
1169
+ ngram_lm_scale_2.0_attention_scale_2.5 22.33
1170
+ ngram_lm_scale_1.1_attention_scale_0.6 22.38
1171
+ ngram_lm_scale_1.9_attention_scale_2.2 22.6
1172
+ ngram_lm_scale_2.3_attention_scale_3.0 22.7
1173
+ ngram_lm_scale_1.3_attention_scale_0.9 22.96
1174
+ ngram_lm_scale_1.5_attention_scale_1.3 22.96
1175
+ ngram_lm_scale_2.1_attention_scale_2.5 22.98
1176
+ ngram_lm_scale_1.9_attention_scale_2.1 22.99
1177
+ ngram_lm_scale_2.0_attention_scale_2.3 22.99
1178
+ ngram_lm_scale_1.7_attention_scale_1.7 23.0
1179
+ ngram_lm_scale_1.2_attention_scale_0.7 23.04
1180
+ ngram_lm_scale_1.1_attention_scale_0.5 23.26
1181
+ ngram_lm_scale_2.0_attention_scale_2.2 23.34
1182
+ ngram_lm_scale_1.9_attention_scale_2.0 23.37
1183
+ ngram_lm_scale_1.5_attention_scale_1.2 23.5
1184
+ ngram_lm_scale_1.0_attention_scale_0.3 23.52
1185
+ ngram_lm_scale_2.2_attention_scale_2.5 23.63
1186
+ ngram_lm_scale_2.1_attention_scale_2.3 23.69
1187
+ ngram_lm_scale_2.0_attention_scale_2.1 23.74
1188
+ ngram_lm_scale_1.9_attention_scale_1.9 23.8
1189
+ ngram_lm_scale_3.0_attention_scale_4.0 23.81
1190
+ ngram_lm_scale_1.2_attention_scale_0.6 23.91
1191
+ ngram_lm_scale_1.7_attention_scale_1.5 23.91
1192
+ ngram_lm_scale_2.5_attention_scale_3.0 23.93
1193
+ ngram_lm_scale_2.1_attention_scale_2.2 24.09
1194
+ ngram_lm_scale_1.5_attention_scale_1.1 24.15
1195
+ ngram_lm_scale_2.0_attention_scale_2.0 24.15
1196
+ ngram_lm_scale_0.9_attention_scale_0.1 24.2
1197
+ ngram_lm_scale_2.3_attention_scale_2.5 24.35
1198
+ ngram_lm_scale_2.2_attention_scale_2.3 24.38
1199
+ ngram_lm_scale_2.1_attention_scale_2.1 24.5
1200
+ ngram_lm_scale_1.3_attention_scale_0.7 24.53
1201
+ ngram_lm_scale_0.9_attention_scale_0.08 24.55
1202
+ ngram_lm_scale_2.0_attention_scale_1.9 24.62
1203
+ ngram_lm_scale_1.2_attention_scale_0.5 24.77
1204
+ ngram_lm_scale_1.9_attention_scale_1.7 24.77
1205
+ ngram_lm_scale_1.5_attention_scale_1.0 24.85
1206
+ ngram_lm_scale_2.2_attention_scale_2.2 24.85
1207
+ ngram_lm_scale_2.1_attention_scale_2.0 24.98
1208
+ ngram_lm_scale_0.9_attention_scale_0.05 25.09
1209
+ ngram_lm_scale_1.7_attention_scale_1.3 25.1
1210
+ ngram_lm_scale_2.3_attention_scale_2.3 25.13
1211
+ ngram_lm_scale_2.2_attention_scale_2.1 25.26
1212
+ ngram_lm_scale_1.1_attention_scale_0.3 25.41
1213
+ ngram_lm_scale_1.3_attention_scale_0.6 25.44
1214
+ ngram_lm_scale_2.1_attention_scale_1.9 25.45
1215
+ ngram_lm_scale_1.5_attention_scale_0.9 25.59
1216
+ ngram_lm_scale_2.3_attention_scale_2.2 25.61
1217
+ ngram_lm_scale_2.0_attention_scale_1.7 25.68
1218
+ ngram_lm_scale_4.0_attention_scale_5.0 25.73
1219
+ ngram_lm_scale_1.7_attention_scale_1.2 25.82
1220
+ ngram_lm_scale_2.5_attention_scale_2.5 25.82
1221
+ ngram_lm_scale_2.2_attention_scale_2.0 25.84
1222
+ ngram_lm_scale_0.9_attention_scale_0.01 25.88
1223
+ ngram_lm_scale_1.9_attention_scale_1.5 26.0
1224
+ ngram_lm_scale_2.3_attention_scale_2.1 26.21
1225
+ ngram_lm_scale_2.2_attention_scale_1.9 26.47
1226
+ ngram_lm_scale_1.0_attention_scale_0.1 26.5
1227
+ ngram_lm_scale_1.3_attention_scale_0.5 26.6
1228
+ ngram_lm_scale_1.7_attention_scale_1.1 26.6
1229
+ ngram_lm_scale_2.1_attention_scale_1.7 26.77
1230
+ ngram_lm_scale_2.3_attention_scale_2.0 26.77
1231
+ ngram_lm_scale_2.5_attention_scale_2.3 26.85
1232
+ ngram_lm_scale_1.0_attention_scale_0.08 26.86
1233
+ ngram_lm_scale_2.0_attention_scale_1.5 27.06
1234
+ ngram_lm_scale_3.0_attention_scale_3.0 27.06
1235
+ ngram_lm_scale_1.9_attention_scale_1.3 27.31
1236
+ ngram_lm_scale_1.7_attention_scale_1.0 27.35
1237
+ ngram_lm_scale_2.5_attention_scale_2.2 27.35
1238
+ ngram_lm_scale_2.3_attention_scale_1.9 27.37
1239
+ ngram_lm_scale_1.2_attention_scale_0.3 27.4
1240
+ ngram_lm_scale_1.0_attention_scale_0.05 27.43
1241
+ ngram_lm_scale_1.5_attention_scale_0.7 27.59
1242
+ ngram_lm_scale_2.2_attention_scale_1.7 27.63
1243
+ ngram_lm_scale_2.5_attention_scale_2.1 27.83
1244
+ ngram_lm_scale_2.1_attention_scale_1.5 27.96
1245
+ ngram_lm_scale_1.9_attention_scale_1.2 28.08
1246
+ ngram_lm_scale_1.0_attention_scale_0.01 28.25
1247
+ ngram_lm_scale_1.7_attention_scale_0.9 28.3
1248
+ ngram_lm_scale_2.5_attention_scale_2.0 28.31
1249
+ ngram_lm_scale_2.0_attention_scale_1.3 28.4
1250
+ ngram_lm_scale_2.3_attention_scale_1.7 28.44
1251
+ ngram_lm_scale_4.0_attention_scale_4.0 28.62
1252
+ ngram_lm_scale_1.1_attention_scale_0.1 28.76
1253
+ ngram_lm_scale_1.5_attention_scale_0.6 28.76
1254
+ ngram_lm_scale_2.5_attention_scale_1.9 28.86
1255
+ ngram_lm_scale_1.9_attention_scale_1.1 28.89
1256
+ ngram_lm_scale_2.2_attention_scale_1.5 28.92
1257
+ ngram_lm_scale_1.1_attention_scale_0.08 29.15
1258
+ ngram_lm_scale_2.0_attention_scale_1.2 29.15
1259
+ ngram_lm_scale_3.0_attention_scale_2.5 29.22
1260
+ ngram_lm_scale_2.1_attention_scale_1.3 29.35
1261
+ ngram_lm_scale_1.3_attention_scale_0.3 29.44
1262
+ ngram_lm_scale_5.0_attention_scale_5.0 29.61
1263
+ ngram_lm_scale_2.3_attention_scale_1.5 29.69
1264
+ ngram_lm_scale_1.9_attention_scale_1.0 29.75
1265
+ ngram_lm_scale_1.1_attention_scale_0.05 29.77
1266
+ ngram_lm_scale_1.5_attention_scale_0.5 29.94
1267
+ ngram_lm_scale_2.0_attention_scale_1.1 29.96
1268
+ ngram_lm_scale_2.5_attention_scale_1.7 30.06
1269
+ ngram_lm_scale_2.1_attention_scale_1.2 30.1
1270
+ ngram_lm_scale_3.0_attention_scale_2.3 30.14
1271
+ ngram_lm_scale_2.2_attention_scale_1.3 30.29
1272
+ ngram_lm_scale_1.7_attention_scale_0.7 30.3
1273
+ ngram_lm_scale_1.1_attention_scale_0.01 30.59
1274
+ ngram_lm_scale_3.0_attention_scale_2.2 30.62
1275
+ ngram_lm_scale_1.9_attention_scale_0.9 30.64
1276
+ ngram_lm_scale_2.0_attention_scale_1.0 30.76
1277
+ ngram_lm_scale_1.2_attention_scale_0.1 30.86
1278
+ ngram_lm_scale_2.1_attention_scale_1.1 30.88
1279
+ ngram_lm_scale_2.2_attention_scale_1.2 30.98
1280
+ ngram_lm_scale_2.3_attention_scale_1.3 31.04
1281
+ ngram_lm_scale_3.0_attention_scale_2.1 31.1
1282
+ ngram_lm_scale_1.2_attention_scale_0.08 31.19
1283
+ ngram_lm_scale_2.5_attention_scale_1.5 31.23
1284
+ ngram_lm_scale_1.7_attention_scale_0.6 31.43
1285
+ ngram_lm_scale_3.0_attention_scale_2.0 31.56
1286
+ ngram_lm_scale_2.0_attention_scale_0.9 31.68
1287
+ ngram_lm_scale_2.1_attention_scale_1.0 31.71
1288
+ ngram_lm_scale_1.2_attention_scale_0.05 31.72
1289
+ ngram_lm_scale_2.2_attention_scale_1.1 31.73
1290
+ ngram_lm_scale_2.3_attention_scale_1.2 31.79
1291
+ ngram_lm_scale_4.0_attention_scale_3.0 31.95
1292
+ ngram_lm_scale_3.0_attention_scale_1.9 32.03
1293
+ ngram_lm_scale_5.0_attention_scale_4.0 32.15
1294
+ ngram_lm_scale_1.2_attention_scale_0.01 32.34
1295
+ ngram_lm_scale_2.1_attention_scale_0.9 32.53
1296
+ ngram_lm_scale_2.3_attention_scale_1.1 32.53
1297
+ ngram_lm_scale_2.5_attention_scale_1.3 32.53
1298
+ ngram_lm_scale_1.9_attention_scale_0.7 32.54
1299
+ ngram_lm_scale_1.7_attention_scale_0.5 32.55
1300
+ ngram_lm_scale_2.2_attention_scale_1.0 32.55
1301
+ ngram_lm_scale_1.3_attention_scale_0.1 32.56
1302
+ ngram_lm_scale_1.5_attention_scale_0.3 32.57
1303
+ ngram_lm_scale_1.3_attention_scale_0.08 32.82
1304
+ ngram_lm_scale_3.0_attention_scale_1.7 33.11
1305
+ ngram_lm_scale_1.3_attention_scale_0.05 33.24
1306
+ ngram_lm_scale_2.5_attention_scale_1.2 33.24
1307
+ ngram_lm_scale_2.3_attention_scale_1.0 33.28
1308
+ ngram_lm_scale_2.0_attention_scale_0.7 33.33
1309
+ ngram_lm_scale_2.2_attention_scale_0.9 33.33
1310
+ ngram_lm_scale_1.9_attention_scale_0.6 33.39
1311
+ ngram_lm_scale_4.0_attention_scale_2.5 33.71
1312
+ ngram_lm_scale_1.3_attention_scale_0.01 33.8
1313
+ ngram_lm_scale_2.5_attention_scale_1.1 33.84
1314
+ ngram_lm_scale_2.3_attention_scale_0.9 33.98
1315
+ ngram_lm_scale_3.0_attention_scale_1.5 34.1
1316
+ ngram_lm_scale_2.1_attention_scale_0.7 34.13
1317
+ ngram_lm_scale_2.0_attention_scale_0.6 34.25
1318
+ ngram_lm_scale_4.0_attention_scale_2.3 34.31
1319
+ ngram_lm_scale_1.9_attention_scale_0.5 34.37
1320
+ ngram_lm_scale_2.5_attention_scale_1.0 34.49
1321
+ ngram_lm_scale_1.7_attention_scale_0.3 34.65
1322
+ ngram_lm_scale_4.0_attention_scale_2.2 34.68
1323
+ ngram_lm_scale_5.0_attention_scale_3.0 34.68
1324
+ ngram_lm_scale_2.2_attention_scale_0.7 34.85
1325
+ ngram_lm_scale_1.5_attention_scale_0.1 34.93
1326
+ ngram_lm_scale_2.1_attention_scale_0.6 34.96
1327
+ ngram_lm_scale_4.0_attention_scale_2.1 35.05
1328
+ ngram_lm_scale_2.0_attention_scale_0.5 35.1
1329
+ ngram_lm_scale_1.5_attention_scale_0.08 35.12
1330
+ ngram_lm_scale_3.0_attention_scale_1.3 35.12
1331
+ ngram_lm_scale_2.5_attention_scale_0.9 35.15
1332
+ ngram_lm_scale_4.0_attention_scale_2.0 35.36
1333
+ ngram_lm_scale_1.5_attention_scale_0.05 35.38
1334
+ ngram_lm_scale_2.3_attention_scale_0.7 35.4
1335
+ ngram_lm_scale_3.0_attention_scale_1.2 35.5
1336
+ ngram_lm_scale_2.2_attention_scale_0.6 35.56
1337
+ ngram_lm_scale_4.0_attention_scale_1.9 35.66
1338
+ ngram_lm_scale_2.1_attention_scale_0.5 35.7
1339
+ ngram_lm_scale_1.5_attention_scale_0.01 35.81
1340
+ ngram_lm_scale_3.0_attention_scale_1.1 35.88
1341
+ ngram_lm_scale_2.3_attention_scale_0.6 35.94
1342
+ ngram_lm_scale_5.0_attention_scale_2.5 35.98
1343
+ ngram_lm_scale_1.9_attention_scale_0.3 35.99
1344
+ ngram_lm_scale_2.2_attention_scale_0.5 36.09
1345
+ ngram_lm_scale_2.5_attention_scale_0.7 36.09
1346
+ ngram_lm_scale_4.0_attention_scale_1.7 36.2
1347
+ ngram_lm_scale_3.0_attention_scale_1.0 36.26
1348
+ ngram_lm_scale_1.7_attention_scale_0.1 36.33
1349
+ ngram_lm_scale_5.0_attention_scale_2.3 36.37
1350
+ ngram_lm_scale_2.3_attention_scale_0.5 36.45
1351
+ ngram_lm_scale_2.0_attention_scale_0.3 36.46
1352
+ ngram_lm_scale_1.7_attention_scale_0.08 36.48
1353
+ ngram_lm_scale_5.0_attention_scale_2.2 36.55
1354
+ ngram_lm_scale_2.5_attention_scale_0.6 36.59
1355
+ ngram_lm_scale_1.7_attention_scale_0.05 36.66
1356
+ ngram_lm_scale_3.0_attention_scale_0.9 36.66
1357
+ ngram_lm_scale_4.0_attention_scale_1.5 36.74
1358
+ ngram_lm_scale_5.0_attention_scale_2.1 36.77
1359
+ ngram_lm_scale_2.1_attention_scale_0.3 36.82
1360
+ ngram_lm_scale_1.7_attention_scale_0.01 36.93
1361
+ ngram_lm_scale_5.0_attention_scale_2.0 36.99
1362
+ ngram_lm_scale_2.5_attention_scale_0.5 37.06
1363
+ ngram_lm_scale_2.2_attention_scale_0.3 37.11
1364
+ ngram_lm_scale_5.0_attention_scale_1.9 37.22
1365
+ ngram_lm_scale_1.9_attention_scale_0.1 37.25
1366
+ ngram_lm_scale_4.0_attention_scale_1.3 37.3
1367
+ ngram_lm_scale_1.9_attention_scale_0.08 37.35
1368
+ ngram_lm_scale_2.3_attention_scale_0.3 37.42
1369
+ ngram_lm_scale_3.0_attention_scale_0.7 37.44
1370
+ ngram_lm_scale_1.9_attention_scale_0.05 37.55
1371
+ ngram_lm_scale_2.0_attention_scale_0.1 37.58
1372
+ ngram_lm_scale_4.0_attention_scale_1.2 37.6
1373
+ ngram_lm_scale_5.0_attention_scale_1.7 37.65
1374
+ ngram_lm_scale_2.0_attention_scale_0.08 37.7
1375
+ ngram_lm_scale_1.9_attention_scale_0.01 37.78
1376
+ ngram_lm_scale_3.0_attention_scale_0.6 37.8
1377
+ ngram_lm_scale_4.0_attention_scale_1.1 37.84
1378
+ ngram_lm_scale_2.1_attention_scale_0.1 37.86
1379
+ ngram_lm_scale_2.0_attention_scale_0.05 37.87
1380
+ ngram_lm_scale_2.5_attention_scale_0.3 37.92
1381
+ ngram_lm_scale_2.1_attention_scale_0.08 37.96
1382
+ ngram_lm_scale_5.0_attention_scale_1.5 38.04
1383
+ ngram_lm_scale_2.0_attention_scale_0.01 38.07
1384
+ ngram_lm_scale_4.0_attention_scale_1.0 38.07
1385
+ ngram_lm_scale_2.2_attention_scale_0.1 38.1
1386
+ ngram_lm_scale_2.1_attention_scale_0.05 38.11
1387
+ ngram_lm_scale_3.0_attention_scale_0.5 38.14
1388
+ ngram_lm_scale_2.2_attention_scale_0.08 38.23
1389
+ ngram_lm_scale_2.1_attention_scale_0.01 38.29
1390
+ ngram_lm_scale_4.0_attention_scale_0.9 38.33
1391
+ ngram_lm_scale_2.2_attention_scale_0.05 38.36
1392
+ ngram_lm_scale_2.3_attention_scale_0.1 38.37
1393
+ ngram_lm_scale_5.0_attention_scale_1.3 38.42
1394
+ ngram_lm_scale_2.3_attention_scale_0.08 38.44
1395
+ ngram_lm_scale_2.2_attention_scale_0.01 38.52
1396
+ ngram_lm_scale_2.3_attention_scale_0.05 38.58
1397
+ ngram_lm_scale_5.0_attention_scale_1.2 38.58
1398
+ ngram_lm_scale_2.5_attention_scale_0.1 38.71
1399
+ ngram_lm_scale_2.3_attention_scale_0.01 38.73
1400
+ ngram_lm_scale_3.0_attention_scale_0.3 38.75
1401
+ ngram_lm_scale_5.0_attention_scale_1.1 38.77
1402
+ ngram_lm_scale_2.5_attention_scale_0.08 38.8
1403
+ ngram_lm_scale_4.0_attention_scale_0.7 38.8
1404
+ ngram_lm_scale_2.5_attention_scale_0.05 38.87
1405
+ ngram_lm_scale_5.0_attention_scale_1.0 38.96
1406
+ ngram_lm_scale_4.0_attention_scale_0.6 38.99
1407
+ ngram_lm_scale_2.5_attention_scale_0.01 39.05
1408
+ ngram_lm_scale_5.0_attention_scale_0.9 39.14
1409
+ ngram_lm_scale_4.0_attention_scale_0.5 39.18
1410
+ ngram_lm_scale_3.0_attention_scale_0.1 39.3
1411
+ ngram_lm_scale_3.0_attention_scale_0.08 39.37
1412
+ ngram_lm_scale_3.0_attention_scale_0.05 39.45
1413
+ ngram_lm_scale_5.0_attention_scale_0.7 39.48
1414
+ ngram_lm_scale_3.0_attention_scale_0.01 39.55
1415
+ ngram_lm_scale_4.0_attention_scale_0.3 39.6
1416
+ ngram_lm_scale_5.0_attention_scale_0.6 39.65
1417
+ ngram_lm_scale_5.0_attention_scale_0.5 39.77
1418
+ ngram_lm_scale_4.0_attention_scale_0.1 39.96
1419
+ ngram_lm_scale_4.0_attention_scale_0.08 40.0
1420
+ ngram_lm_scale_5.0_attention_scale_0.3 40.03
1421
+ ngram_lm_scale_4.0_attention_scale_0.05 40.07
1422
+ ngram_lm_scale_4.0_attention_scale_0.01 40.12
1423
+ ngram_lm_scale_5.0_attention_scale_0.1 40.32
1424
+ ngram_lm_scale_5.0_attention_scale_0.08 40.34
1425
+ ngram_lm_scale_5.0_attention_scale_0.05 40.38
1426
+ ngram_lm_scale_5.0_attention_scale_0.01 40.45
1427
+
1428
+ 2022-06-26 15:43:53,296 INFO [decode.py:695] Done!
decoding-results/log-attention-decoder/log-decode-2022-06-27-18-54-02 ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ 2022-06-27 18:54:02,244 INFO [decode.py:548] Decoding started
2
+ 2022-06-27 18:54:02,245 INFO [decode.py:549] {'subsampling_factor': 4, 'vgg_frontend': False, 'use_feat_batchnorm': True, 'feature_dim': 80, 'nhead': 8, 'attention_dim': 512, 'num_decoder_layers': 6, 'search_beam': 20, 'output_beam': 8, 'min_active_states': 30, 'max_active_states': 5000, 'use_double_scores': True, 'env_info': {'k2-version': '1.16', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '3c606c27045750bbbb7a289d8b2b09825dea521a', 'k2-git-date': 'Mon Jun 27 03:06:58 2022', 'lhotse-version': '1.3.0.dev+git.a07121a.clean', 'torch-version': '1.7.1', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'test', 'icefall-git-sha1': 'e24e6ac-dirty', 'icefall-git-date': 'Mon Jun 27 01:23:06 2022', 'icefall-path': '/alt-arabic/speech/amir/k2/tmp/icefall', 'k2-path': '/alt-arabic/speech/amir/k2/tmp/k2/k2/python/k2/__init__.py', 'lhotse-path': '/alt-arabic/speech/amir/k2/tmp/lhotse/lhotse/__init__.py', 'hostname': 'crimv3srv031', 'IP address': '10.141.0.13'}, 'epoch': 45, 'avg': 10, 'method': 'attention-decoder', 'num_paths': 1000, 'nbest_scale': 0.5, 'exp_dir': PosixPath('conformer_ctc/exp_5000_att0.8'), 'lang_dir': PosixPath('data/lang_bpe_5000'), 'lm_dir': PosixPath('data/lm'), 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 30, 'bucketing_sampler': True, 'num_buckets': 30, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'on_the_fly_feats': False, 'shuffle': False, 'drop_last': True, 'return_cuts': True, 'num_workers': 20, 'enable_spec_aug': False, 'spec_aug_time_warp_factor': 80, 'enable_musan': False}
3
+ 2022-06-27 18:54:02,500 INFO [lexicon.py:177] Loading pre-compiled data/lang_bpe_5000/Linv.pt
4
+ 2022-06-27 18:54:02,544 INFO [decode.py:559] device: cuda:0
5
+ 2022-06-27 18:54:33,545 INFO [decode.py:621] Loading pre-compiled G_4_gram.pt
6
+ 2022-06-27 18:54:34,545 INFO [decode.py:657] averaging ['conformer_ctc/exp_5000_att0.8/epoch-36.pt', 'conformer_ctc/exp_5000_att0.8/epoch-37.pt', 'conformer_ctc/exp_5000_att0.8/epoch-38.pt', 'conformer_ctc/exp_5000_att0.8/epoch-39.pt', 'conformer_ctc/exp_5000_att0.8/epoch-40.pt', 'conformer_ctc/exp_5000_att0.8/epoch-41.pt', 'conformer_ctc/exp_5000_att0.8/epoch-42.pt', 'conformer_ctc/exp_5000_att0.8/epoch-43.pt', 'conformer_ctc/exp_5000_att0.8/epoch-44.pt', 'conformer_ctc/exp_5000_att0.8/epoch-45.pt']
decoding-results/log-attention-decoder/log-decode-2022-06-27-19-04-48 ADDED
@@ -0,0 +1,1308 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2022-06-27 19:04:48,400 INFO [decode.py:548] Decoding started
2
+ 2022-06-27 19:04:48,401 INFO [decode.py:549] {'subsampling_factor': 4, 'vgg_frontend': False, 'use_feat_batchnorm': True, 'feature_dim': 80, 'nhead': 8, 'attention_dim': 512, 'num_decoder_layers': 6, 'search_beam': 20, 'output_beam': 8, 'min_active_states': 30, 'max_active_states': 5000, 'use_double_scores': True, 'env_info': {'k2-version': '1.16', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '3c606c27045750bbbb7a289d8b2b09825dea521a', 'k2-git-date': 'Mon Jun 27 03:06:58 2022', 'lhotse-version': '1.3.0.dev+git.a07121a.clean', 'torch-version': '1.7.1', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'test', 'icefall-git-sha1': 'e24e6ac-dirty', 'icefall-git-date': 'Mon Jun 27 01:23:06 2022', 'icefall-path': '/alt-arabic/speech/amir/k2/tmp/icefall', 'k2-path': '/alt-arabic/speech/amir/k2/tmp/k2/k2/python/k2/__init__.py', 'lhotse-path': '/alt-arabic/speech/amir/k2/tmp/lhotse/lhotse/__init__.py', 'hostname': 'crimv3mgpu016', 'IP address': '10.141.0.3'}, 'epoch': 45, 'avg': 5, 'method': 'attention-decoder', 'num_paths': 1000, 'nbest_scale': 0.5, 'exp_dir': PosixPath('conformer_ctc/exp_5000_att0.8'), 'lang_dir': PosixPath('data/lang_bpe_5000'), 'lm_dir': PosixPath('data/lm'), 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 30, 'bucketing_sampler': True, 'num_buckets': 30, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'on_the_fly_feats': False, 'shuffle': False, 'drop_last': True, 'return_cuts': True, 'num_workers': 20, 'enable_spec_aug': False, 'spec_aug_time_warp_factor': 80, 'enable_musan': False}
3
+ 2022-06-27 19:04:48,655 INFO [lexicon.py:177] Loading pre-compiled data/lang_bpe_5000/Linv.pt
4
+ 2022-06-27 19:04:48,686 INFO [decode.py:559] device: cuda:0
5
+ 2022-06-27 19:05:17,818 INFO [decode.py:621] Loading pre-compiled G_4_gram.pt
6
+ 2022-06-27 19:05:18,546 INFO [decode.py:657] averaging ['conformer_ctc/exp_5000_att0.8/epoch-41.pt', 'conformer_ctc/exp_5000_att0.8/epoch-42.pt', 'conformer_ctc/exp_5000_att0.8/epoch-43.pt', 'conformer_ctc/exp_5000_att0.8/epoch-44.pt', 'conformer_ctc/exp_5000_att0.8/epoch-45.pt']
7
+ 2022-06-27 19:05:21,863 INFO [decode.py:664] Number of model parameters: 90786736
8
+ 2022-06-27 19:05:21,864 INFO [asr_datamodule.py:362] About to get test cuts
9
+ 2022-06-27 19:05:21,867 INFO [asr_datamodule.py:357] About to get dev cuts
10
+ 2022-06-27 19:05:24,421 INFO [decode.py:483] batch 0/?, cuts processed until now is 4
11
+ 2022-06-27 19:07:22,224 INFO [decode.py:783] Caught exception:
12
+ CUDA out of memory. Tried to allocate 1.70 GiB (GPU 0; 31.75 GiB total capacity; 27.32 GiB already allocated; 470.50 MiB free; 30.09 GiB reserved in total by PyTorch)
13
+ Exception raised from malloc at /pytorch/c10/cuda/CUDACachingAllocator.cpp:272 (most recent call first):
14
+ frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x42 (0x2aab0258d8b2 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10.so)
15
+ frame #1: <unknown function> + 0x2021b (0x2aab0232721b in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
16
+ frame #2: <unknown function> + 0x21034 (0x2aab02328034 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
17
+ frame #3: <unknown function> + 0x2167d (0x2aab0232867d in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
18
+ frame #4: k2::PytorchCudaContext::Allocate(unsigned long, void**) + 0x3a (0x2aab1173401a in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
19
+ frame #5: k2::NewRegion(std::shared_ptr<k2::Context>, unsigned long) + 0x112 (0x2aab11465b72 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
20
+ frame #6: k2::Array1<int>::Init(std::shared_ptr<k2::Context>, int, k2::Dtype) + 0x71 (0x2aab11432f51 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
21
+ frame #7: <unknown function> + 0x2472bd (0x2aab115c32bd in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
22
+ frame #8: k2::RaggedShapeFromTotSizes(std::shared_ptr<k2::Context>, int, int const*) + 0x213 (0x2aab115c3b83 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
23
+ frame #9: k2::IndexAxis0(k2::RaggedShape&, k2::Array1<int> const&, k2::Array1<int>*) + 0x32c (0x2aab115d77ec in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
24
+ frame #10: k2::Index(k2::RaggedShape&, int, k2::Array1<int> const&, k2::Array1<int>*) + 0x353 (0x2aab115dc943 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
25
+ frame #11: k2::Ragged<k2::Arc> k2::DeviceIntersector::FormatOutputTpl<k2::Hash::PackedAccessor>(k2::Array1<int>*, k2::Array1<int>*) + 0x407 (0x2aab11552327 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
26
+ frame #12: k2::IntersectDevice(k2::Ragged<k2::Arc>&, int, k2::Ragged<k2::Arc>&, int, k2::Array1<int> const&, k2::Array1<int>*, k2::Array1<int>*, bool) + 0x3a2 (0x2aab11545682 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
27
+ frame #13: <unknown function> + 0x8eb5a (0x2aab1032cb5a in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/_k2.cpython-38-x86_64-linux-gnu.so)
28
+ frame #14: <unknown function> + 0x3628c (0x2aab102d428c in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/_k2.cpython-38-x86_64-linux-gnu.so)
29
+ <omitting python frames>
30
+ frame #44: __libc_start_main + 0xf5 (0x2aaaab616555 in /lib64/libc.so.6)
31
+
32
+
33
+ 2022-06-27 19:07:22,225 INFO [decode.py:789] num_arcs before pruning: 940457
34
+ 2022-06-27 19:07:22,225 INFO [decode.py:792] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
35
+ 2022-06-27 19:07:22,238 INFO [decode.py:803] num_arcs after pruning: 8198
36
+ 2022-06-27 19:08:00,940 INFO [decode.py:483] batch 100/?, cuts processed until now is 407
37
+ 2022-06-27 19:10:32,748 INFO [decode.py:483] batch 200/?, cuts processed until now is 839
38
+ 2022-06-27 19:12:53,794 INFO [decode.py:483] batch 300/?, cuts processed until now is 1272
39
+ 2022-06-27 19:15:03,281 INFO [decode.py:483] batch 400/?, cuts processed until now is 1702
40
+ 2022-06-27 19:17:04,528 INFO [decode.py:483] batch 500/?, cuts processed until now is 2109
41
+ 2022-06-27 19:19:19,810 INFO [decode.py:483] batch 600/?, cuts processed until now is 2544
42
+ 2022-06-27 19:21:50,925 INFO [decode.py:483] batch 700/?, cuts processed until now is 2978
43
+ 2022-06-27 19:24:17,295 INFO [decode.py:483] batch 800/?, cuts processed until now is 3384
44
+ 2022-06-27 19:26:40,070 INFO [decode.py:483] batch 900/?, cuts processed until now is 3811
45
+ 2022-06-27 19:28:46,514 INFO [decode.py:783] Caught exception:
46
+ CUDA out of memory. Tried to allocate 1.67 GiB (GPU 0; 31.75 GiB total capacity; 27.27 GiB already allocated; 990.50 MiB free; 29.58 GiB reserved in total by PyTorch)
47
+ Exception raised from malloc at /pytorch/c10/cuda/CUDACachingAllocator.cpp:272 (most recent call first):
48
+ frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x42 (0x2aab0258d8b2 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10.so)
49
+ frame #1: <unknown function> + 0x2021b (0x2aab0232721b in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
50
+ frame #2: <unknown function> + 0x21034 (0x2aab02328034 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
51
+ frame #3: <unknown function> + 0x2167d (0x2aab0232867d in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
52
+ frame #4: k2::PytorchCudaContext::Allocate(unsigned long, void**) + 0x3a (0x2aab1173401a in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
53
+ frame #5: k2::NewRegion(std::shared_ptr<k2::Context>, unsigned long) + 0x112 (0x2aab11465b72 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
54
+ frame #6: k2::Array1<int>::Init(std::shared_ptr<k2::Context>, int, k2::Dtype) + 0x71 (0x2aab11432f51 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
55
+ frame #7: <unknown function> + 0x2472bd (0x2aab115c32bd in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
56
+ frame #8: k2::RaggedShapeFromTotSizes(std::shared_ptr<k2::Context>, int, int const*) + 0x213 (0x2aab115c3b83 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
57
+ frame #9: k2::IndexAxis0(k2::RaggedShape&, k2::Array1<int> const&, k2::Array1<int>*) + 0x32c (0x2aab115d77ec in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
58
+ frame #10: k2::Index(k2::RaggedShape&, int, k2::Array1<int> const&, k2::Array1<int>*) + 0x353 (0x2aab115dc943 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
59
+ frame #11: k2::Ragged<k2::Arc> k2::DeviceIntersector::FormatOutputTpl<k2::Hash::PackedAccessor>(k2::Array1<int>*, k2::Array1<int>*) + 0x407 (0x2aab11552327 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
60
+ frame #12: k2::IntersectDevice(k2::Ragged<k2::Arc>&, int, k2::Ragged<k2::Arc>&, int, k2::Array1<int> const&, k2::Array1<int>*, k2::Array1<int>*, bool) + 0x3a2 (0x2aab11545682 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
61
+ frame #13: <unknown function> + 0x8eb5a (0x2aab1032cb5a in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/_k2.cpython-38-x86_64-linux-gnu.so)
62
+ frame #14: <unknown function> + 0x3628c (0x2aab102d428c in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/_k2.cpython-38-x86_64-linux-gnu.so)
63
+ <omitting python frames>
64
+ frame #44: __libc_start_main + 0xf5 (0x2aaaab616555 in /lib64/libc.so.6)
65
+
66
+
67
+ 2022-06-27 19:28:46,515 INFO [decode.py:789] num_arcs before pruning: 1034414
68
+ 2022-06-27 19:28:46,515 INFO [decode.py:792] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
69
+ 2022-06-27 19:28:46,527 INFO [decode.py:803] num_arcs after pruning: 5251
70
+ 2022-06-27 19:29:03,366 INFO [decode.py:483] batch 1000/?, cuts processed until now is 4220
71
+ 2022-06-27 19:31:12,021 INFO [decode.py:483] batch 1100/?, cuts processed until now is 4631
72
+ 2022-06-27 19:33:19,271 INFO [decode.py:483] batch 1200/?, cuts processed until now is 5033
73
+ 2022-06-27 19:35:04,562 INFO [decode.py:783] Caught exception:
74
+ CUDA out of memory. Tried to allocate 1.23 GiB (GPU 0; 31.75 GiB total capacity; 26.53 GiB already allocated; 1010.50 MiB free; 29.56 GiB reserved in total by PyTorch)
75
+ Exception raised from malloc at /pytorch/c10/cuda/CUDACachingAllocator.cpp:272 (most recent call first):
76
+ frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x42 (0x2aab0258d8b2 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10.so)
77
+ frame #1: <unknown function> + 0x2021b (0x2aab0232721b in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
78
+ frame #2: <unknown function> + 0x21034 (0x2aab02328034 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
79
+ frame #3: <unknown function> + 0x2167d (0x2aab0232867d in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
80
+ frame #4: k2::PytorchCudaContext::Allocate(unsigned long, void**) + 0x3a (0x2aab1173401a in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
81
+ frame #5: k2::NewRegion(std::shared_ptr<k2::Context>, unsigned long) + 0x112 (0x2aab11465b72 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
82
+ frame #6: k2::Array1<int>::Init(std::shared_ptr<k2::Context>, int, k2::Dtype) + 0x71 (0x2aab11432f51 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
83
+ frame #7: <unknown function> + 0x2472bd (0x2aab115c32bd in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
84
+ frame #8: k2::IndexAxis0(k2::RaggedShape&, k2::Array1<int> const&, k2::Array1<int>*) + 0x2d9 (0x2aab115d7799 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
85
+ frame #9: k2::Index(k2::RaggedShape&, int, k2::Array1<int> const&, k2::Array1<int>*) + 0x353 (0x2aab115dc943 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
86
+ frame #10: k2::Ragged<k2::Arc> k2::DeviceIntersector::FormatOutputTpl<k2::Hash::PackedAccessor>(k2::Array1<int>*, k2::Array1<int>*) + 0x407 (0x2aab11552327 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
87
+ frame #11: k2::IntersectDevice(k2::Ragged<k2::Arc>&, int, k2::Ragged<k2::Arc>&, int, k2::Array1<int> const&, k2::Array1<int>*, k2::Array1<int>*, bool) + 0x3a2 (0x2aab11545682 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
88
+ frame #12: <unknown function> + 0x8eb5a (0x2aab1032cb5a in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/_k2.cpython-38-x86_64-linux-gnu.so)
89
+ frame #13: <unknown function> + 0x3628c (0x2aab102d428c in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/_k2.cpython-38-x86_64-linux-gnu.so)
90
+ <omitting python frames>
91
+ frame #43: __libc_start_main + 0xf5 (0x2aaaab616555 in /lib64/libc.so.6)
92
+
93
+
94
+ 2022-06-27 19:35:04,563 INFO [decode.py:789] num_arcs before pruning: 1081951
95
+ 2022-06-27 19:35:04,563 INFO [decode.py:792] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
96
+ 2022-06-27 19:35:04,576 INFO [decode.py:803] num_arcs after pruning: 6154
97
+ 2022-06-27 19:35:29,511 INFO [decode.py:483] batch 1300/?, cuts processed until now is 5355
98
+ 2022-06-27 19:39:19,317 INFO [decode.py:532]
99
+ For test, WER of different settings are:
100
+ ngram_lm_scale_0.01_attention_scale_0.3 15.08 best for test
101
+ ngram_lm_scale_0.01_attention_scale_0.5 15.1
102
+ ngram_lm_scale_0.01_attention_scale_0.6 15.1
103
+ ngram_lm_scale_0.01_attention_scale_0.7 15.11
104
+ ngram_lm_scale_0.05_attention_scale_0.3 15.11
105
+ ngram_lm_scale_0.05_attention_scale_0.5 15.11
106
+ ngram_lm_scale_0.01_attention_scale_0.9 15.13
107
+ ngram_lm_scale_0.05_attention_scale_0.6 15.13
108
+ ngram_lm_scale_0.05_attention_scale_0.7 15.13
109
+ ngram_lm_scale_0.08_attention_scale_0.5 15.13
110
+ ngram_lm_scale_0.08_attention_scale_0.3 15.14
111
+ ngram_lm_scale_0.08_attention_scale_0.6 15.14
112
+ ngram_lm_scale_0.1_attention_scale_0.3 15.14
113
+ ngram_lm_scale_0.01_attention_scale_1.0 15.15
114
+ ngram_lm_scale_0.08_attention_scale_0.7 15.15
115
+ ngram_lm_scale_0.1_attention_scale_0.6 15.16
116
+ ngram_lm_scale_0.1_attention_scale_0.5 15.17
117
+ ngram_lm_scale_0.1_attention_scale_0.7 15.17
118
+ ngram_lm_scale_0.01_attention_scale_1.1 15.18
119
+ ngram_lm_scale_0.05_attention_scale_0.9 15.18
120
+ ngram_lm_scale_0.01_attention_scale_0.1 15.19
121
+ ngram_lm_scale_0.05_attention_scale_1.0 15.19
122
+ ngram_lm_scale_0.08_attention_scale_0.9 15.19
123
+ ngram_lm_scale_0.01_attention_scale_1.2 15.2
124
+ ngram_lm_scale_0.05_attention_scale_0.1 15.2
125
+ ngram_lm_scale_0.1_attention_scale_0.9 15.2
126
+ ngram_lm_scale_0.05_attention_scale_1.1 15.21
127
+ ngram_lm_scale_0.01_attention_scale_0.08 15.22
128
+ ngram_lm_scale_0.01_attention_scale_1.3 15.22
129
+ ngram_lm_scale_0.08_attention_scale_1.0 15.22
130
+ ngram_lm_scale_0.01_attention_scale_0.05 15.23
131
+ ngram_lm_scale_0.05_attention_scale_0.08 15.23
132
+ ngram_lm_scale_0.05_attention_scale_1.2 15.23
133
+ ngram_lm_scale_0.08_attention_scale_0.1 15.23
134
+ ngram_lm_scale_0.08_attention_scale_1.1 15.23
135
+ ngram_lm_scale_0.1_attention_scale_1.0 15.23
136
+ ngram_lm_scale_0.05_attention_scale_0.05 15.24
137
+ ngram_lm_scale_0.1_attention_scale_0.1 15.24
138
+ ngram_lm_scale_0.1_attention_scale_1.1 15.24
139
+ ngram_lm_scale_0.05_attention_scale_1.3 15.25
140
+ ngram_lm_scale_0.08_attention_scale_0.08 15.25
141
+ ngram_lm_scale_0.08_attention_scale_1.2 15.25
142
+ ngram_lm_scale_0.01_attention_scale_1.5 15.26
143
+ ngram_lm_scale_0.08_attention_scale_1.3 15.26
144
+ ngram_lm_scale_0.05_attention_scale_1.5 15.27
145
+ ngram_lm_scale_0.1_attention_scale_0.08 15.27
146
+ ngram_lm_scale_0.01_attention_scale_0.01 15.28
147
+ ngram_lm_scale_0.08_attention_scale_0.05 15.28
148
+ ngram_lm_scale_0.1_attention_scale_1.2 15.28
149
+ ngram_lm_scale_0.1_attention_scale_1.3 15.28
150
+ ngram_lm_scale_0.01_attention_scale_1.7 15.29
151
+ ngram_lm_scale_0.08_attention_scale_1.5 15.3
152
+ ngram_lm_scale_0.1_attention_scale_1.5 15.3
153
+ ngram_lm_scale_0.05_attention_scale_0.01 15.31
154
+ ngram_lm_scale_0.05_attention_scale_1.7 15.31
155
+ ngram_lm_scale_0.1_attention_scale_0.05 15.32
156
+ ngram_lm_scale_0.08_attention_scale_0.01 15.33
157
+ ngram_lm_scale_0.08_attention_scale_1.7 15.33
158
+ ngram_lm_scale_0.01_attention_scale_1.9 15.34
159
+ ngram_lm_scale_0.01_attention_scale_2.0 15.34
160
+ ngram_lm_scale_0.1_attention_scale_0.01 15.35
161
+ ngram_lm_scale_0.01_attention_scale_2.1 15.37
162
+ ngram_lm_scale_0.05_attention_scale_1.9 15.37
163
+ ngram_lm_scale_0.3_attention_scale_0.5 15.37
164
+ ngram_lm_scale_0.3_attention_scale_0.6 15.37
165
+ ngram_lm_scale_0.05_attention_scale_2.0 15.38
166
+ ngram_lm_scale_0.1_attention_scale_1.7 15.38
167
+ ngram_lm_scale_0.3_attention_scale_0.7 15.38
168
+ ngram_lm_scale_0.01_attention_scale_2.2 15.39
169
+ ngram_lm_scale_0.08_attention_scale_1.9 15.39
170
+ ngram_lm_scale_0.3_attention_scale_0.9 15.39
171
+ ngram_lm_scale_0.01_attention_scale_2.3 15.4
172
+ ngram_lm_scale_0.01_attention_scale_2.5 15.4
173
+ ngram_lm_scale_0.05_attention_scale_2.1 15.4
174
+ ngram_lm_scale_0.05_attention_scale_2.2 15.41
175
+ ngram_lm_scale_0.05_attention_scale_2.3 15.41
176
+ ngram_lm_scale_0.08_attention_scale_2.0 15.41
177
+ ngram_lm_scale_0.1_attention_scale_1.9 15.41
178
+ ngram_lm_scale_0.05_attention_scale_2.5 15.42
179
+ ngram_lm_scale_0.08_attention_scale_2.1 15.42
180
+ ngram_lm_scale_0.08_attention_scale_2.2 15.42
181
+ ngram_lm_scale_0.08_attention_scale_2.3 15.43
182
+ ngram_lm_scale_0.08_attention_scale_2.5 15.43
183
+ ngram_lm_scale_0.1_attention_scale_2.0 15.43
184
+ ngram_lm_scale_0.1_attention_scale_2.1 15.43
185
+ ngram_lm_scale_0.1_attention_scale_2.2 15.43
186
+ ngram_lm_scale_0.1_attention_scale_2.3 15.43
187
+ ngram_lm_scale_0.3_attention_scale_1.0 15.43
188
+ ngram_lm_scale_0.1_attention_scale_2.5 15.44
189
+ ngram_lm_scale_0.01_attention_scale_3.0 15.45
190
+ ngram_lm_scale_0.3_attention_scale_0.3 15.45
191
+ ngram_lm_scale_0.3_attention_scale_1.1 15.45
192
+ ngram_lm_scale_0.3_attention_scale_1.2 15.45
193
+ ngram_lm_scale_0.3_attention_scale_1.3 15.45
194
+ ngram_lm_scale_0.05_attention_scale_3.0 15.48
195
+ ngram_lm_scale_0.3_attention_scale_1.5 15.48
196
+ ngram_lm_scale_0.08_attention_scale_3.0 15.49
197
+ ngram_lm_scale_0.1_attention_scale_3.0 15.5
198
+ ngram_lm_scale_0.3_attention_scale_1.7 15.52
199
+ ngram_lm_scale_0.3_attention_scale_1.9 15.54
200
+ ngram_lm_scale_0.01_attention_scale_4.0 15.56
201
+ ngram_lm_scale_0.3_attention_scale_2.0 15.56
202
+ ngram_lm_scale_0.3_attention_scale_2.1 15.57
203
+ ngram_lm_scale_0.3_attention_scale_2.2 15.57
204
+ ngram_lm_scale_0.05_attention_scale_4.0 15.58
205
+ ngram_lm_scale_0.3_attention_scale_2.3 15.58
206
+ ngram_lm_scale_0.08_attention_scale_4.0 15.59
207
+ ngram_lm_scale_0.1_attention_scale_4.0 15.6
208
+ ngram_lm_scale_0.3_attention_scale_2.5 15.61
209
+ ngram_lm_scale_0.01_attention_scale_5.0 15.64
210
+ ngram_lm_scale_0.3_attention_scale_0.1 15.64
211
+ ngram_lm_scale_0.3_attention_scale_3.0 15.66
212
+ ngram_lm_scale_0.05_attention_scale_5.0 15.67
213
+ ngram_lm_scale_0.08_attention_scale_5.0 15.69
214
+ ngram_lm_scale_0.3_attention_scale_0.08 15.69
215
+ ngram_lm_scale_0.1_attention_scale_5.0 15.7
216
+ ngram_lm_scale_0.3_attention_scale_4.0 15.75
217
+ ngram_lm_scale_0.3_attention_scale_0.05 15.78
218
+ ngram_lm_scale_0.3_attention_scale_5.0 15.81
219
+ ngram_lm_scale_0.5_attention_scale_1.5 15.81
220
+ ngram_lm_scale_0.5_attention_scale_2.0 15.81
221
+ ngram_lm_scale_0.5_attention_scale_2.1 15.81
222
+ ngram_lm_scale_0.5_attention_scale_1.3 15.82
223
+ ngram_lm_scale_0.5_attention_scale_1.7 15.82
224
+ ngram_lm_scale_0.5_attention_scale_1.9 15.82
225
+ ngram_lm_scale_0.5_attention_scale_2.2 15.82
226
+ ngram_lm_scale_0.5_attention_scale_2.3 15.82
227
+ ngram_lm_scale_0.5_attention_scale_0.9 15.84
228
+ ngram_lm_scale_0.5_attention_scale_1.2 15.84
229
+ ngram_lm_scale_0.5_attention_scale_1.0 15.85
230
+ ngram_lm_scale_0.5_attention_scale_1.1 15.85
231
+ ngram_lm_scale_0.5_attention_scale_2.5 15.85
232
+ ngram_lm_scale_0.5_attention_scale_3.0 15.88
233
+ ngram_lm_scale_0.5_attention_scale_4.0 15.89
234
+ ngram_lm_scale_0.5_attention_scale_0.7 15.9
235
+ ngram_lm_scale_0.5_attention_scale_0.6 15.91
236
+ ngram_lm_scale_0.3_attention_scale_0.01 15.94
237
+ ngram_lm_scale_0.5_attention_scale_0.5 15.95
238
+ ngram_lm_scale_0.5_attention_scale_5.0 15.95
239
+ ngram_lm_scale_0.6_attention_scale_2.3 15.97
240
+ ngram_lm_scale_0.6_attention_scale_2.2 15.98
241
+ ngram_lm_scale_0.6_attention_scale_1.9 15.99
242
+ ngram_lm_scale_0.6_attention_scale_2.0 15.99
243
+ ngram_lm_scale_0.6_attention_scale_2.1 15.99
244
+ ngram_lm_scale_0.6_attention_scale_2.5 15.99
245
+ ngram_lm_scale_0.6_attention_scale_3.0 15.99
246
+ ngram_lm_scale_0.6_attention_scale_1.7 16.0
247
+ ngram_lm_scale_0.6_attention_scale_1.5 16.01
248
+ ngram_lm_scale_0.6_attention_scale_4.0 16.01
249
+ ngram_lm_scale_0.6_attention_scale_1.3 16.04
250
+ ngram_lm_scale_0.6_attention_scale_5.0 16.05
251
+ ngram_lm_scale_0.6_attention_scale_1.1 16.07
252
+ ngram_lm_scale_0.6_attention_scale_1.2 16.07
253
+ ngram_lm_scale_0.6_attention_scale_1.0 16.11
254
+ ngram_lm_scale_0.6_attention_scale_0.9 16.13
255
+ ngram_lm_scale_0.7_attention_scale_4.0 16.15
256
+ ngram_lm_scale_0.7_attention_scale_5.0 16.15
257
+ ngram_lm_scale_0.5_attention_scale_0.3 16.16
258
+ ngram_lm_scale_0.7_attention_scale_3.0 16.16
259
+ ngram_lm_scale_0.7_attention_scale_2.5 16.17
260
+ ngram_lm_scale_0.7_attention_scale_2.3 16.18
261
+ ngram_lm_scale_0.7_attention_scale_2.2 16.21
262
+ ngram_lm_scale_0.7_attention_scale_2.1 16.22
263
+ ngram_lm_scale_0.6_attention_scale_0.7 16.23
264
+ ngram_lm_scale_0.7_attention_scale_1.9 16.23
265
+ ngram_lm_scale_0.7_attention_scale_2.0 16.23
266
+ ngram_lm_scale_0.7_attention_scale_1.7 16.25
267
+ ngram_lm_scale_0.7_attention_scale_1.5 16.29
268
+ ngram_lm_scale_0.6_attention_scale_0.6 16.31
269
+ ngram_lm_scale_0.7_attention_scale_1.3 16.33
270
+ ngram_lm_scale_0.9_attention_scale_5.0 16.36
271
+ ngram_lm_scale_0.7_attention_scale_1.2 16.38
272
+ ngram_lm_scale_0.9_attention_scale_4.0 16.38
273
+ ngram_lm_scale_0.7_attention_scale_1.1 16.42
274
+ ngram_lm_scale_0.6_attention_scale_0.5 16.43
275
+ ngram_lm_scale_0.7_attention_scale_1.0 16.45
276
+ ngram_lm_scale_1.0_attention_scale_5.0 16.47
277
+ ngram_lm_scale_0.9_attention_scale_3.0 16.5
278
+ ngram_lm_scale_1.0_attention_scale_4.0 16.55
279
+ ngram_lm_scale_0.7_attention_scale_0.9 16.56
280
+ ngram_lm_scale_1.1_attention_scale_5.0 16.58
281
+ ngram_lm_scale_0.9_attention_scale_2.5 16.59
282
+ ngram_lm_scale_0.9_attention_scale_2.3 16.64
283
+ ngram_lm_scale_0.9_attention_scale_2.2 16.67
284
+ ngram_lm_scale_1.0_attention_scale_3.0 16.67
285
+ ngram_lm_scale_1.1_attention_scale_4.0 16.69
286
+ ngram_lm_scale_1.2_attention_scale_5.0 16.69
287
+ ngram_lm_scale_0.9_attention_scale_2.1 16.7
288
+ ngram_lm_scale_0.9_attention_scale_2.0 16.73
289
+ ngram_lm_scale_0.7_attention_scale_0.7 16.75
290
+ ngram_lm_scale_0.9_attention_scale_1.9 16.76
291
+ ngram_lm_scale_0.5_attention_scale_0.1 16.77
292
+ ngram_lm_scale_1.0_attention_scale_2.5 16.8
293
+ ngram_lm_scale_1.3_attention_scale_5.0 16.81
294
+ ngram_lm_scale_0.6_attention_scale_0.3 16.84
295
+ ngram_lm_scale_1.2_attention_scale_4.0 16.85
296
+ ngram_lm_scale_0.9_attention_scale_1.7 16.86
297
+ ngram_lm_scale_1.1_attention_scale_3.0 16.88
298
+ ngram_lm_scale_0.7_attention_scale_0.6 16.9
299
+ ngram_lm_scale_1.0_attention_scale_2.3 16.91
300
+ ngram_lm_scale_0.5_attention_scale_0.08 16.92
301
+ ngram_lm_scale_1.0_attention_scale_2.2 16.95
302
+ ngram_lm_scale_0.9_attention_scale_1.5 16.99
303
+ ngram_lm_scale_1.0_attention_scale_2.1 17.03
304
+ ngram_lm_scale_1.3_attention_scale_4.0 17.04
305
+ ngram_lm_scale_1.0_attention_scale_2.0 17.06
306
+ ngram_lm_scale_0.7_attention_scale_0.5 17.1
307
+ ngram_lm_scale_1.0_attention_scale_1.9 17.11
308
+ ngram_lm_scale_1.1_attention_scale_2.5 17.11
309
+ ngram_lm_scale_1.5_attention_scale_5.0 17.13
310
+ ngram_lm_scale_0.5_attention_scale_0.05 17.16
311
+ ngram_lm_scale_1.2_attention_scale_3.0 17.16
312
+ ngram_lm_scale_1.1_attention_scale_2.3 17.2
313
+ ngram_lm_scale_1.1_attention_scale_2.2 17.25
314
+ ngram_lm_scale_0.9_attention_scale_1.3 17.26
315
+ ngram_lm_scale_1.0_attention_scale_1.7 17.26
316
+ ngram_lm_scale_1.1_attention_scale_2.1 17.33
317
+ ngram_lm_scale_0.9_attention_scale_1.2 17.34
318
+ ngram_lm_scale_1.3_attention_scale_3.0 17.36
319
+ ngram_lm_scale_1.2_attention_scale_2.5 17.37
320
+ ngram_lm_scale_1.5_attention_scale_4.0 17.38
321
+ ngram_lm_scale_1.1_attention_scale_2.0 17.4
322
+ ngram_lm_scale_1.7_attention_scale_5.0 17.43
323
+ ngram_lm_scale_0.5_attention_scale_0.01 17.45
324
+ ngram_lm_scale_1.0_attention_scale_1.5 17.46
325
+ ngram_lm_scale_1.1_attention_scale_1.9 17.47
326
+ ngram_lm_scale_0.9_attention_scale_1.1 17.48
327
+ ngram_lm_scale_1.2_attention_scale_2.3 17.48
328
+ ngram_lm_scale_1.2_attention_scale_2.2 17.58
329
+ ngram_lm_scale_0.9_attention_scale_1.0 17.61
330
+ ngram_lm_scale_1.1_attention_scale_1.7 17.68
331
+ ngram_lm_scale_1.2_attention_scale_2.1 17.69
332
+ ngram_lm_scale_1.3_attention_scale_2.5 17.71
333
+ ngram_lm_scale_1.0_attention_scale_1.3 17.72
334
+ ngram_lm_scale_0.9_attention_scale_0.9 17.77
335
+ ngram_lm_scale_1.2_attention_scale_2.0 17.77
336
+ ngram_lm_scale_1.9_attention_scale_5.0 17.81
337
+ ngram_lm_scale_0.7_attention_scale_0.3 17.82
338
+ ngram_lm_scale_1.0_attention_scale_1.2 17.88
339
+ ngram_lm_scale_1.7_attention_scale_4.0 17.88
340
+ ngram_lm_scale_0.6_attention_scale_0.1 17.91
341
+ ngram_lm_scale_1.2_attention_scale_1.9 17.91
342
+ ngram_lm_scale_1.3_attention_scale_2.3 17.91
343
+ ngram_lm_scale_1.1_attention_scale_1.5 17.97
344
+ ngram_lm_scale_1.5_attention_scale_3.0 18.02
345
+ ngram_lm_scale_1.3_attention_scale_2.2 18.03
346
+ ngram_lm_scale_2.0_attention_scale_5.0 18.05
347
+ ngram_lm_scale_1.0_attention_scale_1.1 18.06
348
+ ngram_lm_scale_0.6_attention_scale_0.08 18.07
349
+ ngram_lm_scale_1.3_attention_scale_2.1 18.17
350
+ ngram_lm_scale_2.1_attention_scale_5.0 18.21
351
+ ngram_lm_scale_1.2_attention_scale_1.7 18.24
352
+ ngram_lm_scale_0.9_attention_scale_0.7 18.25
353
+ ngram_lm_scale_1.0_attention_scale_1.0 18.27
354
+ ngram_lm_scale_1.3_attention_scale_2.0 18.3
355
+ ngram_lm_scale_1.1_attention_scale_1.3 18.34
356
+ ngram_lm_scale_0.6_attention_scale_0.05 18.35
357
+ ngram_lm_scale_1.9_attention_scale_4.0 18.4
358
+ ngram_lm_scale_1.3_attention_scale_1.9 18.45
359
+ ngram_lm_scale_2.2_attention_scale_5.0 18.47
360
+ ngram_lm_scale_1.5_attention_scale_2.5 18.52
361
+ ngram_lm_scale_1.0_attention_scale_0.9 18.55
362
+ ngram_lm_scale_1.1_attention_scale_1.2 18.56
363
+ ngram_lm_scale_1.2_attention_scale_1.5 18.63
364
+ ngram_lm_scale_0.9_attention_scale_0.6 18.67
365
+ ngram_lm_scale_2.3_attention_scale_5.0 18.72
366
+ ngram_lm_scale_2.0_attention_scale_4.0 18.74
367
+ ngram_lm_scale_1.7_attention_scale_3.0 18.75
368
+ ngram_lm_scale_1.3_attention_scale_1.7 18.77
369
+ ngram_lm_scale_1.5_attention_scale_2.3 18.77
370
+ ngram_lm_scale_1.1_attention_scale_1.1 18.81
371
+ ngram_lm_scale_0.6_attention_scale_0.01 18.82
372
+ ngram_lm_scale_1.5_attention_scale_2.2 18.98
373
+ ngram_lm_scale_1.2_attention_scale_1.3 19.01
374
+ ngram_lm_scale_2.1_attention_scale_4.0 19.04
375
+ ngram_lm_scale_1.1_attention_scale_1.0 19.15
376
+ ngram_lm_scale_1.5_attention_scale_2.1 19.19
377
+ ngram_lm_scale_2.5_attention_scale_5.0 19.21
378
+ ngram_lm_scale_1.3_attention_scale_1.5 19.27
379
+ ngram_lm_scale_0.9_attention_scale_0.5 19.29
380
+ ngram_lm_scale_2.2_attention_scale_4.0 19.33
381
+ ngram_lm_scale_1.2_attention_scale_1.2 19.36
382
+ ngram_lm_scale_1.5_attention_scale_2.0 19.38
383
+ ngram_lm_scale_0.7_attention_scale_0.1 19.4
384
+ ngram_lm_scale_1.0_attention_scale_0.7 19.5
385
+ ngram_lm_scale_1.7_attention_scale_2.5 19.52
386
+ ngram_lm_scale_1.1_attention_scale_0.9 19.59
387
+ ngram_lm_scale_1.9_attention_scale_3.0 19.6
388
+ ngram_lm_scale_0.7_attention_scale_0.08 19.65
389
+ ngram_lm_scale_1.5_attention_scale_1.9 19.66
390
+ ngram_lm_scale_2.3_attention_scale_4.0 19.69
391
+ ngram_lm_scale_1.2_attention_scale_1.1 19.8
392
+ ngram_lm_scale_1.3_attention_scale_1.3 19.97
393
+ ngram_lm_scale_1.7_attention_scale_2.3 19.97
394
+ ngram_lm_scale_1.0_attention_scale_0.6 20.09
395
+ ngram_lm_scale_2.0_attention_scale_3.0 20.09
396
+ ngram_lm_scale_0.7_attention_scale_0.05 20.1
397
+ ngram_lm_scale_1.7_attention_scale_2.2 20.26
398
+ ngram_lm_scale_1.2_attention_scale_1.0 20.33
399
+ ngram_lm_scale_1.5_attention_scale_1.7 20.33
400
+ ngram_lm_scale_1.3_attention_scale_1.2 20.47
401
+ ngram_lm_scale_1.7_attention_scale_2.1 20.54
402
+ ngram_lm_scale_2.5_attention_scale_4.0 20.56
403
+ ngram_lm_scale_2.1_attention_scale_3.0 20.65
404
+ ngram_lm_scale_1.9_attention_scale_2.5 20.74
405
+ ngram_lm_scale_0.7_attention_scale_0.01 20.76
406
+ ngram_lm_scale_1.0_attention_scale_0.5 20.83
407
+ ngram_lm_scale_3.0_attention_scale_5.0 20.84
408
+ ngram_lm_scale_1.1_attention_scale_0.7 20.88
409
+ ngram_lm_scale_1.2_attention_scale_0.9 20.88
410
+ ngram_lm_scale_1.7_attention_scale_2.0 20.92
411
+ ngram_lm_scale_0.9_attention_scale_0.3 20.97
412
+ ngram_lm_scale_1.3_attention_scale_1.1 20.98
413
+ ngram_lm_scale_1.5_attention_scale_1.5 21.09
414
+ ngram_lm_scale_1.7_attention_scale_1.9 21.28
415
+ ngram_lm_scale_2.2_attention_scale_3.0 21.29
416
+ ngram_lm_scale_1.9_attention_scale_2.3 21.41
417
+ ngram_lm_scale_2.0_attention_scale_2.5 21.48
418
+ ngram_lm_scale_1.3_attention_scale_1.0 21.57
419
+ ngram_lm_scale_1.1_attention_scale_0.6 21.61
420
+ ngram_lm_scale_1.9_attention_scale_2.2 21.68
421
+ ngram_lm_scale_2.3_attention_scale_3.0 21.84
422
+ ngram_lm_scale_1.5_attention_scale_1.3 22.04
423
+ ngram_lm_scale_1.7_attention_scale_1.7 22.04
424
+ ngram_lm_scale_1.9_attention_scale_2.1 22.05
425
+ ngram_lm_scale_2.0_attention_scale_2.3 22.05
426
+ ngram_lm_scale_2.1_attention_scale_2.5 22.05
427
+ ngram_lm_scale_1.3_attention_scale_0.9 22.2
428
+ ngram_lm_scale_1.2_attention_scale_0.7 22.32
429
+ ngram_lm_scale_2.0_attention_scale_2.2 22.45
430
+ ngram_lm_scale_1.9_attention_scale_2.0 22.48
431
+ ngram_lm_scale_1.1_attention_scale_0.5 22.53
432
+ ngram_lm_scale_1.5_attention_scale_1.2 22.64
433
+ ngram_lm_scale_2.2_attention_scale_2.5 22.77
434
+ ngram_lm_scale_2.1_attention_scale_2.3 22.86
435
+ ngram_lm_scale_2.0_attention_scale_2.1 22.92
436
+ ngram_lm_scale_3.0_attention_scale_4.0 22.92
437
+ ngram_lm_scale_1.0_attention_scale_0.3 22.94
438
+ ngram_lm_scale_1.9_attention_scale_1.9 22.99
439
+ ngram_lm_scale_2.5_attention_scale_3.0 23.05
440
+ ngram_lm_scale_1.7_attention_scale_1.5 23.11
441
+ ngram_lm_scale_1.2_attention_scale_0.6 23.21
442
+ ngram_lm_scale_2.1_attention_scale_2.2 23.3
443
+ ngram_lm_scale_1.5_attention_scale_1.1 23.35
444
+ ngram_lm_scale_2.0_attention_scale_2.0 23.35
445
+ ngram_lm_scale_2.3_attention_scale_2.5 23.66
446
+ ngram_lm_scale_2.2_attention_scale_2.3 23.72
447
+ ngram_lm_scale_0.9_attention_scale_0.1 23.75
448
+ ngram_lm_scale_1.3_attention_scale_0.7 23.81
449
+ ngram_lm_scale_2.1_attention_scale_2.1 23.81
450
+ ngram_lm_scale_2.0_attention_scale_1.9 23.85
451
+ ngram_lm_scale_1.9_attention_scale_1.7 24.02
452
+ ngram_lm_scale_0.9_attention_scale_0.08 24.16
453
+ ngram_lm_scale_1.5_attention_scale_1.0 24.16
454
+ ngram_lm_scale_2.2_attention_scale_2.2 24.16
455
+ ngram_lm_scale_1.2_attention_scale_0.5 24.29
456
+ ngram_lm_scale_2.1_attention_scale_2.0 24.31
457
+ ngram_lm_scale_1.7_attention_scale_1.3 24.36
458
+ ngram_lm_scale_2.3_attention_scale_2.3 24.49
459
+ ngram_lm_scale_2.2_attention_scale_2.1 24.62
460
+ ngram_lm_scale_0.9_attention_scale_0.05 24.74
461
+ ngram_lm_scale_2.1_attention_scale_1.9 24.83
462
+ ngram_lm_scale_1.3_attention_scale_0.6 24.9
463
+ ngram_lm_scale_2.3_attention_scale_2.2 24.95
464
+ ngram_lm_scale_2.5_attention_scale_2.5 25.02
465
+ ngram_lm_scale_1.1_attention_scale_0.3 25.03
466
+ ngram_lm_scale_2.0_attention_scale_1.7 25.05
467
+ ngram_lm_scale_1.5_attention_scale_0.9 25.06
468
+ ngram_lm_scale_4.0_attention_scale_5.0 25.06
469
+ ngram_lm_scale_2.2_attention_scale_2.0 25.13
470
+ ngram_lm_scale_1.7_attention_scale_1.2 25.22
471
+ ngram_lm_scale_1.9_attention_scale_1.5 25.34
472
+ ngram_lm_scale_2.3_attention_scale_2.1 25.43
473
+ ngram_lm_scale_0.9_attention_scale_0.01 25.64
474
+ ngram_lm_scale_2.2_attention_scale_1.9 25.69
475
+ ngram_lm_scale_2.1_attention_scale_1.7 25.96
476
+ ngram_lm_scale_2.3_attention_scale_2.0 25.98
477
+ ngram_lm_scale_2.5_attention_scale_2.3 26.04
478
+ ngram_lm_scale_1.7_attention_scale_1.1 26.05
479
+ ngram_lm_scale_1.3_attention_scale_0.5 26.14
480
+ ngram_lm_scale_3.0_attention_scale_3.0 26.24
481
+ ngram_lm_scale_2.0_attention_scale_1.5 26.28
482
+ ngram_lm_scale_1.0_attention_scale_0.1 26.31
483
+ ngram_lm_scale_2.5_attention_scale_2.2 26.54
484
+ ngram_lm_scale_2.3_attention_scale_1.9 26.56
485
+ ngram_lm_scale_1.9_attention_scale_1.3 26.7
486
+ ngram_lm_scale_1.0_attention_scale_0.08 26.74
487
+ ngram_lm_scale_1.7_attention_scale_1.0 26.88
488
+ ngram_lm_scale_2.2_attention_scale_1.7 26.92
489
+ ngram_lm_scale_2.5_attention_scale_2.1 27.08
490
+ ngram_lm_scale_1.5_attention_scale_0.7 27.13
491
+ ngram_lm_scale_1.2_attention_scale_0.3 27.25
492
+ ngram_lm_scale_1.0_attention_scale_0.05 27.33
493
+ ngram_lm_scale_2.1_attention_scale_1.5 27.36
494
+ ngram_lm_scale_1.9_attention_scale_1.2 27.52
495
+ ngram_lm_scale_2.5_attention_scale_2.0 27.71
496
+ ngram_lm_scale_1.7_attention_scale_0.9 27.86
497
+ ngram_lm_scale_2.0_attention_scale_1.3 27.87
498
+ ngram_lm_scale_4.0_attention_scale_4.0 27.91
499
+ ngram_lm_scale_2.3_attention_scale_1.7 27.92
500
+ ngram_lm_scale_1.0_attention_scale_0.01 28.24
501
+ ngram_lm_scale_2.5_attention_scale_1.9 28.37
502
+ ngram_lm_scale_1.5_attention_scale_0.6 28.42
503
+ ngram_lm_scale_2.2_attention_scale_1.5 28.43
504
+ ngram_lm_scale_1.9_attention_scale_1.1 28.49
505
+ ngram_lm_scale_3.0_attention_scale_2.5 28.64
506
+ ngram_lm_scale_1.1_attention_scale_0.1 28.71
507
+ ngram_lm_scale_2.0_attention_scale_1.2 28.76
508
+ ngram_lm_scale_5.0_attention_scale_5.0 28.94
509
+ ngram_lm_scale_2.1_attention_scale_1.3 28.98
510
+ ngram_lm_scale_1.1_attention_scale_0.08 29.13
511
+ ngram_lm_scale_1.3_attention_scale_0.3 29.25
512
+ ngram_lm_scale_2.3_attention_scale_1.5 29.34
513
+ ngram_lm_scale_1.9_attention_scale_1.0 29.45
514
+ ngram_lm_scale_2.0_attention_scale_1.1 29.62
515
+ ngram_lm_scale_2.5_attention_scale_1.7 29.63
516
+ ngram_lm_scale_3.0_attention_scale_2.3 29.63
517
+ ngram_lm_scale_1.5_attention_scale_0.5 29.71
518
+ ngram_lm_scale_1.1_attention_scale_0.05 29.74
519
+ ngram_lm_scale_2.1_attention_scale_1.2 29.78
520
+ ngram_lm_scale_2.2_attention_scale_1.3 29.92
521
+ ngram_lm_scale_1.7_attention_scale_0.7 30.08
522
+ ngram_lm_scale_3.0_attention_scale_2.2 30.18
523
+ ngram_lm_scale_1.9_attention_scale_0.9 30.39
524
+ ngram_lm_scale_2.0_attention_scale_1.0 30.48
525
+ ngram_lm_scale_1.1_attention_scale_0.01 30.56
526
+ ngram_lm_scale_2.1_attention_scale_1.1 30.61
527
+ ngram_lm_scale_3.0_attention_scale_2.1 30.65
528
+ ngram_lm_scale_2.2_attention_scale_1.2 30.76
529
+ ngram_lm_scale_1.2_attention_scale_0.1 30.8
530
+ ngram_lm_scale_2.3_attention_scale_1.3 30.82
531
+ ngram_lm_scale_2.5_attention_scale_1.5 30.9
532
+ ngram_lm_scale_3.0_attention_scale_2.0 31.15
533
+ ngram_lm_scale_1.2_attention_scale_0.08 31.18
534
+ ngram_lm_scale_1.7_attention_scale_0.6 31.23
535
+ ngram_lm_scale_2.0_attention_scale_0.9 31.43
536
+ ngram_lm_scale_4.0_attention_scale_3.0 31.46
537
+ ngram_lm_scale_2.1_attention_scale_1.0 31.48
538
+ ngram_lm_scale_2.2_attention_scale_1.1 31.52
539
+ ngram_lm_scale_2.3_attention_scale_1.2 31.56
540
+ ngram_lm_scale_5.0_attention_scale_4.0 31.6
541
+ ngram_lm_scale_3.0_attention_scale_1.9 31.72
542
+ ngram_lm_scale_1.2_attention_scale_0.05 31.77
543
+ ngram_lm_scale_2.3_attention_scale_1.1 32.34
544
+ ngram_lm_scale_2.5_attention_scale_1.3 32.37
545
+ ngram_lm_scale_2.1_attention_scale_0.9 32.38
546
+ ngram_lm_scale_2.2_attention_scale_1.0 32.39
547
+ ngram_lm_scale_1.9_attention_scale_0.7 32.41
548
+ ngram_lm_scale_1.7_attention_scale_0.5 32.42
549
+ ngram_lm_scale_1.2_attention_scale_0.01 32.46
550
+ ngram_lm_scale_1.5_attention_scale_0.3 32.5
551
+ ngram_lm_scale_1.3_attention_scale_0.1 32.52
552
+ ngram_lm_scale_1.3_attention_scale_0.08 32.84
553
+ ngram_lm_scale_3.0_attention_scale_1.7 32.85
554
+ ngram_lm_scale_2.5_attention_scale_1.2 33.03
555
+ ngram_lm_scale_2.2_attention_scale_0.9 33.12
556
+ ngram_lm_scale_2.3_attention_scale_1.0 33.12
557
+ ngram_lm_scale_2.0_attention_scale_0.7 33.27
558
+ ngram_lm_scale_4.0_attention_scale_2.5 33.32
559
+ ngram_lm_scale_1.3_attention_scale_0.05 33.33
560
+ ngram_lm_scale_1.9_attention_scale_0.6 33.39
561
+ ngram_lm_scale_2.5_attention_scale_1.1 33.65
562
+ ngram_lm_scale_2.3_attention_scale_0.9 33.81
563
+ ngram_lm_scale_1.3_attention_scale_0.01 33.86
564
+ ngram_lm_scale_3.0_attention_scale_1.5 33.86
565
+ ngram_lm_scale_2.1_attention_scale_0.7 33.99
566
+ ngram_lm_scale_4.0_attention_scale_2.3 34.1
567
+ ngram_lm_scale_2.0_attention_scale_0.6 34.13
568
+ ngram_lm_scale_2.5_attention_scale_1.0 34.24
569
+ ngram_lm_scale_1.9_attention_scale_0.5 34.3
570
+ ngram_lm_scale_4.0_attention_scale_2.2 34.4
571
+ ngram_lm_scale_5.0_attention_scale_3.0 34.41
572
+ ngram_lm_scale_1.7_attention_scale_0.3 34.6
573
+ ngram_lm_scale_2.2_attention_scale_0.7 34.63
574
+ ngram_lm_scale_4.0_attention_scale_2.1 34.72
575
+ ngram_lm_scale_3.0_attention_scale_1.3 34.74
576
+ ngram_lm_scale_2.1_attention_scale_0.6 34.76
577
+ ngram_lm_scale_2.5_attention_scale_0.9 34.83
578
+ ngram_lm_scale_1.5_attention_scale_0.1 34.93
579
+ ngram_lm_scale_2.0_attention_scale_0.5 34.93
580
+ ngram_lm_scale_4.0_attention_scale_2.0 35.0
581
+ ngram_lm_scale_1.5_attention_scale_0.08 35.12
582
+ ngram_lm_scale_2.3_attention_scale_0.7 35.17
583
+ ngram_lm_scale_3.0_attention_scale_1.2 35.25
584
+ ngram_lm_scale_4.0_attention_scale_1.9 35.32
585
+ ngram_lm_scale_2.2_attention_scale_0.6 35.34
586
+ ngram_lm_scale_1.5_attention_scale_0.05 35.44
587
+ ngram_lm_scale_2.1_attention_scale_0.5 35.46
588
+ ngram_lm_scale_3.0_attention_scale_1.1 35.65
589
+ ngram_lm_scale_5.0_attention_scale_2.5 35.65
590
+ ngram_lm_scale_2.3_attention_scale_0.6 35.73
591
+ ngram_lm_scale_1.5_attention_scale_0.01 35.85
592
+ ngram_lm_scale_1.9_attention_scale_0.3 35.87
593
+ ngram_lm_scale_2.2_attention_scale_0.5 35.9
594
+ ngram_lm_scale_2.5_attention_scale_0.7 35.95
595
+ ngram_lm_scale_4.0_attention_scale_1.7 35.95
596
+ ngram_lm_scale_3.0_attention_scale_1.0 36.05
597
+ ngram_lm_scale_5.0_attention_scale_2.3 36.07
598
+ ngram_lm_scale_1.7_attention_scale_0.1 36.25
599
+ ngram_lm_scale_2.3_attention_scale_0.5 36.28
600
+ ngram_lm_scale_2.0_attention_scale_0.3 36.29
601
+ ngram_lm_scale_5.0_attention_scale_2.2 36.3
602
+ ngram_lm_scale_2.5_attention_scale_0.6 36.38
603
+ ngram_lm_scale_1.7_attention_scale_0.08 36.42
604
+ ngram_lm_scale_3.0_attention_scale_0.9 36.47
605
+ ngram_lm_scale_4.0_attention_scale_1.5 36.52
606
+ ngram_lm_scale_5.0_attention_scale_2.1 36.53
607
+ ngram_lm_scale_1.7_attention_scale_0.05 36.62
608
+ ngram_lm_scale_2.1_attention_scale_0.3 36.67
609
+ ngram_lm_scale_5.0_attention_scale_2.0 36.72
610
+ ngram_lm_scale_2.5_attention_scale_0.5 36.89
611
+ ngram_lm_scale_5.0_attention_scale_1.9 36.94
612
+ ngram_lm_scale_1.7_attention_scale_0.01 36.95
613
+ ngram_lm_scale_2.2_attention_scale_0.3 36.99
614
+ ngram_lm_scale_4.0_attention_scale_1.3 37.0
615
+ ngram_lm_scale_1.9_attention_scale_0.1 37.16
616
+ ngram_lm_scale_3.0_attention_scale_0.7 37.17
617
+ ngram_lm_scale_2.3_attention_scale_0.3 37.28
618
+ ngram_lm_scale_4.0_attention_scale_1.2 37.28
619
+ ngram_lm_scale_1.9_attention_scale_0.08 37.3
620
+ ngram_lm_scale_5.0_attention_scale_1.7 37.35
621
+ ngram_lm_scale_1.9_attention_scale_0.05 37.5
622
+ ngram_lm_scale_2.0_attention_scale_0.1 37.51
623
+ ngram_lm_scale_3.0_attention_scale_0.6 37.51
624
+ ngram_lm_scale_4.0_attention_scale_1.1 37.55
625
+ ngram_lm_scale_2.0_attention_scale_0.08 37.6
626
+ ngram_lm_scale_1.9_attention_scale_0.01 37.67
627
+ ngram_lm_scale_5.0_attention_scale_1.5 37.73
628
+ ngram_lm_scale_2.5_attention_scale_0.3 37.74
629
+ ngram_lm_scale_2.0_attention_scale_0.05 37.75
630
+ ngram_lm_scale_2.1_attention_scale_0.1 37.77
631
+ ngram_lm_scale_4.0_attention_scale_1.0 37.79
632
+ ngram_lm_scale_2.1_attention_scale_0.08 37.85
633
+ ngram_lm_scale_3.0_attention_scale_0.5 37.85
634
+ ngram_lm_scale_2.0_attention_scale_0.01 37.94
635
+ ngram_lm_scale_2.2_attention_scale_0.1 37.96
636
+ ngram_lm_scale_2.1_attention_scale_0.05 37.98
637
+ ngram_lm_scale_4.0_attention_scale_0.9 38.03
638
+ ngram_lm_scale_2.2_attention_scale_0.08 38.05
639
+ ngram_lm_scale_5.0_attention_scale_1.3 38.1
640
+ ngram_lm_scale_2.3_attention_scale_0.1 38.18
641
+ ngram_lm_scale_2.1_attention_scale_0.01 38.19
642
+ ngram_lm_scale_2.2_attention_scale_0.05 38.21
643
+ ngram_lm_scale_2.3_attention_scale_0.08 38.28
644
+ ngram_lm_scale_5.0_attention_scale_1.2 38.3
645
+ ngram_lm_scale_2.2_attention_scale_0.01 38.36
646
+ ngram_lm_scale_2.3_attention_scale_0.05 38.38
647
+ ngram_lm_scale_5.0_attention_scale_1.1 38.45
648
+ ngram_lm_scale_2.5_attention_scale_0.1 38.5
649
+ ngram_lm_scale_4.0_attention_scale_0.7 38.53
650
+ ngram_lm_scale_2.3_attention_scale_0.01 38.54
651
+ ngram_lm_scale_3.0_attention_scale_0.3 38.55
652
+ ngram_lm_scale_2.5_attention_scale_0.08 38.57
653
+ ngram_lm_scale_2.5_attention_scale_0.05 38.67
654
+ ngram_lm_scale_5.0_attention_scale_1.0 38.7
655
+ ngram_lm_scale_4.0_attention_scale_0.6 38.76
656
+ ngram_lm_scale_2.5_attention_scale_0.01 38.83
657
+ ngram_lm_scale_5.0_attention_scale_0.9 38.89
658
+ ngram_lm_scale_4.0_attention_scale_0.5 39.02
659
+ ngram_lm_scale_3.0_attention_scale_0.1 39.18
660
+ ngram_lm_scale_5.0_attention_scale_0.7 39.22
661
+ ngram_lm_scale_3.0_attention_scale_0.08 39.23
662
+ ngram_lm_scale_3.0_attention_scale_0.05 39.31
663
+ ngram_lm_scale_5.0_attention_scale_0.6 39.39
664
+ ngram_lm_scale_3.0_attention_scale_0.01 39.4
665
+ ngram_lm_scale_4.0_attention_scale_0.3 39.4
666
+ ngram_lm_scale_5.0_attention_scale_0.5 39.52
667
+ ngram_lm_scale_4.0_attention_scale_0.1 39.75
668
+ ngram_lm_scale_4.0_attention_scale_0.08 39.77
669
+ ngram_lm_scale_5.0_attention_scale_0.3 39.77
670
+ ngram_lm_scale_4.0_attention_scale_0.05 39.84
671
+ ngram_lm_scale_4.0_attention_scale_0.01 39.92
672
+ ngram_lm_scale_5.0_attention_scale_0.1 40.05
673
+ ngram_lm_scale_5.0_attention_scale_0.08 40.07
674
+ ngram_lm_scale_5.0_attention_scale_0.05 40.1
675
+ ngram_lm_scale_5.0_attention_scale_0.01 40.17
676
+
677
+ 2022-06-27 19:39:22,402 INFO [decode.py:483] batch 0/?, cuts processed until now is 4
678
+ 2022-06-27 19:41:38,635 INFO [decode.py:483] batch 100/?, cuts processed until now is 428
679
+ 2022-06-27 19:43:58,275 INFO [decode.py:483] batch 200/?, cuts processed until now is 888
680
+ 2022-06-27 19:46:15,303 INFO [decode.py:483] batch 300/?, cuts processed until now is 1363
681
+ 2022-06-27 19:48:20,796 INFO [decode.py:483] batch 400/?, cuts processed until now is 1815
682
+ 2022-06-27 19:50:41,802 INFO [decode.py:483] batch 500/?, cuts processed until now is 2243
683
+ 2022-06-27 19:52:47,093 INFO [decode.py:483] batch 600/?, cuts processed until now is 2717
684
+ 2022-06-27 19:54:50,131 INFO [decode.py:483] batch 700/?, cuts processed until now is 3192
685
+ 2022-06-27 19:56:50,896 INFO [decode.py:783] Caught exception:
686
+ CUDA out of memory. Tried to allocate 8.00 GiB (GPU 0; 31.75 GiB total capacity; 16.89 GiB already allocated; 3.44 GiB free; 27.11 GiB reserved in total by PyTorch)
687
+ Exception raised from malloc at /pytorch/c10/cuda/CUDACachingAllocator.cpp:272 (most recent call first):
688
+ frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x42 (0x2aab0258d8b2 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10.so)
689
+ frame #1: <unknown function> + 0x2021b (0x2aab0232721b in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
690
+ frame #2: <unknown function> + 0x21034 (0x2aab02328034 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
691
+ frame #3: <unknown function> + 0x2167d (0x2aab0232867d in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
692
+ frame #4: k2::PytorchCudaContext::Allocate(unsigned long, void**) + 0x3a (0x2aab1173401a in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
693
+ frame #5: k2::NewRegion(std::shared_ptr<k2::Context>, unsigned long) + 0x112 (0x2aab11465b72 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
694
+ frame #6: k2::Hash::Hash(std::shared_ptr<k2::Context>, int, int, int) + 0x2f7 (0x2aab11538957 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
695
+ frame #7: k2::Hash::Resize(int, int, int, bool) + 0x1b4 (0x2aab1152e464 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
696
+ frame #8: k2::DeviceIntersector::ForwardSortedA() + 0x53e (0x2aab1156355e in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
697
+ frame #9: k2::IntersectDevice(k2::Ragged<k2::Arc>&, int, k2::Ragged<k2::Arc>&, int, k2::Array1<int> const&, k2::Array1<int>*, k2::Array1<int>*, bool) + 0x4cd (0x2aab115457ad in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
698
+ frame #10: <unknown function> + 0x8eb5a (0x2aab1032cb5a in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/_k2.cpython-38-x86_64-linux-gnu.so)
699
+ frame #11: <unknown function> + 0x3628c (0x2aab102d428c in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/_k2.cpython-38-x86_64-linux-gnu.so)
700
+ <omitting python frames>
701
+ frame #41: __libc_start_main + 0xf5 (0x2aaaab616555 in /lib64/libc.so.6)
702
+
703
+
704
+ 2022-06-27 19:56:50,897 INFO [decode.py:789] num_arcs before pruning: 1001553
705
+ 2022-06-27 19:56:50,897 INFO [decode.py:792] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
706
+ 2022-06-27 19:56:50,909 INFO [decode.py:803] num_arcs after pruning: 5815
707
+ 2022-06-27 19:57:04,620 INFO [decode.py:483] batch 800/?, cuts processed until now is 3635
708
+ 2022-06-27 19:59:30,814 INFO [decode.py:483] batch 900/?, cuts processed until now is 4082
709
+ 2022-06-27 20:00:05,093 INFO [decode.py:888] Caught exception:
710
+
711
+ Some bad things happened. Please read the above error messages and stack
712
+ trace. If you are using Python, the following command may be helpful:
713
+
714
+ gdb --args python /path/to/your/code.py
715
+
716
+ (You can use `gdb` to debug the code. Please consider compiling
717
+ a debug version of k2.).
718
+
719
+ If you are unable to fix it, please open an issue at:
720
+
721
+ https://github.com/k2-fsa/k2/issues/new
722
+
723
+
724
+ 2022-06-27 20:00:05,094 INFO [decode.py:889] num_paths before decreasing: 1000
725
+ 2022-06-27 20:00:05,094 INFO [decode.py:896] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
726
+ 2022-06-27 20:00:05,094 INFO [decode.py:902] num_paths after decreasing: 500
727
+ 2022-06-27 20:02:41,083 INFO [decode.py:483] batch 1000/?, cuts processed until now is 4500
728
+ 2022-06-27 20:05:44,787 INFO [decode.py:483] batch 1100/?, cuts processed until now is 4869
729
+ 2022-06-27 20:10:33,885 INFO [decode.py:532]
730
+ For dev, WER of different settings are:
731
+ ngram_lm_scale_0.01_attention_scale_0.5 15.89 best for dev
732
+ ngram_lm_scale_0.01_attention_scale_0.3 15.9
733
+ ngram_lm_scale_0.01_attention_scale_0.6 15.9
734
+ ngram_lm_scale_0.01_attention_scale_0.7 15.9
735
+ ngram_lm_scale_0.05_attention_scale_0.6 15.92
736
+ ngram_lm_scale_0.08_attention_scale_0.6 15.92
737
+ ngram_lm_scale_0.01_attention_scale_0.9 15.93
738
+ ngram_lm_scale_0.08_attention_scale_0.5 15.93
739
+ ngram_lm_scale_0.05_attention_scale_0.5 15.94
740
+ ngram_lm_scale_0.05_attention_scale_0.7 15.94
741
+ ngram_lm_scale_0.05_attention_scale_0.3 15.95
742
+ ngram_lm_scale_0.01_attention_scale_1.0 15.96
743
+ ngram_lm_scale_0.1_attention_scale_0.5 15.96
744
+ ngram_lm_scale_0.1_attention_scale_0.6 15.96
745
+ ngram_lm_scale_0.05_attention_scale_0.9 15.97
746
+ ngram_lm_scale_0.08_attention_scale_0.7 15.97
747
+ ngram_lm_scale_0.08_attention_scale_0.3 15.98
748
+ ngram_lm_scale_0.1_attention_scale_0.7 15.98
749
+ ngram_lm_scale_0.01_attention_scale_0.1 16.0
750
+ ngram_lm_scale_0.01_attention_scale_1.1 16.0
751
+ ngram_lm_scale_0.05_attention_scale_1.0 16.01
752
+ ngram_lm_scale_0.1_attention_scale_0.3 16.01
753
+ ngram_lm_scale_0.01_attention_scale_0.08 16.02
754
+ ngram_lm_scale_0.08_attention_scale_0.9 16.02
755
+ ngram_lm_scale_0.01_attention_scale_1.2 16.03
756
+ ngram_lm_scale_0.08_attention_scale_1.0 16.03
757
+ ngram_lm_scale_0.1_attention_scale_0.9 16.03
758
+ ngram_lm_scale_0.1_attention_scale_1.0 16.03
759
+ ngram_lm_scale_0.05_attention_scale_1.1 16.04
760
+ ngram_lm_scale_0.08_attention_scale_1.1 16.04
761
+ ngram_lm_scale_0.05_attention_scale_0.1 16.05
762
+ ngram_lm_scale_0.05_attention_scale_0.08 16.06
763
+ ngram_lm_scale_0.01_attention_scale_0.05 16.07
764
+ ngram_lm_scale_0.05_attention_scale_1.2 16.07
765
+ ngram_lm_scale_0.08_attention_scale_0.1 16.07
766
+ ngram_lm_scale_0.1_attention_scale_1.1 16.08
767
+ ngram_lm_scale_0.01_attention_scale_1.3 16.09
768
+ ngram_lm_scale_0.08_attention_scale_0.08 16.09
769
+ ngram_lm_scale_0.1_attention_scale_0.1 16.09
770
+ ngram_lm_scale_0.08_attention_scale_1.2 16.1
771
+ ngram_lm_scale_0.1_attention_scale_0.08 16.1
772
+ ngram_lm_scale_0.05_attention_scale_0.05 16.11
773
+ ngram_lm_scale_0.05_attention_scale_1.3 16.13
774
+ ngram_lm_scale_0.1_attention_scale_1.2 16.13
775
+ ngram_lm_scale_0.01_attention_scale_1.5 16.15
776
+ ngram_lm_scale_0.08_attention_scale_0.05 16.15
777
+ ngram_lm_scale_0.08_attention_scale_1.3 16.15
778
+ ngram_lm_scale_0.01_attention_scale_0.01 16.16
779
+ ngram_lm_scale_0.1_attention_scale_1.3 16.16
780
+ ngram_lm_scale_0.1_attention_scale_0.05 16.17
781
+ ngram_lm_scale_0.05_attention_scale_0.01 16.18
782
+ ngram_lm_scale_0.05_attention_scale_1.5 16.19
783
+ ngram_lm_scale_0.01_attention_scale_1.7 16.21
784
+ ngram_lm_scale_0.08_attention_scale_1.5 16.21
785
+ ngram_lm_scale_0.01_attention_scale_1.9 16.22
786
+ ngram_lm_scale_0.05_attention_scale_1.7 16.22
787
+ ngram_lm_scale_0.08_attention_scale_0.01 16.23
788
+ ngram_lm_scale_0.1_attention_scale_1.5 16.24
789
+ ngram_lm_scale_0.01_attention_scale_2.0 16.26
790
+ ngram_lm_scale_0.08_attention_scale_1.7 16.26
791
+ ngram_lm_scale_0.1_attention_scale_0.01 16.28
792
+ ngram_lm_scale_0.05_attention_scale_1.9 16.29
793
+ ngram_lm_scale_0.1_attention_scale_1.7 16.29
794
+ ngram_lm_scale_0.01_attention_scale_2.1 16.31
795
+ ngram_lm_scale_0.05_attention_scale_2.0 16.31
796
+ ngram_lm_scale_0.08_attention_scale_1.9 16.32
797
+ ngram_lm_scale_0.3_attention_scale_0.3 16.32
798
+ ngram_lm_scale_0.3_attention_scale_0.5 16.32
799
+ ngram_lm_scale_0.3_attention_scale_0.6 16.32
800
+ ngram_lm_scale_0.3_attention_scale_0.7 16.33
801
+ ngram_lm_scale_0.05_attention_scale_2.1 16.34
802
+ ngram_lm_scale_0.08_attention_scale_2.0 16.34
803
+ ngram_lm_scale_0.01_attention_scale_2.2 16.35
804
+ ngram_lm_scale_0.01_attention_scale_2.3 16.36
805
+ ngram_lm_scale_0.1_attention_scale_1.9 16.36
806
+ ngram_lm_scale_0.1_attention_scale_2.0 16.36
807
+ ngram_lm_scale_0.3_attention_scale_0.9 16.36
808
+ ngram_lm_scale_0.08_attention_scale_2.1 16.37
809
+ ngram_lm_scale_0.05_attention_scale_2.2 16.38
810
+ ngram_lm_scale_0.1_attention_scale_2.1 16.4
811
+ ngram_lm_scale_0.01_attention_scale_2.5 16.41
812
+ ngram_lm_scale_0.05_attention_scale_2.3 16.41
813
+ ngram_lm_scale_0.08_attention_scale_2.2 16.41
814
+ ngram_lm_scale_0.3_attention_scale_1.0 16.41
815
+ ngram_lm_scale_0.08_attention_scale_2.3 16.43
816
+ ngram_lm_scale_0.1_attention_scale_2.2 16.43
817
+ ngram_lm_scale_0.1_attention_scale_2.3 16.43
818
+ ngram_lm_scale_0.3_attention_scale_1.1 16.44
819
+ ngram_lm_scale_0.05_attention_scale_2.5 16.45
820
+ ngram_lm_scale_0.08_attention_scale_2.5 16.45
821
+ ngram_lm_scale_0.01_attention_scale_3.0 16.47
822
+ ngram_lm_scale_0.1_attention_scale_2.5 16.47
823
+ ngram_lm_scale_0.3_attention_scale_1.2 16.49
824
+ ngram_lm_scale_0.3_attention_scale_1.3 16.49
825
+ ngram_lm_scale_0.3_attention_scale_1.5 16.52
826
+ ngram_lm_scale_0.05_attention_scale_3.0 16.53
827
+ ngram_lm_scale_0.08_attention_scale_3.0 16.54
828
+ ngram_lm_scale_0.1_attention_scale_3.0 16.56
829
+ ngram_lm_scale_0.3_attention_scale_0.1 16.57
830
+ ngram_lm_scale_0.3_attention_scale_1.7 16.57
831
+ ngram_lm_scale_0.3_attention_scale_1.9 16.63
832
+ ngram_lm_scale_0.01_attention_scale_4.0 16.64
833
+ ngram_lm_scale_0.3_attention_scale_0.08 16.64
834
+ ngram_lm_scale_0.3_attention_scale_2.0 16.64
835
+ ngram_lm_scale_0.3_attention_scale_2.1 16.66
836
+ ngram_lm_scale_0.05_attention_scale_4.0 16.67
837
+ ngram_lm_scale_0.08_attention_scale_4.0 16.69
838
+ ngram_lm_scale_0.3_attention_scale_0.05 16.69
839
+ ngram_lm_scale_0.3_attention_scale_2.2 16.69
840
+ ngram_lm_scale_0.3_attention_scale_2.3 16.7
841
+ ngram_lm_scale_0.1_attention_scale_4.0 16.71
842
+ ngram_lm_scale_0.3_attention_scale_2.5 16.74
843
+ ngram_lm_scale_0.01_attention_scale_5.0 16.75
844
+ ngram_lm_scale_0.05_attention_scale_5.0 16.77
845
+ ngram_lm_scale_0.08_attention_scale_5.0 16.79
846
+ ngram_lm_scale_0.1_attention_scale_5.0 16.79
847
+ ngram_lm_scale_0.3_attention_scale_3.0 16.82
848
+ ngram_lm_scale_0.3_attention_scale_0.01 16.84
849
+ ngram_lm_scale_0.3_attention_scale_4.0 16.86
850
+ ngram_lm_scale_0.3_attention_scale_5.0 16.94
851
+ ngram_lm_scale_0.5_attention_scale_1.5 16.94
852
+ ngram_lm_scale_0.5_attention_scale_1.3 16.95
853
+ ngram_lm_scale_0.5_attention_scale_1.2 16.96
854
+ ngram_lm_scale_0.5_attention_scale_1.0 16.97
855
+ ngram_lm_scale_0.5_attention_scale_1.1 16.98
856
+ ngram_lm_scale_0.5_attention_scale_1.7 16.99
857
+ ngram_lm_scale_0.5_attention_scale_0.9 17.0
858
+ ngram_lm_scale_0.5_attention_scale_1.9 17.0
859
+ ngram_lm_scale_0.5_attention_scale_2.0 17.01
860
+ ngram_lm_scale_0.5_attention_scale_2.2 17.01
861
+ ngram_lm_scale_0.5_attention_scale_2.1 17.02
862
+ ngram_lm_scale_0.5_attention_scale_0.7 17.04
863
+ ngram_lm_scale_0.5_attention_scale_2.3 17.04
864
+ ngram_lm_scale_0.5_attention_scale_2.5 17.05
865
+ ngram_lm_scale_0.5_attention_scale_3.0 17.05
866
+ ngram_lm_scale_0.5_attention_scale_0.6 17.06
867
+ ngram_lm_scale_0.5_attention_scale_4.0 17.09
868
+ ngram_lm_scale_0.5_attention_scale_0.5 17.1
869
+ ngram_lm_scale_0.5_attention_scale_5.0 17.11
870
+ ngram_lm_scale_0.6_attention_scale_3.0 17.22
871
+ ngram_lm_scale_0.6_attention_scale_5.0 17.22
872
+ ngram_lm_scale_0.6_attention_scale_1.7 17.23
873
+ ngram_lm_scale_0.6_attention_scale_2.3 17.23
874
+ ngram_lm_scale_0.6_attention_scale_2.5 17.23
875
+ ngram_lm_scale_0.6_attention_scale_4.0 17.23
876
+ ngram_lm_scale_0.6_attention_scale_1.5 17.24
877
+ ngram_lm_scale_0.6_attention_scale_1.9 17.24
878
+ ngram_lm_scale_0.6_attention_scale_2.1 17.24
879
+ ngram_lm_scale_0.6_attention_scale_2.2 17.24
880
+ ngram_lm_scale_0.6_attention_scale_2.0 17.25
881
+ ngram_lm_scale_0.6_attention_scale_1.3 17.26
882
+ ngram_lm_scale_0.6_attention_scale_1.1 17.27
883
+ ngram_lm_scale_0.5_attention_scale_0.3 17.28
884
+ ngram_lm_scale_0.6_attention_scale_1.2 17.28
885
+ ngram_lm_scale_0.6_attention_scale_1.0 17.31
886
+ ngram_lm_scale_0.6_attention_scale_0.9 17.33
887
+ ngram_lm_scale_0.7_attention_scale_5.0 17.33
888
+ ngram_lm_scale_0.7_attention_scale_4.0 17.37
889
+ ngram_lm_scale_0.6_attention_scale_0.7 17.44
890
+ ngram_lm_scale_0.7_attention_scale_3.0 17.44
891
+ ngram_lm_scale_0.7_attention_scale_2.2 17.49
892
+ ngram_lm_scale_0.7_attention_scale_2.3 17.49
893
+ ngram_lm_scale_0.7_attention_scale_2.5 17.49
894
+ ngram_lm_scale_0.7_attention_scale_2.0 17.5
895
+ ngram_lm_scale_0.7_attention_scale_2.1 17.5
896
+ ngram_lm_scale_0.7_attention_scale_1.7 17.51
897
+ ngram_lm_scale_0.7_attention_scale_1.9 17.51
898
+ ngram_lm_scale_0.6_attention_scale_0.6 17.55
899
+ ngram_lm_scale_0.7_attention_scale_1.5 17.56
900
+ ngram_lm_scale_0.9_attention_scale_5.0 17.61
901
+ ngram_lm_scale_0.7_attention_scale_1.3 17.62
902
+ ngram_lm_scale_0.7_attention_scale_1.2 17.68
903
+ ngram_lm_scale_0.9_attention_scale_4.0 17.69
904
+ ngram_lm_scale_0.6_attention_scale_0.5 17.72
905
+ ngram_lm_scale_0.7_attention_scale_1.1 17.74
906
+ ngram_lm_scale_0.9_attention_scale_3.0 17.77
907
+ ngram_lm_scale_1.0_attention_scale_5.0 17.77
908
+ ngram_lm_scale_0.7_attention_scale_1.0 17.79
909
+ ngram_lm_scale_1.0_attention_scale_4.0 17.82
910
+ ngram_lm_scale_0.7_attention_scale_0.9 17.84
911
+ ngram_lm_scale_0.9_attention_scale_2.5 17.87
912
+ ngram_lm_scale_1.1_attention_scale_5.0 17.88
913
+ ngram_lm_scale_0.9_attention_scale_2.3 17.93
914
+ ngram_lm_scale_0.5_attention_scale_0.1 17.95
915
+ ngram_lm_scale_0.9_attention_scale_2.2 17.96
916
+ ngram_lm_scale_0.9_attention_scale_2.1 18.0
917
+ ngram_lm_scale_1.0_attention_scale_3.0 18.0
918
+ ngram_lm_scale_1.1_attention_scale_4.0 18.01
919
+ ngram_lm_scale_1.2_attention_scale_5.0 18.02
920
+ ngram_lm_scale_0.9_attention_scale_2.0 18.03
921
+ ngram_lm_scale_0.5_attention_scale_0.08 18.05
922
+ ngram_lm_scale_0.7_attention_scale_0.7 18.07
923
+ ngram_lm_scale_0.9_attention_scale_1.9 18.09
924
+ ngram_lm_scale_0.6_attention_scale_0.3 18.1
925
+ ngram_lm_scale_1.3_attention_scale_5.0 18.15
926
+ ngram_lm_scale_1.0_attention_scale_2.5 18.17
927
+ ngram_lm_scale_1.2_attention_scale_4.0 18.18
928
+ ngram_lm_scale_0.9_attention_scale_1.7 18.2
929
+ ngram_lm_scale_0.5_attention_scale_0.05 18.24
930
+ ngram_lm_scale_0.7_attention_scale_0.6 18.24
931
+ ngram_lm_scale_1.0_attention_scale_2.3 18.28
932
+ ngram_lm_scale_1.1_attention_scale_3.0 18.28
933
+ ngram_lm_scale_1.0_attention_scale_2.2 18.31
934
+ ngram_lm_scale_0.9_attention_scale_1.5 18.36
935
+ ngram_lm_scale_1.0_attention_scale_2.1 18.37
936
+ ngram_lm_scale_1.0_attention_scale_2.0 18.41
937
+ ngram_lm_scale_1.3_attention_scale_4.0 18.42
938
+ ngram_lm_scale_0.7_attention_scale_0.5 18.45
939
+ ngram_lm_scale_1.0_attention_scale_1.9 18.48
940
+ ngram_lm_scale_1.1_attention_scale_2.5 18.49
941
+ ngram_lm_scale_1.5_attention_scale_5.0 18.5
942
+ ngram_lm_scale_1.2_attention_scale_3.0 18.52
943
+ ngram_lm_scale_0.9_attention_scale_1.3 18.57
944
+ ngram_lm_scale_1.1_attention_scale_2.3 18.58
945
+ ngram_lm_scale_0.5_attention_scale_0.01 18.61
946
+ ngram_lm_scale_0.9_attention_scale_1.2 18.66
947
+ ngram_lm_scale_1.0_attention_scale_1.7 18.66
948
+ ngram_lm_scale_1.1_attention_scale_2.2 18.66
949
+ ngram_lm_scale_1.1_attention_scale_2.1 18.78
950
+ ngram_lm_scale_1.2_attention_scale_2.5 18.83
951
+ ngram_lm_scale_1.3_attention_scale_3.0 18.83
952
+ ngram_lm_scale_0.9_attention_scale_1.1 18.85
953
+ ngram_lm_scale_1.5_attention_scale_4.0 18.88
954
+ ngram_lm_scale_1.1_attention_scale_2.0 18.89
955
+ ngram_lm_scale_1.7_attention_scale_5.0 18.89
956
+ ngram_lm_scale_1.0_attention_scale_1.5 18.94
957
+ ngram_lm_scale_1.1_attention_scale_1.9 19.04
958
+ ngram_lm_scale_0.9_attention_scale_1.0 19.07
959
+ ngram_lm_scale_1.2_attention_scale_2.3 19.07
960
+ ngram_lm_scale_1.2_attention_scale_2.2 19.15
961
+ ngram_lm_scale_0.6_attention_scale_0.1 19.17
962
+ ngram_lm_scale_0.7_attention_scale_0.3 19.17
963
+ ngram_lm_scale_1.0_attention_scale_1.3 19.21
964
+ ngram_lm_scale_1.1_attention_scale_1.7 19.24
965
+ ngram_lm_scale_1.2_attention_scale_2.1 19.28
966
+ ngram_lm_scale_0.9_attention_scale_0.9 19.29
967
+ ngram_lm_scale_1.3_attention_scale_2.5 19.3
968
+ ngram_lm_scale_1.9_attention_scale_5.0 19.32
969
+ ngram_lm_scale_0.6_attention_scale_0.08 19.35
970
+ ngram_lm_scale_1.2_attention_scale_2.0 19.36
971
+ ngram_lm_scale_1.7_attention_scale_4.0 19.41
972
+ ngram_lm_scale_1.0_attention_scale_1.2 19.43
973
+ ngram_lm_scale_1.2_attention_scale_1.9 19.49
974
+ ngram_lm_scale_1.3_attention_scale_2.3 19.52
975
+ ngram_lm_scale_1.1_attention_scale_1.5 19.54
976
+ ngram_lm_scale_2.0_attention_scale_5.0 19.56
977
+ ngram_lm_scale_1.5_attention_scale_3.0 19.59
978
+ ngram_lm_scale_1.3_attention_scale_2.2 19.66
979
+ ngram_lm_scale_0.6_attention_scale_0.05 19.67
980
+ ngram_lm_scale_1.0_attention_scale_1.1 19.68
981
+ ngram_lm_scale_2.1_attention_scale_5.0 19.79
982
+ ngram_lm_scale_1.3_attention_scale_2.1 19.8
983
+ ngram_lm_scale_1.2_attention_scale_1.7 19.86
984
+ ngram_lm_scale_0.9_attention_scale_0.7 19.92
985
+ ngram_lm_scale_1.3_attention_scale_2.0 19.93
986
+ ngram_lm_scale_1.0_attention_scale_1.0 19.94
987
+ ngram_lm_scale_1.1_attention_scale_1.3 20.03
988
+ ngram_lm_scale_1.9_attention_scale_4.0 20.08
989
+ ngram_lm_scale_1.3_attention_scale_1.9 20.11
990
+ ngram_lm_scale_2.2_attention_scale_5.0 20.16
991
+ ngram_lm_scale_1.5_attention_scale_2.5 20.2
992
+ ngram_lm_scale_0.6_attention_scale_0.01 20.22
993
+ ngram_lm_scale_1.0_attention_scale_0.9 20.26
994
+ ngram_lm_scale_1.1_attention_scale_1.2 20.27
995
+ ngram_lm_scale_0.9_attention_scale_0.6 20.29
996
+ ngram_lm_scale_1.2_attention_scale_1.5 20.34
997
+ ngram_lm_scale_2.3_attention_scale_5.0 20.42
998
+ ngram_lm_scale_2.0_attention_scale_4.0 20.47
999
+ ngram_lm_scale_1.7_attention_scale_3.0 20.48
1000
+ ngram_lm_scale_1.3_attention_scale_1.7 20.56
1001
+ ngram_lm_scale_1.1_attention_scale_1.1 20.57
1002
+ ngram_lm_scale_1.5_attention_scale_2.3 20.61
1003
+ ngram_lm_scale_1.2_attention_scale_1.3 20.8
1004
+ ngram_lm_scale_1.5_attention_scale_2.2 20.8
1005
+ ngram_lm_scale_2.1_attention_scale_4.0 20.81
1006
+ ngram_lm_scale_0.9_attention_scale_0.5 20.85
1007
+ ngram_lm_scale_1.1_attention_scale_1.0 20.87
1008
+ ngram_lm_scale_0.7_attention_scale_0.1 20.93
1009
+ ngram_lm_scale_1.5_attention_scale_2.1 20.98
1010
+ ngram_lm_scale_2.5_attention_scale_5.0 21.02
1011
+ ngram_lm_scale_1.3_attention_scale_1.5 21.06
1012
+ ngram_lm_scale_1.0_attention_scale_0.7 21.09
1013
+ ngram_lm_scale_2.2_attention_scale_4.0 21.14
1014
+ ngram_lm_scale_1.2_attention_scale_1.2 21.21
1015
+ ngram_lm_scale_1.5_attention_scale_2.0 21.24
1016
+ ngram_lm_scale_0.7_attention_scale_0.08 21.25
1017
+ ngram_lm_scale_1.7_attention_scale_2.5 21.32
1018
+ ngram_lm_scale_1.1_attention_scale_0.9 21.38
1019
+ ngram_lm_scale_1.9_attention_scale_3.0 21.4
1020
+ ngram_lm_scale_1.5_attention_scale_1.9 21.5
1021
+ ngram_lm_scale_2.3_attention_scale_4.0 21.53
1022
+ ngram_lm_scale_1.2_attention_scale_1.1 21.61
1023
+ ngram_lm_scale_0.7_attention_scale_0.05 21.66
1024
+ ngram_lm_scale_1.0_attention_scale_0.6 21.68
1025
+ ngram_lm_scale_1.7_attention_scale_2.3 21.76
1026
+ ngram_lm_scale_1.3_attention_scale_1.3 21.78
1027
+ ngram_lm_scale_2.0_attention_scale_3.0 21.87
1028
+ ngram_lm_scale_1.7_attention_scale_2.2 21.99
1029
+ ngram_lm_scale_1.5_attention_scale_1.7 22.06
1030
+ ngram_lm_scale_1.2_attention_scale_1.0 22.09
1031
+ ngram_lm_scale_1.3_attention_scale_1.2 22.2
1032
+ ngram_lm_scale_2.5_attention_scale_4.0 22.25
1033
+ ngram_lm_scale_1.7_attention_scale_2.1 22.26
1034
+ ngram_lm_scale_0.7_attention_scale_0.01 22.32
1035
+ ngram_lm_scale_2.1_attention_scale_3.0 22.35
1036
+ ngram_lm_scale_1.9_attention_scale_2.5 22.46
1037
+ ngram_lm_scale_3.0_attention_scale_5.0 22.52
1038
+ ngram_lm_scale_1.1_attention_scale_0.7 22.56
1039
+ ngram_lm_scale_1.0_attention_scale_0.5 22.58
1040
+ ngram_lm_scale_0.9_attention_scale_0.3 22.61
1041
+ ngram_lm_scale_1.2_attention_scale_0.9 22.62
1042
+ ngram_lm_scale_1.7_attention_scale_2.0 22.63
1043
+ ngram_lm_scale_1.3_attention_scale_1.1 22.66
1044
+ ngram_lm_scale_1.5_attention_scale_1.5 22.76
1045
+ ngram_lm_scale_1.7_attention_scale_1.9 22.94
1046
+ ngram_lm_scale_2.2_attention_scale_3.0 22.94
1047
+ ngram_lm_scale_1.9_attention_scale_2.3 23.06
1048
+ ngram_lm_scale_2.0_attention_scale_2.5 23.14
1049
+ ngram_lm_scale_1.3_attention_scale_1.0 23.29
1050
+ ngram_lm_scale_1.1_attention_scale_0.6 23.36
1051
+ ngram_lm_scale_1.9_attention_scale_2.2 23.44
1052
+ ngram_lm_scale_2.3_attention_scale_3.0 23.55
1053
+ ngram_lm_scale_1.9_attention_scale_2.1 23.78
1054
+ ngram_lm_scale_1.7_attention_scale_1.7 23.79
1055
+ ngram_lm_scale_2.0_attention_scale_2.3 23.79
1056
+ ngram_lm_scale_1.5_attention_scale_1.3 23.8
1057
+ ngram_lm_scale_2.1_attention_scale_2.5 23.83
1058
+ ngram_lm_scale_1.3_attention_scale_0.9 23.96
1059
+ ngram_lm_scale_1.2_attention_scale_0.7 24.08
1060
+ ngram_lm_scale_1.9_attention_scale_2.0 24.22
1061
+ ngram_lm_scale_2.0_attention_scale_2.2 24.22
1062
+ ngram_lm_scale_1.1_attention_scale_0.5 24.3
1063
+ ngram_lm_scale_1.5_attention_scale_1.2 24.47
1064
+ ngram_lm_scale_2.2_attention_scale_2.5 24.56
1065
+ ngram_lm_scale_1.0_attention_scale_0.3 24.59
1066
+ ngram_lm_scale_2.1_attention_scale_2.3 24.6
1067
+ ngram_lm_scale_2.0_attention_scale_2.1 24.66
1068
+ ngram_lm_scale_3.0_attention_scale_4.0 24.67
1069
+ ngram_lm_scale_1.9_attention_scale_1.9 24.74
1070
+ ngram_lm_scale_2.5_attention_scale_3.0 24.82
1071
+ ngram_lm_scale_1.7_attention_scale_1.5 24.87
1072
+ ngram_lm_scale_1.2_attention_scale_0.6 25.02
1073
+ ngram_lm_scale_2.1_attention_scale_2.2 25.05
1074
+ ngram_lm_scale_2.0_attention_scale_2.0 25.16
1075
+ ngram_lm_scale_1.5_attention_scale_1.1 25.21
1076
+ ngram_lm_scale_2.3_attention_scale_2.5 25.31
1077
+ ngram_lm_scale_2.2_attention_scale_2.3 25.44
1078
+ ngram_lm_scale_2.1_attention_scale_2.1 25.49
1079
+ ngram_lm_scale_0.9_attention_scale_0.1 25.53
1080
+ ngram_lm_scale_2.0_attention_scale_1.9 25.63
1081
+ ngram_lm_scale_1.3_attention_scale_0.7 25.68
1082
+ ngram_lm_scale_1.9_attention_scale_1.7 25.78
1083
+ ngram_lm_scale_2.2_attention_scale_2.2 25.85
1084
+ ngram_lm_scale_0.9_attention_scale_0.08 25.91
1085
+ ngram_lm_scale_1.5_attention_scale_1.0 25.93
1086
+ ngram_lm_scale_2.1_attention_scale_2.0 26.0
1087
+ ngram_lm_scale_1.2_attention_scale_0.5 26.09
1088
+ ngram_lm_scale_1.7_attention_scale_1.3 26.16
1089
+ ngram_lm_scale_2.3_attention_scale_2.3 26.19
1090
+ ngram_lm_scale_2.2_attention_scale_2.1 26.34
1091
+ ngram_lm_scale_0.9_attention_scale_0.05 26.44
1092
+ ngram_lm_scale_2.1_attention_scale_1.9 26.5
1093
+ ngram_lm_scale_4.0_attention_scale_5.0 26.63
1094
+ ngram_lm_scale_2.3_attention_scale_2.2 26.64
1095
+ ngram_lm_scale_2.0_attention_scale_1.7 26.7
1096
+ ngram_lm_scale_1.3_attention_scale_0.6 26.71
1097
+ ngram_lm_scale_1.5_attention_scale_0.9 26.72
1098
+ ngram_lm_scale_2.5_attention_scale_2.5 26.78
1099
+ ngram_lm_scale_1.1_attention_scale_0.3 26.79
1100
+ ngram_lm_scale_2.2_attention_scale_2.0 26.83
1101
+ ngram_lm_scale_1.7_attention_scale_1.2 26.89
1102
+ ngram_lm_scale_1.9_attention_scale_1.5 26.98
1103
+ ngram_lm_scale_2.3_attention_scale_2.1 27.17
1104
+ ngram_lm_scale_0.9_attention_scale_0.01 27.35
1105
+ ngram_lm_scale_2.2_attention_scale_1.9 27.42
1106
+ ngram_lm_scale_1.7_attention_scale_1.1 27.64
1107
+ ngram_lm_scale_2.1_attention_scale_1.7 27.66
1108
+ ngram_lm_scale_2.3_attention_scale_2.0 27.69
1109
+ ngram_lm_scale_2.5_attention_scale_2.3 27.71
1110
+ ngram_lm_scale_1.0_attention_scale_0.1 27.84
1111
+ ngram_lm_scale_3.0_attention_scale_3.0 27.9
1112
+ ngram_lm_scale_1.3_attention_scale_0.5 27.92
1113
+ ngram_lm_scale_2.0_attention_scale_1.5 28.0
1114
+ ngram_lm_scale_2.5_attention_scale_2.2 28.22
1115
+ ngram_lm_scale_1.0_attention_scale_0.08 28.24
1116
+ ngram_lm_scale_2.3_attention_scale_1.9 28.24
1117
+ ngram_lm_scale_1.9_attention_scale_1.3 28.34
1118
+ ngram_lm_scale_1.7_attention_scale_1.0 28.47
1119
+ ngram_lm_scale_2.2_attention_scale_1.7 28.6
1120
+ ngram_lm_scale_2.5_attention_scale_2.1 28.74
1121
+ ngram_lm_scale_1.5_attention_scale_0.7 28.75
1122
+ ngram_lm_scale_1.2_attention_scale_0.3 28.83
1123
+ ngram_lm_scale_1.0_attention_scale_0.05 28.91
1124
+ ngram_lm_scale_2.1_attention_scale_1.5 28.96
1125
+ ngram_lm_scale_1.9_attention_scale_1.2 29.09
1126
+ ngram_lm_scale_2.5_attention_scale_2.0 29.23
1127
+ ngram_lm_scale_2.0_attention_scale_1.3 29.38
1128
+ ngram_lm_scale_2.3_attention_scale_1.7 29.39
1129
+ ngram_lm_scale_1.7_attention_scale_0.9 29.4
1130
+ ngram_lm_scale_4.0_attention_scale_4.0 29.41
1131
+ ngram_lm_scale_1.0_attention_scale_0.01 29.71
1132
+ ngram_lm_scale_2.5_attention_scale_1.9 29.79
1133
+ ngram_lm_scale_1.5_attention_scale_0.6 29.84
1134
+ ngram_lm_scale_2.2_attention_scale_1.5 29.84
1135
+ ngram_lm_scale_1.9_attention_scale_1.1 29.97
1136
+ ngram_lm_scale_3.0_attention_scale_2.5 30.05
1137
+ ngram_lm_scale_2.0_attention_scale_1.2 30.18
1138
+ ngram_lm_scale_1.1_attention_scale_0.1 30.22
1139
+ ngram_lm_scale_5.0_attention_scale_5.0 30.24
1140
+ ngram_lm_scale_2.1_attention_scale_1.3 30.34
1141
+ ngram_lm_scale_1.1_attention_scale_0.08 30.61
1142
+ ngram_lm_scale_1.3_attention_scale_0.3 30.69
1143
+ ngram_lm_scale_2.3_attention_scale_1.5 30.69
1144
+ ngram_lm_scale_1.9_attention_scale_1.0 30.8
1145
+ ngram_lm_scale_2.5_attention_scale_1.7 30.96
1146
+ ngram_lm_scale_3.0_attention_scale_2.3 31.0
1147
+ ngram_lm_scale_2.0_attention_scale_1.1 31.01
1148
+ ngram_lm_scale_1.5_attention_scale_0.5 31.06
1149
+ ngram_lm_scale_1.1_attention_scale_0.05 31.14
1150
+ ngram_lm_scale_2.1_attention_scale_1.2 31.14
1151
+ ngram_lm_scale_2.2_attention_scale_1.3 31.28
1152
+ ngram_lm_scale_1.7_attention_scale_0.7 31.42
1153
+ ngram_lm_scale_3.0_attention_scale_2.2 31.48
1154
+ ngram_lm_scale_1.9_attention_scale_0.9 31.72
1155
+ ngram_lm_scale_2.0_attention_scale_1.0 31.83
1156
+ ngram_lm_scale_2.1_attention_scale_1.1 31.92
1157
+ ngram_lm_scale_1.1_attention_scale_0.01 31.94
1158
+ ngram_lm_scale_3.0_attention_scale_2.1 31.98
1159
+ ngram_lm_scale_2.2_attention_scale_1.2 32.0
1160
+ ngram_lm_scale_2.3_attention_scale_1.3 32.07
1161
+ ngram_lm_scale_1.2_attention_scale_0.1 32.24
1162
+ ngram_lm_scale_2.5_attention_scale_1.5 32.27
1163
+ ngram_lm_scale_3.0_attention_scale_2.0 32.49
1164
+ ngram_lm_scale_1.7_attention_scale_0.6 32.53
1165
+ ngram_lm_scale_1.2_attention_scale_0.08 32.63
1166
+ ngram_lm_scale_2.0_attention_scale_0.9 32.71
1167
+ ngram_lm_scale_2.1_attention_scale_1.0 32.79
1168
+ ngram_lm_scale_2.2_attention_scale_1.1 32.82
1169
+ ngram_lm_scale_4.0_attention_scale_3.0 32.82
1170
+ ngram_lm_scale_2.3_attention_scale_1.2 32.89
1171
+ ngram_lm_scale_5.0_attention_scale_4.0 32.97
1172
+ ngram_lm_scale_3.0_attention_scale_1.9 33.03
1173
+ ngram_lm_scale_1.2_attention_scale_0.05 33.09
1174
+ ngram_lm_scale_2.5_attention_scale_1.3 33.56
1175
+ ngram_lm_scale_2.1_attention_scale_0.9 33.57
1176
+ ngram_lm_scale_2.2_attention_scale_1.0 33.58
1177
+ ngram_lm_scale_2.3_attention_scale_1.1 33.58
1178
+ ngram_lm_scale_1.9_attention_scale_0.7 33.66
1179
+ ngram_lm_scale_1.7_attention_scale_0.5 33.67
1180
+ ngram_lm_scale_1.5_attention_scale_0.3 33.73
1181
+ ngram_lm_scale_1.2_attention_scale_0.01 33.78
1182
+ ngram_lm_scale_1.3_attention_scale_0.1 33.86
1183
+ ngram_lm_scale_3.0_attention_scale_1.7 34.01
1184
+ ngram_lm_scale_1.3_attention_scale_0.08 34.16
1185
+ ngram_lm_scale_2.5_attention_scale_1.2 34.22
1186
+ ngram_lm_scale_2.3_attention_scale_1.0 34.32
1187
+ ngram_lm_scale_2.2_attention_scale_0.9 34.37
1188
+ ngram_lm_scale_2.0_attention_scale_0.7 34.46
1189
+ ngram_lm_scale_1.9_attention_scale_0.6 34.54
1190
+ ngram_lm_scale_4.0_attention_scale_2.5 34.54
1191
+ ngram_lm_scale_1.3_attention_scale_0.05 34.59
1192
+ ngram_lm_scale_2.5_attention_scale_1.1 34.83
1193
+ ngram_lm_scale_2.3_attention_scale_0.9 34.99
1194
+ ngram_lm_scale_3.0_attention_scale_1.5 35.02
1195
+ ngram_lm_scale_1.3_attention_scale_0.01 35.15
1196
+ ngram_lm_scale_2.1_attention_scale_0.7 35.19
1197
+ ngram_lm_scale_4.0_attention_scale_2.3 35.21
1198
+ ngram_lm_scale_2.0_attention_scale_0.6 35.3
1199
+ ngram_lm_scale_1.9_attention_scale_0.5 35.46
1200
+ ngram_lm_scale_2.5_attention_scale_1.0 35.46
1201
+ ngram_lm_scale_4.0_attention_scale_2.2 35.58
1202
+ ngram_lm_scale_5.0_attention_scale_3.0 35.61
1203
+ ngram_lm_scale_2.2_attention_scale_0.7 35.8
1204
+ ngram_lm_scale_1.7_attention_scale_0.3 35.87
1205
+ ngram_lm_scale_4.0_attention_scale_2.1 35.92
1206
+ ngram_lm_scale_2.1_attention_scale_0.6 35.98
1207
+ ngram_lm_scale_3.0_attention_scale_1.3 36.05
1208
+ ngram_lm_scale_2.5_attention_scale_0.9 36.08
1209
+ ngram_lm_scale_2.0_attention_scale_0.5 36.11
1210
+ ngram_lm_scale_1.5_attention_scale_0.1 36.23
1211
+ ngram_lm_scale_4.0_attention_scale_2.0 36.31
1212
+ ngram_lm_scale_2.3_attention_scale_0.7 36.34
1213
+ ngram_lm_scale_1.5_attention_scale_0.08 36.47
1214
+ ngram_lm_scale_3.0_attention_scale_1.2 36.47
1215
+ ngram_lm_scale_2.2_attention_scale_0.6 36.51
1216
+ ngram_lm_scale_4.0_attention_scale_1.9 36.59
1217
+ ngram_lm_scale_2.1_attention_scale_0.5 36.72
1218
+ ngram_lm_scale_1.5_attention_scale_0.05 36.74
1219
+ ngram_lm_scale_5.0_attention_scale_2.5 36.86
1220
+ ngram_lm_scale_3.0_attention_scale_1.1 36.93
1221
+ ngram_lm_scale_2.3_attention_scale_0.6 36.98
1222
+ ngram_lm_scale_1.5_attention_scale_0.01 37.11
1223
+ ngram_lm_scale_1.9_attention_scale_0.3 37.17
1224
+ ngram_lm_scale_2.5_attention_scale_0.7 37.17
1225
+ ngram_lm_scale_2.2_attention_scale_0.5 37.2
1226
+ ngram_lm_scale_4.0_attention_scale_1.7 37.21
1227
+ ngram_lm_scale_5.0_attention_scale_2.3 37.29
1228
+ ngram_lm_scale_3.0_attention_scale_1.0 37.35
1229
+ ngram_lm_scale_5.0_attention_scale_2.2 37.51
1230
+ ngram_lm_scale_2.3_attention_scale_0.5 37.61
1231
+ ngram_lm_scale_1.7_attention_scale_0.1 37.63
1232
+ ngram_lm_scale_2.0_attention_scale_0.3 37.63
1233
+ ngram_lm_scale_3.0_attention_scale_0.9 37.73
1234
+ ngram_lm_scale_5.0_attention_scale_2.1 37.75
1235
+ ngram_lm_scale_4.0_attention_scale_1.5 37.76
1236
+ ngram_lm_scale_2.5_attention_scale_0.6 37.78
1237
+ ngram_lm_scale_1.7_attention_scale_0.08 37.81
1238
+ ngram_lm_scale_5.0_attention_scale_2.0 37.96
1239
+ ngram_lm_scale_2.1_attention_scale_0.3 37.98
1240
+ ngram_lm_scale_1.7_attention_scale_0.05 38.01
1241
+ ngram_lm_scale_2.5_attention_scale_0.5 38.19
1242
+ ngram_lm_scale_5.0_attention_scale_1.9 38.2
1243
+ ngram_lm_scale_1.7_attention_scale_0.01 38.25
1244
+ ngram_lm_scale_2.2_attention_scale_0.3 38.28
1245
+ ngram_lm_scale_4.0_attention_scale_1.3 38.3
1246
+ ngram_lm_scale_1.9_attention_scale_0.1 38.43
1247
+ ngram_lm_scale_3.0_attention_scale_0.7 38.48
1248
+ ngram_lm_scale_2.3_attention_scale_0.3 38.53
1249
+ ngram_lm_scale_1.9_attention_scale_0.08 38.55
1250
+ ngram_lm_scale_4.0_attention_scale_1.2 38.56
1251
+ ngram_lm_scale_5.0_attention_scale_1.7 38.62
1252
+ ngram_lm_scale_1.9_attention_scale_0.05 38.69
1253
+ ngram_lm_scale_2.0_attention_scale_0.1 38.74
1254
+ ngram_lm_scale_3.0_attention_scale_0.6 38.82
1255
+ ngram_lm_scale_2.0_attention_scale_0.08 38.83
1256
+ ngram_lm_scale_4.0_attention_scale_1.1 38.84
1257
+ ngram_lm_scale_1.9_attention_scale_0.01 38.91
1258
+ ngram_lm_scale_2.0_attention_scale_0.05 38.96
1259
+ ngram_lm_scale_2.1_attention_scale_0.1 38.97
1260
+ ngram_lm_scale_2.5_attention_scale_0.3 38.97
1261
+ ngram_lm_scale_5.0_attention_scale_1.5 39.04
1262
+ ngram_lm_scale_2.1_attention_scale_0.08 39.05
1263
+ ngram_lm_scale_4.0_attention_scale_1.0 39.07
1264
+ ngram_lm_scale_3.0_attention_scale_0.5 39.14
1265
+ ngram_lm_scale_2.0_attention_scale_0.01 39.18
1266
+ ngram_lm_scale_2.2_attention_scale_0.1 39.22
1267
+ ngram_lm_scale_2.1_attention_scale_0.05 39.24
1268
+ ngram_lm_scale_2.2_attention_scale_0.08 39.31
1269
+ ngram_lm_scale_4.0_attention_scale_0.9 39.33
1270
+ ngram_lm_scale_5.0_attention_scale_1.3 39.41
1271
+ ngram_lm_scale_2.1_attention_scale_0.01 39.43
1272
+ ngram_lm_scale_2.3_attention_scale_0.1 39.46
1273
+ ngram_lm_scale_2.2_attention_scale_0.05 39.48
1274
+ ngram_lm_scale_2.3_attention_scale_0.08 39.52
1275
+ ngram_lm_scale_5.0_attention_scale_1.2 39.58
1276
+ ngram_lm_scale_2.2_attention_scale_0.01 39.6
1277
+ ngram_lm_scale_2.3_attention_scale_0.05 39.62
1278
+ ngram_lm_scale_2.5_attention_scale_0.1 39.71
1279
+ ngram_lm_scale_3.0_attention_scale_0.3 39.72
1280
+ ngram_lm_scale_2.3_attention_scale_0.01 39.74
1281
+ ngram_lm_scale_4.0_attention_scale_0.7 39.74
1282
+ ngram_lm_scale_5.0_attention_scale_1.1 39.74
1283
+ ngram_lm_scale_2.5_attention_scale_0.08 39.77
1284
+ ngram_lm_scale_2.5_attention_scale_0.05 39.89
1285
+ ngram_lm_scale_5.0_attention_scale_1.0 39.91
1286
+ ngram_lm_scale_4.0_attention_scale_0.6 39.95
1287
+ ngram_lm_scale_2.5_attention_scale_0.01 40.03
1288
+ ngram_lm_scale_5.0_attention_scale_0.9 40.07
1289
+ ngram_lm_scale_4.0_attention_scale_0.5 40.14
1290
+ ngram_lm_scale_3.0_attention_scale_0.1 40.27
1291
+ ngram_lm_scale_3.0_attention_scale_0.08 40.31
1292
+ ngram_lm_scale_5.0_attention_scale_0.7 40.34
1293
+ ngram_lm_scale_3.0_attention_scale_0.05 40.38
1294
+ ngram_lm_scale_3.0_attention_scale_0.01 40.47
1295
+ ngram_lm_scale_4.0_attention_scale_0.3 40.47
1296
+ ngram_lm_scale_5.0_attention_scale_0.6 40.49
1297
+ ngram_lm_scale_5.0_attention_scale_0.5 40.64
1298
+ ngram_lm_scale_4.0_attention_scale_0.1 40.84
1299
+ ngram_lm_scale_4.0_attention_scale_0.08 40.87
1300
+ ngram_lm_scale_4.0_attention_scale_0.05 40.91
1301
+ ngram_lm_scale_5.0_attention_scale_0.3 40.91
1302
+ ngram_lm_scale_4.0_attention_scale_0.01 40.96
1303
+ ngram_lm_scale_5.0_attention_scale_0.1 41.12
1304
+ ngram_lm_scale_5.0_attention_scale_0.08 41.17
1305
+ ngram_lm_scale_5.0_attention_scale_0.05 41.22
1306
+ ngram_lm_scale_5.0_attention_scale_0.01 41.29
1307
+
1308
+ 2022-06-27 20:10:33,885 INFO [decode.py:695] Done!
decoding-results/log-whole-lattice-rescoring/log-decode-2022-06-26-22-37-17 ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2022-06-26 22:37:17,938 INFO [decode.py:548] Decoding started
2
+ 2022-06-26 22:37:17,939 INFO [decode.py:549] {'subsampling_factor': 4, 'vgg_frontend': False, 'use_feat_batchnorm': True, 'feature_dim': 80, 'nhead': 8, 'attention_dim': 512, 'num_decoder_layers': 6, 'search_beam': 20, 'output_beam': 4, 'min_active_states': 30, 'max_active_states': 1000, 'use_double_scores': True, 'env_info': {'k2-version': '1.11', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '', 'k2-git-date': '', 'lhotse-version': '1.3.0.dev+git.a07121a.clean', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'test', 'icefall-git-sha1': '7e72d78-dirty', 'icefall-git-date': 'Sat May 28 19:13:53 2022', 'icefall-path': '/alt-arabic/speech/amir/k2/tmp/icefall', 'k2-path': '/home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/k2/__init__.py', 'lhotse-path': '/alt-arabic/speech/amir/k2/tmp/lhotse/lhotse/__init__.py', 'hostname': 'crimv3mgpu008', 'IP address': '10.141.0.6'}, 'epoch': 44, 'avg': 5, 'method': 'whole-lattice-rescoring', 'num_paths': 100, 'nbest_scale': 0.2, 'exp_dir': PosixPath('conformer_ctc/exp_5000_att0.8'), 'lang_dir': PosixPath('data/lang_bpe_5000'), 'lm_dir': PosixPath('data/lm'), 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 30, 'bucketing_sampler': True, 'num_buckets': 30, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'on_the_fly_feats': False, 'shuffle': True, 'drop_last': True, 'return_cuts': True, 'num_workers': 8, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True}
3
+ 2022-06-26 22:37:18,176 INFO [lexicon.py:177] Loading pre-compiled data/lang_bpe_5000/Linv.pt
4
+ 2022-06-26 22:37:18,206 INFO [decode.py:559] device: cuda:0
5
+ 2022-06-26 22:37:22,549 INFO [decode.py:621] Loading pre-compiled G_4_gram.pt
6
+ 2022-06-26 22:37:23,265 INFO [decode.py:657] averaging ['conformer_ctc/exp_5000_att0.8/epoch-40.pt', 'conformer_ctc/exp_5000_att0.8/epoch-41.pt', 'conformer_ctc/exp_5000_att0.8/epoch-42.pt', 'conformer_ctc/exp_5000_att0.8/epoch-43.pt', 'conformer_ctc/exp_5000_att0.8/epoch-44.pt']
7
+ 2022-06-26 22:38:42,767 INFO [decode.py:664] Number of model parameters: 90786736
8
+ 2022-06-26 22:38:42,768 INFO [asr_datamodule.py:374] About to get test cuts
9
+ 2022-06-26 22:38:42,799 INFO [asr_datamodule.py:367] About to get dev cuts
10
+ 2022-06-26 22:38:46,538 INFO [decode.py:483] batch 0/?, cuts processed until now is 4
11
+ 2022-06-26 22:39:19,008 INFO [decode.py:483] batch 100/?, cuts processed until now is 407
12
+ 2022-06-26 22:39:51,978 INFO [decode.py:483] batch 200/?, cuts processed until now is 839
13
+ 2022-06-26 22:40:24,708 INFO [decode.py:483] batch 300/?, cuts processed until now is 1272
14
+ 2022-06-26 22:40:58,103 INFO [decode.py:483] batch 400/?, cuts processed until now is 1702
15
+ 2022-06-26 22:41:30,530 INFO [decode.py:483] batch 500/?, cuts processed until now is 2109
16
+ 2022-06-26 22:42:03,513 INFO [decode.py:483] batch 600/?, cuts processed until now is 2544
17
+ 2022-06-26 22:42:35,204 INFO [decode.py:483] batch 700/?, cuts processed until now is 2978
18
+ 2022-06-26 22:43:08,567 INFO [decode.py:483] batch 800/?, cuts processed until now is 3384
19
+ 2022-06-26 22:43:40,827 INFO [decode.py:483] batch 900/?, cuts processed until now is 3811
20
+ 2022-06-26 22:44:13,123 INFO [decode.py:483] batch 1000/?, cuts processed until now is 4220
21
+ 2022-06-26 22:44:45,400 INFO [decode.py:483] batch 1100/?, cuts processed until now is 4631
22
+ 2022-06-26 22:45:18,544 INFO [decode.py:483] batch 1200/?, cuts processed until now is 5033
23
+ 2022-06-26 22:45:52,136 INFO [decode.py:483] batch 1300/?, cuts processed until now is 5355
24
+ 2022-06-26 22:45:53,135 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.1.txt
25
+ 2022-06-26 22:45:53,234 INFO [utils.py:404] [test-lm_scale_0.1] %WER 18.57% [11956 / 64388, 172 ins, 7090 del, 4694 sub ]
26
+ 2022-06-26 22:45:53,689 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.1.txt
27
+ 2022-06-26 22:45:53,741 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.2.txt
28
+ 2022-06-26 22:45:53,828 INFO [utils.py:404] [test-lm_scale_0.2] %WER 18.59% [11972 / 64388, 167 ins, 7142 del, 4663 sub ]
29
+ 2022-06-26 22:45:54,029 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.2.txt
30
+ 2022-06-26 22:45:54,073 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.3.txt
31
+ 2022-06-26 22:45:54,157 INFO [utils.py:404] [test-lm_scale_0.3] %WER 18.68% [12030 / 64388, 161 ins, 7235 del, 4634 sub ]
32
+ 2022-06-26 22:45:54,357 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.3.txt
33
+ 2022-06-26 22:45:54,401 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.4.txt
34
+ 2022-06-26 22:45:54,496 INFO [utils.py:404] [test-lm_scale_0.4] %WER 18.90% [12170 / 64388, 156 ins, 7412 del, 4602 sub ]
35
+ 2022-06-26 22:45:54,914 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.4.txt
36
+ 2022-06-26 22:45:54,961 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.5.txt
37
+ 2022-06-26 22:45:55,046 INFO [utils.py:404] [test-lm_scale_0.5] %WER 19.38% [12477 / 64388, 145 ins, 7769 del, 4563 sub ]
38
+ 2022-06-26 22:45:55,247 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.5.txt
39
+ 2022-06-26 22:45:55,302 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.6.txt
40
+ 2022-06-26 22:45:55,389 INFO [utils.py:404] [test-lm_scale_0.6] %WER 20.31% [13075 / 64388, 134 ins, 8461 del, 4480 sub ]
41
+ 2022-06-26 22:45:55,589 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.6.txt
42
+ 2022-06-26 22:45:55,640 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.7.txt
43
+ 2022-06-26 22:45:55,732 INFO [utils.py:404] [test-lm_scale_0.7] %WER 22.15% [14262 / 64388, 120 ins, 9817 del, 4325 sub ]
44
+ 2022-06-26 22:45:55,933 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.7.txt
45
+ 2022-06-26 22:45:55,980 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.8.txt
46
+ 2022-06-26 22:45:56,274 INFO [utils.py:404] [test-lm_scale_0.8] %WER 24.89% [16029 / 64388, 99 ins, 11848 del, 4082 sub ]
47
+ 2022-06-26 22:45:56,482 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.8.txt
48
+ 2022-06-26 22:45:56,526 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.9.txt
49
+ 2022-06-26 22:45:56,621 INFO [utils.py:404] [test-lm_scale_0.9] %WER 28.21% [18166 / 64388, 87 ins, 14210 del, 3869 sub ]
50
+ 2022-06-26 22:45:56,822 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.9.txt
51
+ 2022-06-26 22:45:56,862 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.0.txt
52
+ 2022-06-26 22:45:56,945 INFO [utils.py:404] [test-lm_scale_1.0] %WER 32.19% [20725 / 64388, 73 ins, 17007 del, 3645 sub ]
53
+ 2022-06-26 22:45:57,148 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.0.txt
54
+ 2022-06-26 22:45:57,192 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.1.txt
55
+ 2022-06-26 22:45:57,282 INFO [utils.py:404] [test-lm_scale_1.1] %WER 36.29% [23365 / 64388, 57 ins, 19898 del, 3410 sub ]
56
+ 2022-06-26 22:45:57,487 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.1.txt
57
+ 2022-06-26 22:45:57,526 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.2.txt
58
+ 2022-06-26 22:45:57,816 INFO [utils.py:404] [test-lm_scale_1.2] %WER 40.47% [26055 / 64388, 48 ins, 22805 del, 3202 sub ]
59
+ 2022-06-26 22:45:58,024 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.2.txt
60
+ 2022-06-26 22:45:58,066 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.3.txt
61
+ 2022-06-26 22:45:58,146 INFO [utils.py:404] [test-lm_scale_1.3] %WER 43.89% [28261 / 64388, 42 ins, 25155 del, 3064 sub ]
62
+ 2022-06-26 22:45:58,352 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.3.txt
63
+ 2022-06-26 22:45:58,390 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.4.txt
64
+ 2022-06-26 22:45:58,481 INFO [utils.py:404] [test-lm_scale_1.4] %WER 46.33% [29833 / 64388, 38 ins, 26847 del, 2948 sub ]
65
+ 2022-06-26 22:45:58,686 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.4.txt
66
+ 2022-06-26 22:45:58,725 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.5.txt
67
+ 2022-06-26 22:45:58,805 INFO [utils.py:404] [test-lm_scale_1.5] %WER 48.03% [30928 / 64388, 35 ins, 27980 del, 2913 sub ]
68
+ 2022-06-26 22:45:59,011 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.5.txt
69
+ 2022-06-26 22:45:59,048 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.6.txt
70
+ 2022-06-26 22:45:59,339 INFO [utils.py:404] [test-lm_scale_1.6] %WER 49.31% [31747 / 64388, 34 ins, 28810 del, 2903 sub ]
71
+ 2022-06-26 22:45:59,546 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.6.txt
72
+ 2022-06-26 22:45:59,585 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.7.txt
73
+ 2022-06-26 22:45:59,666 INFO [utils.py:404] [test-lm_scale_1.7] %WER 50.30% [32385 / 64388, 29 ins, 29439 del, 2917 sub ]
74
+ 2022-06-26 22:45:59,872 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.7.txt
75
+ 2022-06-26 22:45:59,919 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.8.txt
76
+ 2022-06-26 22:46:00,003 INFO [utils.py:404] [test-lm_scale_1.8] %WER 51.09% [32893 / 64388, 30 ins, 29947 del, 2916 sub ]
77
+ 2022-06-26 22:46:00,224 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.8.txt
78
+ 2022-06-26 22:46:00,264 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.9.txt
79
+ 2022-06-26 22:46:00,345 INFO [utils.py:404] [test-lm_scale_1.9] %WER 51.73% [33308 / 64388, 25 ins, 30381 del, 2902 sub ]
80
+ 2022-06-26 22:46:00,769 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.9.txt
81
+ 2022-06-26 22:46:00,817 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_2.0.txt
82
+ 2022-06-26 22:46:00,898 INFO [utils.py:404] [test-lm_scale_2.0] %WER 52.17% [33592 / 64388, 27 ins, 30665 del, 2900 sub ]
83
+ 2022-06-26 22:46:01,107 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_2.0.txt
84
+ 2022-06-26 22:46:01,113 INFO [decode.py:532]
85
+ For test, WER of different settings are:
86
+ lm_scale_0.1 18.57 best for test
87
+ lm_scale_0.2 18.59
88
+ lm_scale_0.3 18.68
89
+ lm_scale_0.4 18.9
90
+ lm_scale_0.5 19.38
91
+ lm_scale_0.6 20.31
92
+ lm_scale_0.7 22.15
93
+ lm_scale_0.8 24.89
94
+ lm_scale_0.9 28.21
95
+ lm_scale_1.0 32.19
96
+ lm_scale_1.1 36.29
97
+ lm_scale_1.2 40.47
98
+ lm_scale_1.3 43.89
99
+ lm_scale_1.4 46.33
100
+ lm_scale_1.5 48.03
101
+ lm_scale_1.6 49.31
102
+ lm_scale_1.7 50.3
103
+ lm_scale_1.8 51.09
104
+ lm_scale_1.9 51.73
105
+ lm_scale_2.0 52.17
106
+
107
+ 2022-06-26 22:46:02,285 INFO [decode.py:483] batch 0/?, cuts processed until now is 4
108
+ 2022-06-26 22:46:35,071 INFO [decode.py:483] batch 100/?, cuts processed until now is 428
109
+ 2022-06-26 22:47:08,653 INFO [decode.py:483] batch 200/?, cuts processed until now is 888
110
+ 2022-06-26 22:47:40,809 INFO [decode.py:483] batch 300/?, cuts processed until now is 1363
111
+ 2022-06-26 22:48:12,930 INFO [decode.py:483] batch 400/?, cuts processed until now is 1815
112
+ 2022-06-26 22:48:46,311 INFO [decode.py:483] batch 500/?, cuts processed until now is 2243
113
+ 2022-06-26 22:49:18,323 INFO [decode.py:483] batch 600/?, cuts processed until now is 2717
114
+ 2022-06-26 22:49:51,593 INFO [decode.py:483] batch 700/?, cuts processed until now is 3192
115
+ 2022-06-26 22:50:24,480 INFO [decode.py:483] batch 800/?, cuts processed until now is 3635
116
+ 2022-06-26 22:50:56,383 INFO [decode.py:483] batch 900/?, cuts processed until now is 4082
117
+ 2022-06-26 22:51:28,560 INFO [decode.py:483] batch 1000/?, cuts processed until now is 4500
118
+ 2022-06-26 22:52:02,855 INFO [decode.py:483] batch 1100/?, cuts processed until now is 4869
119
+ 2022-06-26 22:52:16,282 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.1.txt
120
+ 2022-06-26 22:52:16,364 INFO [utils.py:404] [dev-lm_scale_0.1] %WER 20.01% [12039 / 60169, 166 ins, 7746 del, 4127 sub ]
121
+ 2022-06-26 22:52:16,555 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.1.txt
122
+ 2022-06-26 22:52:16,596 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.2.txt
123
+ 2022-06-26 22:52:16,671 INFO [utils.py:404] [dev-lm_scale_0.2] %WER 20.07% [12077 / 60169, 155 ins, 7797 del, 4125 sub ]
124
+ 2022-06-26 22:52:16,857 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.2.txt
125
+ 2022-06-26 22:52:16,894 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.3.txt
126
+ 2022-06-26 22:52:16,980 INFO [utils.py:404] [dev-lm_scale_0.3] %WER 20.20% [12155 / 60169, 150 ins, 7886 del, 4119 sub ]
127
+ 2022-06-26 22:52:17,166 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.3.txt
128
+ 2022-06-26 22:52:17,215 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.4.txt
129
+ 2022-06-26 22:52:17,290 INFO [utils.py:404] [dev-lm_scale_0.4] %WER 20.44% [12301 / 60169, 142 ins, 8066 del, 4093 sub ]
130
+ 2022-06-26 22:52:17,484 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.4.txt
131
+ 2022-06-26 22:52:17,523 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.5.txt
132
+ 2022-06-26 22:52:17,598 INFO [utils.py:404] [dev-lm_scale_0.5] %WER 20.92% [12589 / 60169, 133 ins, 8415 del, 4041 sub ]
133
+ 2022-06-26 22:52:17,789 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.5.txt
134
+ 2022-06-26 22:52:17,826 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.6.txt
135
+ 2022-06-26 22:52:18,051 INFO [utils.py:404] [dev-lm_scale_0.6] %WER 21.95% [13210 / 60169, 114 ins, 9123 del, 3973 sub ]
136
+ 2022-06-26 22:52:18,244 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.6.txt
137
+ 2022-06-26 22:52:18,281 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.7.txt
138
+ 2022-06-26 22:52:18,357 INFO [utils.py:404] [dev-lm_scale_0.7] %WER 23.88% [14368 / 60169, 105 ins, 10478 del, 3785 sub ]
139
+ 2022-06-26 22:52:18,549 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.7.txt
140
+ 2022-06-26 22:52:18,584 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.8.txt
141
+ 2022-06-26 22:52:18,658 INFO [utils.py:404] [dev-lm_scale_0.8] %WER 26.64% [16029 / 60169, 89 ins, 12382 del, 3558 sub ]
142
+ 2022-06-26 22:52:18,850 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.8.txt
143
+ 2022-06-26 22:52:18,884 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.9.txt
144
+ 2022-06-26 22:52:18,959 INFO [utils.py:404] [dev-lm_scale_0.9] %WER 30.08% [18099 / 60169, 76 ins, 14702 del, 3321 sub ]
145
+ 2022-06-26 22:52:19,278 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.9.txt
146
+ 2022-06-26 22:52:19,324 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.0.txt
147
+ 2022-06-26 22:52:19,396 INFO [utils.py:404] [dev-lm_scale_1.0] %WER 33.80% [20337 / 60169, 67 ins, 17194 del, 3076 sub ]
148
+ 2022-06-26 22:52:19,587 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.0.txt
149
+ 2022-06-26 22:52:19,628 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.1.txt
150
+ 2022-06-26 22:52:19,704 INFO [utils.py:404] [dev-lm_scale_1.1] %WER 38.05% [22893 / 60169, 59 ins, 19979 del, 2855 sub ]
151
+ 2022-06-26 22:52:19,893 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.1.txt
152
+ 2022-06-26 22:52:19,926 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.2.txt
153
+ 2022-06-26 22:52:19,999 INFO [utils.py:404] [dev-lm_scale_1.2] %WER 42.04% [25295 / 60169, 47 ins, 22608 del, 2640 sub ]
154
+ 2022-06-26 22:52:20,188 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.2.txt
155
+ 2022-06-26 22:52:20,220 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.3.txt
156
+ 2022-06-26 22:52:20,413 INFO [utils.py:404] [dev-lm_scale_1.3] %WER 45.64% [27462 / 60169, 43 ins, 24922 del, 2497 sub ]
157
+ 2022-06-26 22:52:20,605 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.3.txt
158
+ 2022-06-26 22:52:20,638 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.4.txt
159
+ 2022-06-26 22:52:20,709 INFO [utils.py:404] [dev-lm_scale_1.4] %WER 48.05% [28913 / 60169, 42 ins, 26468 del, 2403 sub ]
160
+ 2022-06-26 22:52:20,900 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.4.txt
161
+ 2022-06-26 22:52:20,945 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.5.txt
162
+ 2022-06-26 22:52:21,026 INFO [utils.py:404] [dev-lm_scale_1.5] %WER 49.60% [29841 / 60169, 40 ins, 27451 del, 2350 sub ]
163
+ 2022-06-26 22:52:21,224 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.5.txt
164
+ 2022-06-26 22:52:21,258 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.6.txt
165
+ 2022-06-26 22:52:21,339 INFO [utils.py:404] [dev-lm_scale_1.6] %WER 50.69% [30499 / 60169, 40 ins, 28138 del, 2321 sub ]
166
+ 2022-06-26 22:52:21,662 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.6.txt
167
+ 2022-06-26 22:52:21,697 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.7.txt
168
+ 2022-06-26 22:52:21,769 INFO [utils.py:404] [dev-lm_scale_1.7] %WER 51.66% [31086 / 60169, 41 ins, 28748 del, 2297 sub ]
169
+ 2022-06-26 22:52:21,966 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.7.txt
170
+ 2022-06-26 22:52:21,999 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.8.txt
171
+ 2022-06-26 22:52:22,070 INFO [utils.py:404] [dev-lm_scale_1.8] %WER 52.42% [31540 / 60169, 39 ins, 29206 del, 2295 sub ]
172
+ 2022-06-26 22:52:22,268 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.8.txt
173
+ 2022-06-26 22:52:22,300 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.9.txt
174
+ 2022-06-26 22:52:22,370 INFO [utils.py:404] [dev-lm_scale_1.9] %WER 52.97% [31873 / 60169, 36 ins, 29517 del, 2320 sub ]
175
+ 2022-06-26 22:52:22,568 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.9.txt
176
+ 2022-06-26 22:52:22,608 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_2.0.txt
177
+ 2022-06-26 22:52:22,688 INFO [utils.py:404] [dev-lm_scale_2.0] %WER 53.42% [32142 / 60169, 37 ins, 29787 del, 2318 sub ]
178
+ 2022-06-26 22:52:23,010 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_2.0.txt
179
+ 2022-06-26 22:52:23,017 INFO [decode.py:532]
180
+ For dev, WER of different settings are:
181
+ lm_scale_0.1 20.01 best for dev
182
+ lm_scale_0.2 20.07
183
+ lm_scale_0.3 20.2
184
+ lm_scale_0.4 20.44
185
+ lm_scale_0.5 20.92
186
+ lm_scale_0.6 21.95
187
+ lm_scale_0.7 23.88
188
+ lm_scale_0.8 26.64
189
+ lm_scale_0.9 30.08
190
+ lm_scale_1.0 33.8
191
+ lm_scale_1.1 38.05
192
+ lm_scale_1.2 42.04
193
+ lm_scale_1.3 45.64
194
+ lm_scale_1.4 48.05
195
+ lm_scale_1.5 49.6
196
+ lm_scale_1.6 50.69
197
+ lm_scale_1.7 51.66
198
+ lm_scale_1.8 52.42
199
+ lm_scale_1.9 52.97
200
+ lm_scale_2.0 53.42
201
+
202
+ 2022-06-26 22:52:23,017 INFO [decode.py:695] Done!
decoding-results/log-whole-lattice-rescoring/log-decode-2022-06-26-23-11-51 ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2022-06-26 23:11:51,173 INFO [decode.py:548] Decoding started
2
+ 2022-06-26 23:11:51,174 INFO [decode.py:549] {'subsampling_factor': 4, 'vgg_frontend': False, 'use_feat_batchnorm': True, 'feature_dim': 80, 'nhead': 8, 'attention_dim': 512, 'num_decoder_layers': 6, 'search_beam': 20, 'output_beam': 4, 'min_active_states': 30, 'max_active_states': 1000, 'use_double_scores': True, 'env_info': {'k2-version': '1.11', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '', 'k2-git-date': '', 'lhotse-version': '1.3.0.dev+git.a07121a.clean', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'test', 'icefall-git-sha1': '7e72d78-dirty', 'icefall-git-date': 'Sat May 28 19:13:53 2022', 'icefall-path': '/alt-arabic/speech/amir/k2/tmp/icefall', 'k2-path': '/home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/k2/__init__.py', 'lhotse-path': '/alt-arabic/speech/amir/k2/tmp/lhotse/lhotse/__init__.py', 'hostname': 'crimv3mgpu008', 'IP address': '10.141.0.6'}, 'epoch': 44, 'avg': 5, 'method': 'whole-lattice-rescoring', 'num_paths': 1000, 'nbest_scale': 0.5, 'exp_dir': PosixPath('conformer_ctc/exp_5000_att0.8'), 'lang_dir': PosixPath('data/lang_bpe_5000'), 'lm_dir': PosixPath('data/lm'), 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 30, 'bucketing_sampler': True, 'num_buckets': 30, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'on_the_fly_feats': False, 'shuffle': True, 'drop_last': True, 'return_cuts': True, 'num_workers': 8, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True}
3
+ 2022-06-26 23:11:51,423 INFO [lexicon.py:177] Loading pre-compiled data/lang_bpe_5000/Linv.pt
4
+ 2022-06-26 23:11:51,454 INFO [decode.py:559] device: cuda:0
5
+ 2022-06-26 23:11:55,905 INFO [decode.py:621] Loading pre-compiled G_4_gram.pt
6
+ 2022-06-26 23:11:56,647 INFO [decode.py:657] averaging ['conformer_ctc/exp_5000_att0.8/epoch-40.pt', 'conformer_ctc/exp_5000_att0.8/epoch-41.pt', 'conformer_ctc/exp_5000_att0.8/epoch-42.pt', 'conformer_ctc/exp_5000_att0.8/epoch-43.pt', 'conformer_ctc/exp_5000_att0.8/epoch-44.pt']
7
+ 2022-06-26 23:12:00,210 INFO [decode.py:664] Number of model parameters: 90786736
8
+ 2022-06-26 23:12:00,210 INFO [asr_datamodule.py:374] About to get test cuts
9
+ 2022-06-26 23:12:00,214 INFO [asr_datamodule.py:367] About to get dev cuts
10
+ 2022-06-26 23:12:01,821 INFO [decode.py:483] batch 0/?, cuts processed until now is 4
11
+ 2022-06-26 23:12:34,348 INFO [decode.py:483] batch 100/?, cuts processed until now is 407
12
+ 2022-06-26 23:13:06,294 INFO [decode.py:483] batch 200/?, cuts processed until now is 839
13
+ 2022-06-26 23:13:38,794 INFO [decode.py:483] batch 300/?, cuts processed until now is 1272
14
+ 2022-06-26 23:14:11,304 INFO [decode.py:483] batch 400/?, cuts processed until now is 1702
15
+ 2022-06-26 23:14:43,455 INFO [decode.py:483] batch 500/?, cuts processed until now is 2109
16
+ 2022-06-26 23:15:16,309 INFO [decode.py:483] batch 600/?, cuts processed until now is 2544
17
+ 2022-06-26 23:15:50,970 INFO [decode.py:483] batch 700/?, cuts processed until now is 2978
18
+ 2022-06-26 23:16:25,113 INFO [decode.py:483] batch 800/?, cuts processed until now is 3384
19
+ 2022-06-26 23:16:58,532 INFO [decode.py:483] batch 900/?, cuts processed until now is 3811
20
+ 2022-06-26 23:17:33,161 INFO [decode.py:483] batch 1000/?, cuts processed until now is 4220
21
+ 2022-06-26 23:18:07,464 INFO [decode.py:483] batch 1100/?, cuts processed until now is 4631
22
+ 2022-06-26 23:18:42,532 INFO [decode.py:483] batch 1200/?, cuts processed until now is 5033
23
+ 2022-06-26 23:19:18,733 INFO [decode.py:483] batch 1300/?, cuts processed until now is 5355
24
+ 2022-06-26 23:19:19,776 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.1.txt
25
+ 2022-06-26 23:19:19,874 INFO [utils.py:404] [test-lm_scale_0.1] %WER 18.57% [11956 / 64388, 172 ins, 7090 del, 4694 sub ]
26
+ 2022-06-26 23:19:20,304 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.1.txt
27
+ 2022-06-26 23:19:20,367 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.2.txt
28
+ 2022-06-26 23:19:20,463 INFO [utils.py:404] [test-lm_scale_0.2] %WER 18.59% [11972 / 64388, 167 ins, 7142 del, 4663 sub ]
29
+ 2022-06-26 23:19:20,662 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.2.txt
30
+ 2022-06-26 23:19:20,707 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.3.txt
31
+ 2022-06-26 23:19:20,794 INFO [utils.py:404] [test-lm_scale_0.3] %WER 18.68% [12030 / 64388, 161 ins, 7235 del, 4634 sub ]
32
+ 2022-06-26 23:19:20,991 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.3.txt
33
+ 2022-06-26 23:19:21,035 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.4.txt
34
+ 2022-06-26 23:19:21,119 INFO [utils.py:404] [test-lm_scale_0.4] %WER 18.90% [12170 / 64388, 156 ins, 7412 del, 4602 sub ]
35
+ 2022-06-26 23:19:21,318 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.4.txt
36
+ 2022-06-26 23:19:21,361 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.5.txt
37
+ 2022-06-26 23:19:21,629 INFO [utils.py:404] [test-lm_scale_0.5] %WER 19.38% [12477 / 64388, 145 ins, 7769 del, 4563 sub ]
38
+ 2022-06-26 23:19:21,833 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.5.txt
39
+ 2022-06-26 23:19:21,877 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.6.txt
40
+ 2022-06-26 23:19:21,967 INFO [utils.py:404] [test-lm_scale_0.6] %WER 20.31% [13075 / 64388, 134 ins, 8461 del, 4480 sub ]
41
+ 2022-06-26 23:19:22,164 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.6.txt
42
+ 2022-06-26 23:19:22,210 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.7.txt
43
+ 2022-06-26 23:19:22,312 INFO [utils.py:404] [test-lm_scale_0.7] %WER 22.15% [14262 / 64388, 120 ins, 9817 del, 4325 sub ]
44
+ 2022-06-26 23:19:22,514 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.7.txt
45
+ 2022-06-26 23:19:22,568 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.8.txt
46
+ 2022-06-26 23:19:22,653 INFO [utils.py:404] [test-lm_scale_0.8] %WER 24.89% [16029 / 64388, 99 ins, 11848 del, 4082 sub ]
47
+ 2022-06-26 23:19:23,035 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.8.txt
48
+ 2022-06-26 23:19:23,081 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.9.txt
49
+ 2022-06-26 23:19:23,168 INFO [utils.py:404] [test-lm_scale_0.9] %WER 28.21% [18166 / 64388, 87 ins, 14210 del, 3869 sub ]
50
+ 2022-06-26 23:19:23,369 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.9.txt
51
+ 2022-06-26 23:19:23,422 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.0.txt
52
+ 2022-06-26 23:19:23,517 INFO [utils.py:404] [test-lm_scale_1.0] %WER 32.19% [20725 / 64388, 73 ins, 17007 del, 3645 sub ]
53
+ 2022-06-26 23:19:23,720 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.0.txt
54
+ 2022-06-26 23:19:23,764 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.1.txt
55
+ 2022-06-26 23:19:23,857 INFO [utils.py:404] [test-lm_scale_1.1] %WER 36.29% [23365 / 64388, 57 ins, 19898 del, 3410 sub ]
56
+ 2022-06-26 23:19:24,060 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.1.txt
57
+ 2022-06-26 23:19:24,105 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.2.txt
58
+ 2022-06-26 23:19:24,197 INFO [utils.py:404] [test-lm_scale_1.2] %WER 40.47% [26055 / 64388, 48 ins, 22805 del, 3202 sub ]
59
+ 2022-06-26 23:19:24,583 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.2.txt
60
+ 2022-06-26 23:19:24,627 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.3.txt
61
+ 2022-06-26 23:19:24,712 INFO [utils.py:404] [test-lm_scale_1.3] %WER 43.89% [28261 / 64388, 42 ins, 25155 del, 3064 sub ]
62
+ 2022-06-26 23:19:24,916 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.3.txt
63
+ 2022-06-26 23:19:24,956 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.4.txt
64
+ 2022-06-26 23:19:25,047 INFO [utils.py:404] [test-lm_scale_1.4] %WER 46.33% [29833 / 64388, 38 ins, 26847 del, 2948 sub ]
65
+ 2022-06-26 23:19:25,252 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.4.txt
66
+ 2022-06-26 23:19:25,297 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.5.txt
67
+ 2022-06-26 23:19:25,377 INFO [utils.py:404] [test-lm_scale_1.5] %WER 48.03% [30928 / 64388, 35 ins, 27980 del, 2913 sub ]
68
+ 2022-06-26 23:19:25,581 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.5.txt
69
+ 2022-06-26 23:19:25,626 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.6.txt
70
+ 2022-06-26 23:19:25,893 INFO [utils.py:404] [test-lm_scale_1.6] %WER 49.31% [31747 / 64388, 34 ins, 28810 del, 2903 sub ]
71
+ 2022-06-26 23:19:26,107 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.6.txt
72
+ 2022-06-26 23:19:26,150 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.7.txt
73
+ 2022-06-26 23:19:26,232 INFO [utils.py:404] [test-lm_scale_1.7] %WER 50.30% [32385 / 64388, 29 ins, 29439 del, 2917 sub ]
74
+ 2022-06-26 23:19:26,437 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.7.txt
75
+ 2022-06-26 23:19:26,476 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.8.txt
76
+ 2022-06-26 23:19:26,556 INFO [utils.py:404] [test-lm_scale_1.8] %WER 51.09% [32893 / 64388, 30 ins, 29947 del, 2916 sub ]
77
+ 2022-06-26 23:19:26,763 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.8.txt
78
+ 2022-06-26 23:19:26,804 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.9.txt
79
+ 2022-06-26 23:19:26,894 INFO [utils.py:404] [test-lm_scale_1.9] %WER 51.73% [33308 / 64388, 25 ins, 30381 del, 2902 sub ]
80
+ 2022-06-26 23:19:27,100 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.9.txt
81
+ 2022-06-26 23:19:27,141 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_2.0.txt
82
+ 2022-06-26 23:19:27,221 INFO [utils.py:404] [test-lm_scale_2.0] %WER 52.17% [33592 / 64388, 27 ins, 30665 del, 2900 sub ]
83
+ 2022-06-26 23:19:27,604 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_2.0.txt
84
+ 2022-06-26 23:19:27,609 INFO [decode.py:532]
85
+ For test, WER of different settings are:
86
+ lm_scale_0.1 18.57 best for test
87
+ lm_scale_0.2 18.59
88
+ lm_scale_0.3 18.68
89
+ lm_scale_0.4 18.9
90
+ lm_scale_0.5 19.38
91
+ lm_scale_0.6 20.31
92
+ lm_scale_0.7 22.15
93
+ lm_scale_0.8 24.89
94
+ lm_scale_0.9 28.21
95
+ lm_scale_1.0 32.19
96
+ lm_scale_1.1 36.29
97
+ lm_scale_1.2 40.47
98
+ lm_scale_1.3 43.89
99
+ lm_scale_1.4 46.33
100
+ lm_scale_1.5 48.03
101
+ lm_scale_1.6 49.31
102
+ lm_scale_1.7 50.3
103
+ lm_scale_1.8 51.09
104
+ lm_scale_1.9 51.73
105
+ lm_scale_2.0 52.17
106
+
107
+ 2022-06-26 23:19:28,540 INFO [decode.py:483] batch 0/?, cuts processed until now is 4
108
+ 2022-06-26 23:20:02,462 INFO [decode.py:483] batch 100/?, cuts processed until now is 428
109
+ 2022-06-26 23:20:38,014 INFO [decode.py:483] batch 200/?, cuts processed until now is 888
110
+ 2022-06-26 23:21:11,425 INFO [decode.py:483] batch 300/?, cuts processed until now is 1363
111
+ 2022-06-26 23:21:44,755 INFO [decode.py:483] batch 400/?, cuts processed until now is 1815
112
+ 2022-06-26 23:22:18,702 INFO [decode.py:483] batch 500/?, cuts processed until now is 2243
113
+ 2022-06-26 23:22:52,000 INFO [decode.py:483] batch 600/?, cuts processed until now is 2717
114
+ 2022-06-26 23:23:25,539 INFO [decode.py:483] batch 700/?, cuts processed until now is 3192
115
+ 2022-06-26 23:23:59,158 INFO [decode.py:483] batch 800/?, cuts processed until now is 3635
116
+ 2022-06-26 23:24:33,611 INFO [decode.py:483] batch 900/?, cuts processed until now is 4082
117
+ 2022-06-26 23:25:05,267 INFO [decode.py:483] batch 1000/?, cuts processed until now is 4500
118
+ 2022-06-26 23:25:38,452 INFO [decode.py:483] batch 1100/?, cuts processed until now is 4869
119
+ 2022-06-26 23:25:53,215 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.1.txt
120
+ 2022-06-26 23:25:53,522 INFO [utils.py:404] [dev-lm_scale_0.1] %WER 20.01% [12039 / 60169, 166 ins, 7746 del, 4127 sub ]
121
+ 2022-06-26 23:25:53,718 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.1.txt
122
+ 2022-06-26 23:25:53,767 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.2.txt
123
+ 2022-06-26 23:25:53,851 INFO [utils.py:404] [dev-lm_scale_0.2] %WER 20.07% [12077 / 60169, 155 ins, 7797 del, 4125 sub ]
124
+ 2022-06-26 23:25:54,041 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.2.txt
125
+ 2022-06-26 23:25:54,089 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.3.txt
126
+ 2022-06-26 23:25:54,170 INFO [utils.py:404] [dev-lm_scale_0.3] %WER 20.20% [12155 / 60169, 150 ins, 7886 del, 4119 sub ]
127
+ 2022-06-26 23:25:54,361 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.3.txt
128
+ 2022-06-26 23:25:54,409 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.4.txt
129
+ 2022-06-26 23:25:54,672 INFO [utils.py:404] [dev-lm_scale_0.4] %WER 20.44% [12301 / 60169, 142 ins, 8066 del, 4093 sub ]
130
+ 2022-06-26 23:25:54,866 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.4.txt
131
+ 2022-06-26 23:25:54,914 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.5.txt
132
+ 2022-06-26 23:25:54,994 INFO [utils.py:404] [dev-lm_scale_0.5] %WER 20.92% [12589 / 60169, 133 ins, 8415 del, 4041 sub ]
133
+ 2022-06-26 23:25:55,184 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.5.txt
134
+ 2022-06-26 23:25:55,234 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.6.txt
135
+ 2022-06-26 23:25:55,323 INFO [utils.py:404] [dev-lm_scale_0.6] %WER 21.95% [13210 / 60169, 114 ins, 9123 del, 3973 sub ]
136
+ 2022-06-26 23:25:55,512 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.6.txt
137
+ 2022-06-26 23:25:55,552 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.7.txt
138
+ 2022-06-26 23:25:55,632 INFO [utils.py:404] [dev-lm_scale_0.7] %WER 23.88% [14368 / 60169, 105 ins, 10478 del, 3785 sub ]
139
+ 2022-06-26 23:25:56,005 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.7.txt
140
+ 2022-06-26 23:25:56,053 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.8.txt
141
+ 2022-06-26 23:25:56,133 INFO [utils.py:404] [dev-lm_scale_0.8] %WER 26.64% [16029 / 60169, 89 ins, 12382 del, 3558 sub ]
142
+ 2022-06-26 23:25:56,324 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.8.txt
143
+ 2022-06-26 23:25:56,374 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.9.txt
144
+ 2022-06-26 23:25:56,463 INFO [utils.py:404] [dev-lm_scale_0.9] %WER 30.08% [18099 / 60169, 76 ins, 14702 del, 3321 sub ]
145
+ 2022-06-26 23:25:56,668 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.9.txt
146
+ 2022-06-26 23:25:56,720 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.0.txt
147
+ 2022-06-26 23:25:56,799 INFO [utils.py:404] [dev-lm_scale_1.0] %WER 33.80% [20337 / 60169, 67 ins, 17194 del, 3076 sub ]
148
+ 2022-06-26 23:25:57,000 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.0.txt
149
+ 2022-06-26 23:25:57,041 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.1.txt
150
+ 2022-06-26 23:25:57,119 INFO [utils.py:404] [dev-lm_scale_1.1] %WER 38.05% [22893 / 60169, 59 ins, 19979 del, 2855 sub ]
151
+ 2022-06-26 23:25:57,495 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.1.txt
152
+ 2022-06-26 23:25:57,535 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.2.txt
153
+ 2022-06-26 23:25:57,613 INFO [utils.py:404] [dev-lm_scale_1.2] %WER 42.04% [25295 / 60169, 47 ins, 22608 del, 2640 sub ]
154
+ 2022-06-26 23:25:57,806 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.2.txt
155
+ 2022-06-26 23:25:57,845 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.3.txt
156
+ 2022-06-26 23:25:57,931 INFO [utils.py:404] [dev-lm_scale_1.3] %WER 45.64% [27462 / 60169, 43 ins, 24922 del, 2497 sub ]
157
+ 2022-06-26 23:25:58,126 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.3.txt
158
+ 2022-06-26 23:25:58,167 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.4.txt
159
+ 2022-06-26 23:25:58,254 INFO [utils.py:404] [dev-lm_scale_1.4] %WER 48.05% [28913 / 60169, 42 ins, 26468 del, 2403 sub ]
160
+ 2022-06-26 23:25:58,462 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.4.txt
161
+ 2022-06-26 23:25:58,514 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.5.txt
162
+ 2022-06-26 23:25:58,762 INFO [utils.py:404] [dev-lm_scale_1.5] %WER 49.60% [29841 / 60169, 40 ins, 27451 del, 2350 sub ]
163
+ 2022-06-26 23:25:58,963 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.5.txt
164
+ 2022-06-26 23:25:59,014 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.6.txt
165
+ 2022-06-26 23:25:59,100 INFO [utils.py:404] [dev-lm_scale_1.6] %WER 50.69% [30499 / 60169, 40 ins, 28138 del, 2321 sub ]
166
+ 2022-06-26 23:25:59,297 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.6.txt
167
+ 2022-06-26 23:25:59,344 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.7.txt
168
+ 2022-06-26 23:25:59,431 INFO [utils.py:404] [dev-lm_scale_1.7] %WER 51.66% [31086 / 60169, 41 ins, 28748 del, 2297 sub ]
169
+ 2022-06-26 23:25:59,626 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.7.txt
170
+ 2022-06-26 23:25:59,665 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.8.txt
171
+ 2022-06-26 23:25:59,742 INFO [utils.py:404] [dev-lm_scale_1.8] %WER 52.42% [31540 / 60169, 39 ins, 29206 del, 2295 sub ]
172
+ 2022-06-26 23:25:59,936 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.8.txt
173
+ 2022-06-26 23:25:59,971 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.9.txt
174
+ 2022-06-26 23:26:00,234 INFO [utils.py:404] [dev-lm_scale_1.9] %WER 52.97% [31873 / 60169, 36 ins, 29517 del, 2320 sub ]
175
+ 2022-06-26 23:26:00,436 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.9.txt
176
+ 2022-06-26 23:26:00,486 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_2.0.txt
177
+ 2022-06-26 23:26:00,563 INFO [utils.py:404] [dev-lm_scale_2.0] %WER 53.42% [32142 / 60169, 37 ins, 29787 del, 2318 sub ]
178
+ 2022-06-26 23:26:00,760 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_2.0.txt
179
+ 2022-06-26 23:26:00,765 INFO [decode.py:532]
180
+ For dev, WER of different settings are:
181
+ lm_scale_0.1 20.01 best for dev
182
+ lm_scale_0.2 20.07
183
+ lm_scale_0.3 20.2
184
+ lm_scale_0.4 20.44
185
+ lm_scale_0.5 20.92
186
+ lm_scale_0.6 21.95
187
+ lm_scale_0.7 23.88
188
+ lm_scale_0.8 26.64
189
+ lm_scale_0.9 30.08
190
+ lm_scale_1.0 33.8
191
+ lm_scale_1.1 38.05
192
+ lm_scale_1.2 42.04
193
+ lm_scale_1.3 45.64
194
+ lm_scale_1.4 48.05
195
+ lm_scale_1.5 49.6
196
+ lm_scale_1.6 50.69
197
+ lm_scale_1.7 51.66
198
+ lm_scale_1.8 52.42
199
+ lm_scale_1.9 52.97
200
+ lm_scale_2.0 53.42
201
+
202
+ 2022-06-26 23:26:00,765 INFO [decode.py:695] Done!
decoding-results/log-whole-lattice-rescoring/log-decode-2022-06-26-23-21-46 ADDED
The diff for this file is too large to render. See raw diff
 
decoding-results/log-whole-lattice-rescoring/log-decode-2022-06-27-18-46-45 ADDED
@@ -0,0 +1,299 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2022-06-27 18:46:45,715 INFO [decode.py:548] Decoding started
2
+ 2022-06-27 18:46:45,716 INFO [decode.py:549] {'subsampling_factor': 4, 'vgg_frontend': False, 'use_feat_batchnorm': True, 'feature_dim': 80, 'nhead': 8, 'attention_dim': 512, 'num_decoder_layers': 6, 'search_beam': 20, 'output_beam': 8, 'min_active_states': 30, 'max_active_states': 5000, 'use_double_scores': True, 'env_info': {'k2-version': '1.16', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '3c606c27045750bbbb7a289d8b2b09825dea521a', 'k2-git-date': 'Mon Jun 27 03:06:58 2022', 'lhotse-version': '1.3.0.dev+git.a07121a.clean', 'torch-version': '1.7.1', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'test', 'icefall-git-sha1': 'e24e6ac-dirty', 'icefall-git-date': 'Mon Jun 27 01:23:06 2022', 'icefall-path': '/alt-arabic/speech/amir/k2/tmp/icefall', 'k2-path': '/alt-arabic/speech/amir/k2/tmp/k2/k2/python/k2/__init__.py', 'lhotse-path': '/alt-arabic/speech/amir/k2/tmp/lhotse/lhotse/__init__.py', 'hostname': 'crimv3mgpu016', 'IP address': '10.141.0.3'}, 'epoch': 45, 'avg': 5, 'method': 'whole-lattice-rescoring', 'num_paths': 1000, 'nbest_scale': 0.5, 'exp_dir': PosixPath('conformer_ctc/exp_5000_att0.8'), 'lang_dir': PosixPath('data/lang_bpe_5000'), 'lm_dir': PosixPath('data/lm'), 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 30, 'bucketing_sampler': True, 'num_buckets': 30, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'on_the_fly_feats': False, 'shuffle': True, 'drop_last': True, 'return_cuts': True, 'num_workers': 20, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True}
3
+ 2022-06-27 18:46:46,037 INFO [lexicon.py:177] Loading pre-compiled data/lang_bpe_5000/Linv.pt
4
+ 2022-06-27 18:46:46,339 INFO [decode.py:559] device: cuda:0
5
+ 2022-06-27 18:47:36,089 INFO [decode.py:621] Loading pre-compiled G_4_gram.pt
6
+ 2022-06-27 18:47:39,038 INFO [decode.py:657] averaging ['conformer_ctc/exp_5000_att0.8/epoch-41.pt', 'conformer_ctc/exp_5000_att0.8/epoch-42.pt', 'conformer_ctc/exp_5000_att0.8/epoch-43.pt', 'conformer_ctc/exp_5000_att0.8/epoch-44.pt', 'conformer_ctc/exp_5000_att0.8/epoch-45.pt']
7
+ 2022-06-27 18:48:49,442 INFO [decode.py:664] Number of model parameters: 90786736
8
+ 2022-06-27 18:48:49,442 INFO [asr_datamodule.py:362] About to get test cuts
9
+ 2022-06-27 18:48:49,491 INFO [asr_datamodule.py:357] About to get dev cuts
10
+ 2022-06-27 18:48:51,979 INFO [decode.py:483] batch 0/?, cuts processed until now is 4
11
+ 2022-06-27 18:49:23,121 INFO [decode.py:783] Caught exception:
12
+ CUDA out of memory. Tried to allocate 1.70 GiB (GPU 0; 31.75 GiB total capacity; 27.32 GiB already allocated; 400.50 MiB free; 30.16 GiB reserved in total by PyTorch)
13
+ Exception raised from malloc at /pytorch/c10/cuda/CUDACachingAllocator.cpp:272 (most recent call first):
14
+ frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x42 (0x2aab0258d8b2 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10.so)
15
+ frame #1: <unknown function> + 0x2021b (0x2aab0232721b in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
16
+ frame #2: <unknown function> + 0x21034 (0x2aab02328034 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
17
+ frame #3: <unknown function> + 0x2167d (0x2aab0232867d in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
18
+ frame #4: k2::PytorchCudaContext::Allocate(unsigned long, void**) + 0x3a (0x2aab1173401a in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
19
+ frame #5: k2::NewRegion(std::shared_ptr<k2::Context>, unsigned long) + 0x112 (0x2aab11465b72 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
20
+ frame #6: k2::Array1<int>::Init(std::shared_ptr<k2::Context>, int, k2::Dtype) + 0x71 (0x2aab11432f51 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
21
+ frame #7: <unknown function> + 0x2472bd (0x2aab115c32bd in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
22
+ frame #8: k2::RaggedShapeFromTotSizes(std::shared_ptr<k2::Context>, int, int const*) + 0x213 (0x2aab115c3b83 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
23
+ frame #9: k2::IndexAxis0(k2::RaggedShape&, k2::Array1<int> const&, k2::Array1<int>*) + 0x32c (0x2aab115d77ec in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
24
+ frame #10: k2::Index(k2::RaggedShape&, int, k2::Array1<int> const&, k2::Array1<int>*) + 0x353 (0x2aab115dc943 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
25
+ frame #11: k2::Ragged<k2::Arc> k2::DeviceIntersector::FormatOutputTpl<k2::Hash::PackedAccessor>(k2::Array1<int>*, k2::Array1<int>*) + 0x407 (0x2aab11552327 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
26
+ frame #12: k2::IntersectDevice(k2::Ragged<k2::Arc>&, int, k2::Ragged<k2::Arc>&, int, k2::Array1<int> const&, k2::Array1<int>*, k2::Array1<int>*, bool) + 0x3a2 (0x2aab11545682 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
27
+ frame #13: <unknown function> + 0x8eb5a (0x2aab1032cb5a in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/_k2.cpython-38-x86_64-linux-gnu.so)
28
+ frame #14: <unknown function> + 0x3628c (0x2aab102d428c in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/_k2.cpython-38-x86_64-linux-gnu.so)
29
+ <omitting python frames>
30
+ frame #44: __libc_start_main + 0xf5 (0x2aaaab616555 in /lib64/libc.so.6)
31
+
32
+
33
+ 2022-06-27 18:49:23,122 INFO [decode.py:789] num_arcs before pruning: 940457
34
+ 2022-06-27 18:49:23,123 INFO [decode.py:792] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
35
+ 2022-06-27 18:49:23,136 INFO [decode.py:803] num_arcs after pruning: 8198
36
+ 2022-06-27 18:49:32,702 INFO [decode.py:483] batch 100/?, cuts processed until now is 407
37
+ 2022-06-27 18:50:09,988 INFO [decode.py:483] batch 200/?, cuts processed until now is 839
38
+ 2022-06-27 18:50:47,777 INFO [decode.py:483] batch 300/?, cuts processed until now is 1272
39
+ 2022-06-27 18:51:25,388 INFO [decode.py:483] batch 400/?, cuts processed until now is 1702
40
+ 2022-06-27 18:52:00,700 INFO [decode.py:483] batch 500/?, cuts processed until now is 2109
41
+ 2022-06-27 18:52:39,254 INFO [decode.py:483] batch 600/?, cuts processed until now is 2544
42
+ 2022-06-27 18:53:16,422 INFO [decode.py:483] batch 700/?, cuts processed until now is 2978
43
+ 2022-06-27 18:53:54,612 INFO [decode.py:483] batch 800/?, cuts processed until now is 3384
44
+ 2022-06-27 18:54:31,824 INFO [decode.py:483] batch 900/?, cuts processed until now is 3811
45
+ 2022-06-27 18:55:06,114 INFO [decode.py:783] Caught exception:
46
+ CUDA out of memory. Tried to allocate 1.67 GiB (GPU 0; 31.75 GiB total capacity; 27.27 GiB already allocated; 1.06 GiB free; 29.49 GiB reserved in total by PyTorch)
47
+ Exception raised from malloc at /pytorch/c10/cuda/CUDACachingAllocator.cpp:272 (most recent call first):
48
+ frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x42 (0x2aab0258d8b2 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10.so)
49
+ frame #1: <unknown function> + 0x2021b (0x2aab0232721b in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
50
+ frame #2: <unknown function> + 0x21034 (0x2aab02328034 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
51
+ frame #3: <unknown function> + 0x2167d (0x2aab0232867d in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
52
+ frame #4: k2::PytorchCudaContext::Allocate(unsigned long, void**) + 0x3a (0x2aab1173401a in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
53
+ frame #5: k2::NewRegion(std::shared_ptr<k2::Context>, unsigned long) + 0x112 (0x2aab11465b72 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
54
+ frame #6: k2::Array1<int>::Init(std::shared_ptr<k2::Context>, int, k2::Dtype) + 0x71 (0x2aab11432f51 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
55
+ frame #7: <unknown function> + 0x2472bd (0x2aab115c32bd in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
56
+ frame #8: k2::RaggedShapeFromTotSizes(std::shared_ptr<k2::Context>, int, int const*) + 0x213 (0x2aab115c3b83 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
57
+ frame #9: k2::IndexAxis0(k2::RaggedShape&, k2::Array1<int> const&, k2::Array1<int>*) + 0x32c (0x2aab115d77ec in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
58
+ frame #10: k2::Index(k2::RaggedShape&, int, k2::Array1<int> const&, k2::Array1<int>*) + 0x353 (0x2aab115dc943 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
59
+ frame #11: k2::Ragged<k2::Arc> k2::DeviceIntersector::FormatOutputTpl<k2::Hash::PackedAccessor>(k2::Array1<int>*, k2::Array1<int>*) + 0x407 (0x2aab11552327 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
60
+ frame #12: k2::IntersectDevice(k2::Ragged<k2::Arc>&, int, k2::Ragged<k2::Arc>&, int, k2::Array1<int> const&, k2::Array1<int>*, k2::Array1<int>*, bool) + 0x3a2 (0x2aab11545682 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
61
+ frame #13: <unknown function> + 0x8eb5a (0x2aab1032cb5a in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/_k2.cpython-38-x86_64-linux-gnu.so)
62
+ frame #14: <unknown function> + 0x3628c (0x2aab102d428c in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/_k2.cpython-38-x86_64-linux-gnu.so)
63
+ <omitting python frames>
64
+ frame #44: __libc_start_main + 0xf5 (0x2aaaab616555 in /lib64/libc.so.6)
65
+
66
+
67
+ 2022-06-27 18:55:06,115 INFO [decode.py:789] num_arcs before pruning: 1034414
68
+ 2022-06-27 18:55:06,115 INFO [decode.py:792] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
69
+ 2022-06-27 18:55:06,128 INFO [decode.py:803] num_arcs after pruning: 5251
70
+ 2022-06-27 18:55:09,967 INFO [decode.py:483] batch 1000/?, cuts processed until now is 4220
71
+ 2022-06-27 18:55:48,299 INFO [decode.py:483] batch 1100/?, cuts processed until now is 4631
72
+ 2022-06-27 18:56:25,270 INFO [decode.py:483] batch 1200/?, cuts processed until now is 5033
73
+ 2022-06-27 18:56:56,411 INFO [decode.py:783] Caught exception:
74
+ CUDA out of memory. Tried to allocate 1.81 GiB (GPU 0; 31.75 GiB total capacity; 27.76 GiB already allocated; 1.06 GiB free; 29.49 GiB reserved in total by PyTorch)
75
+ Exception raised from malloc at /pytorch/c10/cuda/CUDACachingAllocator.cpp:272 (most recent call first):
76
+ frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x42 (0x2aab0258d8b2 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10.so)
77
+ frame #1: <unknown function> + 0x2021b (0x2aab0232721b in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
78
+ frame #2: <unknown function> + 0x21034 (0x2aab02328034 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
79
+ frame #3: <unknown function> + 0x2167d (0x2aab0232867d in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
80
+ frame #4: k2::PytorchCudaContext::Allocate(unsigned long, void**) + 0x3a (0x2aab1173401a in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
81
+ frame #5: k2::NewRegion(std::shared_ptr<k2::Context>, unsigned long) + 0x112 (0x2aab11465b72 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
82
+ frame #6: k2::Array1<int>::Init(std::shared_ptr<k2::Context>, int, k2::Dtype) + 0x71 (0x2aab11432f51 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
83
+ frame #7: <unknown function> + 0x2472bd (0x2aab115c32bd in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
84
+ frame #8: k2::RaggedShapeFromTotSizes(std::shared_ptr<k2::Context>, int, int const*) + 0x213 (0x2aab115c3b83 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
85
+ frame #9: k2::IndexAxis0(k2::RaggedShape&, k2::Array1<int> const&, k2::Array1<int>*) + 0x32c (0x2aab115d77ec in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
86
+ frame #10: k2::Index(k2::RaggedShape&, int, k2::Array1<int> const&, k2::Array1<int>*) + 0x353 (0x2aab115dc943 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
87
+ frame #11: k2::Ragged<k2::Arc> k2::DeviceIntersector::FormatOutputTpl<k2::Hash::PackedAccessor>(k2::Array1<int>*, k2::Array1<int>*) + 0x407 (0x2aab11552327 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
88
+ frame #12: k2::IntersectDevice(k2::Ragged<k2::Arc>&, int, k2::Ragged<k2::Arc>&, int, k2::Array1<int> const&, k2::Array1<int>*, k2::Array1<int>*, bool) + 0x3a2 (0x2aab11545682 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
89
+ frame #13: <unknown function> + 0x8eb5a (0x2aab1032cb5a in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/_k2.cpython-38-x86_64-linux-gnu.so)
90
+ frame #14: <unknown function> + 0x3628c (0x2aab102d428c in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/_k2.cpython-38-x86_64-linux-gnu.so)
91
+ <omitting python frames>
92
+ frame #44: __libc_start_main + 0xf5 (0x2aaaab616555 in /lib64/libc.so.6)
93
+
94
+
95
+ 2022-06-27 18:56:56,411 INFO [decode.py:789] num_arcs before pruning: 1081951
96
+ 2022-06-27 18:56:56,411 INFO [decode.py:792] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
97
+ 2022-06-27 18:56:56,424 INFO [decode.py:803] num_arcs after pruning: 6154
98
+ 2022-06-27 18:57:04,430 INFO [decode.py:483] batch 1300/?, cuts processed until now is 5355
99
+ 2022-06-27 18:57:05,509 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.1.txt
100
+ 2022-06-27 18:57:05,613 INFO [utils.py:418] [test-lm_scale_0.1] %WER 15.01% [9667 / 64388, 371 ins, 3551 del, 5745 sub ]
101
+ 2022-06-27 18:57:06,082 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.1.txt
102
+ 2022-06-27 18:57:06,131 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.2.txt
103
+ 2022-06-27 18:57:06,226 INFO [utils.py:418] [test-lm_scale_0.2] %WER 15.29% [9842 / 64388, 323 ins, 3881 del, 5638 sub ]
104
+ 2022-06-27 18:57:06,450 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.2.txt
105
+ 2022-06-27 18:57:06,495 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.3.txt
106
+ 2022-06-27 18:57:06,590 INFO [utils.py:418] [test-lm_scale_0.3] %WER 15.70% [10106 / 64388, 262 ins, 4365 del, 5479 sub ]
107
+ 2022-06-27 18:57:06,814 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.3.txt
108
+ 2022-06-27 18:57:06,858 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.4.txt
109
+ 2022-06-27 18:57:06,951 INFO [utils.py:418] [test-lm_scale_0.4] %WER 16.37% [10541 / 64388, 219 ins, 5051 del, 5271 sub ]
110
+ 2022-06-27 18:57:07,175 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.4.txt
111
+ 2022-06-27 18:57:07,468 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.5.txt
112
+ 2022-06-27 18:57:07,703 INFO [utils.py:418] [test-lm_scale_0.5] %WER 17.47% [11250 / 64388, 176 ins, 6024 del, 5050 sub ]
113
+ 2022-06-27 18:57:07,927 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.5.txt
114
+ 2022-06-27 18:57:07,972 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.6.txt
115
+ 2022-06-27 18:57:08,066 INFO [utils.py:418] [test-lm_scale_0.6] %WER 19.07% [12282 / 64388, 152 ins, 7369 del, 4761 sub ]
116
+ 2022-06-27 18:57:08,290 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.6.txt
117
+ 2022-06-27 18:57:08,334 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.7.txt
118
+ 2022-06-27 18:57:08,426 INFO [utils.py:418] [test-lm_scale_0.7] %WER 21.32% [13725 / 64388, 136 ins, 9100 del, 4489 sub ]
119
+ 2022-06-27 18:57:08,649 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.7.txt
120
+ 2022-06-27 18:57:08,692 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.8.txt
121
+ 2022-06-27 18:57:08,783 INFO [utils.py:418] [test-lm_scale_0.8] %WER 24.17% [15565 / 64388, 105 ins, 11228 del, 4232 sub ]
122
+ 2022-06-27 18:57:09,153 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.8.txt
123
+ 2022-06-27 18:57:09,198 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_0.9.txt
124
+ 2022-06-27 18:57:09,290 INFO [utils.py:418] [test-lm_scale_0.9] %WER 27.59% [17766 / 64388, 91 ins, 13706 del, 3969 sub ]
125
+ 2022-06-27 18:57:09,516 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_0.9.txt
126
+ 2022-06-27 18:57:09,560 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.0.txt
127
+ 2022-06-27 18:57:09,651 INFO [utils.py:418] [test-lm_scale_1.0] %WER 31.58% [20336 / 64388, 80 ins, 16475 del, 3781 sub ]
128
+ 2022-06-27 18:57:09,879 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.0.txt
129
+ 2022-06-27 18:57:09,920 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.1.txt
130
+ 2022-06-27 18:57:10,011 INFO [utils.py:418] [test-lm_scale_1.1] %WER 35.69% [22983 / 64388, 65 ins, 19363 del, 3555 sub ]
131
+ 2022-06-27 18:57:10,240 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.1.txt
132
+ 2022-06-27 18:57:10,280 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.2.txt
133
+ 2022-06-27 18:57:10,510 INFO [utils.py:418] [test-lm_scale_1.2] %WER 39.91% [25700 / 64388, 54 ins, 22286 del, 3360 sub ]
134
+ 2022-06-27 18:57:10,744 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.2.txt
135
+ 2022-06-27 18:57:10,790 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.3.txt
136
+ 2022-06-27 18:57:10,880 INFO [utils.py:418] [test-lm_scale_1.3] %WER 44.32% [28534 / 64388, 47 ins, 25350 del, 3137 sub ]
137
+ 2022-06-27 18:57:11,109 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.3.txt
138
+ 2022-06-27 18:57:11,149 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.4.txt
139
+ 2022-06-27 18:57:11,236 INFO [utils.py:418] [test-lm_scale_1.4] %WER 48.37% [31142 / 64388, 40 ins, 28142 del, 2960 sub ]
140
+ 2022-06-27 18:57:11,468 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.4.txt
141
+ 2022-06-27 18:57:11,505 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.5.txt
142
+ 2022-06-27 18:57:11,592 INFO [utils.py:418] [test-lm_scale_1.5] %WER 52.01% [33491 / 64388, 38 ins, 30627 del, 2826 sub ]
143
+ 2022-06-27 18:57:11,825 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.5.txt
144
+ 2022-06-27 18:57:11,863 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.6.txt
145
+ 2022-06-27 18:57:12,093 INFO [utils.py:418] [test-lm_scale_1.6] %WER 55.07% [35457 / 64388, 35 ins, 32740 del, 2682 sub ]
146
+ 2022-06-27 18:57:12,325 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.6.txt
147
+ 2022-06-27 18:57:12,363 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.7.txt
148
+ 2022-06-27 18:57:12,450 INFO [utils.py:418] [test-lm_scale_1.7] %WER 57.74% [37177 / 64388, 33 ins, 34549 del, 2595 sub ]
149
+ 2022-06-27 18:57:12,683 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.7.txt
150
+ 2022-06-27 18:57:12,720 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.8.txt
151
+ 2022-06-27 18:57:12,807 INFO [utils.py:418] [test-lm_scale_1.8] %WER 59.81% [38508 / 64388, 28 ins, 35972 del, 2508 sub ]
152
+ 2022-06-27 18:57:13,040 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.8.txt
153
+ 2022-06-27 18:57:13,077 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_1.9.txt
154
+ 2022-06-27 18:57:13,162 INFO [utils.py:418] [test-lm_scale_1.9] %WER 61.70% [39730 / 64388, 22 ins, 37267 del, 2441 sub ]
155
+ 2022-06-27 18:57:13,539 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_1.9.txt
156
+ 2022-06-27 18:57:13,577 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-test-lm_scale_2.0.txt
157
+ 2022-06-27 18:57:13,663 INFO [utils.py:418] [test-lm_scale_2.0] %WER 63.24% [40719 / 64388, 20 ins, 38322 del, 2377 sub ]
158
+ 2022-06-27 18:57:13,897 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-test-lm_scale_2.0.txt
159
+ 2022-06-27 18:57:13,904 INFO [decode.py:532]
160
+ For test, WER of different settings are:
161
+ lm_scale_0.1 15.01 best for test
162
+ lm_scale_0.2 15.29
163
+ lm_scale_0.3 15.7
164
+ lm_scale_0.4 16.37
165
+ lm_scale_0.5 17.47
166
+ lm_scale_0.6 19.07
167
+ lm_scale_0.7 21.32
168
+ lm_scale_0.8 24.17
169
+ lm_scale_0.9 27.59
170
+ lm_scale_1.0 31.58
171
+ lm_scale_1.1 35.69
172
+ lm_scale_1.2 39.91
173
+ lm_scale_1.3 44.32
174
+ lm_scale_1.4 48.37
175
+ lm_scale_1.5 52.01
176
+ lm_scale_1.6 55.07
177
+ lm_scale_1.7 57.74
178
+ lm_scale_1.8 59.81
179
+ lm_scale_1.9 61.7
180
+ lm_scale_2.0 63.24
181
+
182
+ 2022-06-27 18:57:15,579 INFO [decode.py:483] batch 0/?, cuts processed until now is 4
183
+ 2022-06-27 18:57:52,780 INFO [decode.py:483] batch 100/?, cuts processed until now is 428
184
+ 2022-06-27 18:58:29,361 INFO [decode.py:483] batch 200/?, cuts processed until now is 888
185
+ 2022-06-27 18:59:07,002 INFO [decode.py:483] batch 300/?, cuts processed until now is 1363
186
+ 2022-06-27 18:59:42,509 INFO [decode.py:483] batch 400/?, cuts processed until now is 1815
187
+ 2022-06-27 19:00:20,896 INFO [decode.py:483] batch 500/?, cuts processed until now is 2243
188
+ 2022-06-27 19:00:55,648 INFO [decode.py:483] batch 600/?, cuts processed until now is 2717
189
+ 2022-06-27 19:01:30,559 INFO [decode.py:483] batch 700/?, cuts processed until now is 3192
190
+ 2022-06-27 19:02:03,949 INFO [decode.py:783] Caught exception:
191
+ CUDA out of memory. Tried to allocate 8.00 GiB (GPU 0; 31.75 GiB total capacity; 16.89 GiB already allocated; 7.38 GiB free; 23.17 GiB reserved in total by PyTorch)
192
+ Exception raised from malloc at /pytorch/c10/cuda/CUDACachingAllocator.cpp:272 (most recent call first):
193
+ frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x42 (0x2aab0258d8b2 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10.so)
194
+ frame #1: <unknown function> + 0x2021b (0x2aab0232721b in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
195
+ frame #2: <unknown function> + 0x21034 (0x2aab02328034 in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
196
+ frame #3: <unknown function> + 0x2167d (0x2aab0232867d in /home/local/QCRI/ahussein/anaconda3/envs/k2/lib/python3.8/site-packages/torch/lib/libc10_cuda.so)
197
+ frame #4: k2::PytorchCudaContext::Allocate(unsigned long, void**) + 0x3a (0x2aab1173401a in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
198
+ frame #5: k2::NewRegion(std::shared_ptr<k2::Context>, unsigned long) + 0x112 (0x2aab11465b72 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
199
+ frame #6: k2::Hash::Hash(std::shared_ptr<k2::Context>, int, int, int) + 0x2f7 (0x2aab11538957 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
200
+ frame #7: k2::Hash::Resize(int, int, int, bool) + 0x1b4 (0x2aab1152e464 in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
201
+ frame #8: k2::DeviceIntersector::ForwardSortedA() + 0x53e (0x2aab1156355e in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
202
+ frame #9: k2::IntersectDevice(k2::Ragged<k2::Arc>&, int, k2::Ragged<k2::Arc>&, int, k2::Array1<int> const&, k2::Array1<int>*, k2::Array1<int>*, bool) + 0x4cd (0x2aab115457ad in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/libk2context.so)
203
+ frame #10: <unknown function> + 0x8eb5a (0x2aab1032cb5a in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/_k2.cpython-38-x86_64-linux-gnu.so)
204
+ frame #11: <unknown function> + 0x3628c (0x2aab102d428c in /alt-arabic/speech/amir/k2/tmp/k2/build/lib/_k2.cpython-38-x86_64-linux-gnu.so)
205
+ <omitting python frames>
206
+ frame #41: __libc_start_main + 0xf5 (0x2aaaab616555 in /lib64/libc.so.6)
207
+
208
+
209
+ 2022-06-27 19:02:03,950 INFO [decode.py:789] num_arcs before pruning: 1001553
210
+ 2022-06-27 19:02:03,950 INFO [decode.py:792] This OOM is not an error. You can ignore it. If your model does not converge well, or --max-duration is too large, or the input sound file is difficult to decode, you will meet this exception.
211
+ 2022-06-27 19:02:03,962 INFO [decode.py:803] num_arcs after pruning: 5815
212
+ 2022-06-27 19:02:07,376 INFO [decode.py:483] batch 800/?, cuts processed until now is 3635
213
+ 2022-06-27 19:02:44,139 INFO [decode.py:483] batch 900/?, cuts processed until now is 4082
214
+ 2022-06-27 19:03:28,596 INFO [decode.py:483] batch 1000/?, cuts processed until now is 4500
215
+ 2022-06-27 19:04:15,771 INFO [decode.py:483] batch 1100/?, cuts processed until now is 4869
216
+ 2022-06-27 19:04:33,221 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.1.txt
217
+ 2022-06-27 19:04:33,318 INFO [utils.py:418] [dev-lm_scale_0.1] %WER 15.62% [9398 / 60169, 338 ins, 3783 del, 5277 sub ]
218
+ 2022-06-27 19:04:33,545 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.1.txt
219
+ 2022-06-27 19:04:33,593 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.2.txt
220
+ 2022-06-27 19:04:33,688 INFO [utils.py:418] [dev-lm_scale_0.2] %WER 15.85% [9536 / 60169, 295 ins, 4101 del, 5140 sub ]
221
+ 2022-06-27 19:04:33,920 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.2.txt
222
+ 2022-06-27 19:04:33,966 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.3.txt
223
+ 2022-06-27 19:04:34,061 INFO [utils.py:418] [dev-lm_scale_0.3] %WER 16.38% [9856 / 60169, 249 ins, 4624 del, 4983 sub ]
224
+ 2022-06-27 19:04:34,291 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.3.txt
225
+ 2022-06-27 19:04:34,335 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.4.txt
226
+ 2022-06-27 19:04:34,426 INFO [utils.py:418] [dev-lm_scale_0.4] %WER 17.22% [10364 / 60169, 219 ins, 5377 del, 4768 sub ]
227
+ 2022-06-27 19:04:34,650 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.4.txt
228
+ 2022-06-27 19:04:34,695 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.5.txt
229
+ 2022-06-27 19:04:34,935 INFO [utils.py:418] [dev-lm_scale_0.5] %WER 18.58% [11178 / 60169, 176 ins, 6457 del, 4545 sub ]
230
+ 2022-06-27 19:04:35,152 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.5.txt
231
+ 2022-06-27 19:04:35,193 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.6.txt
232
+ 2022-06-27 19:04:35,283 INFO [utils.py:418] [dev-lm_scale_0.6] %WER 20.50% [12337 / 60169, 142 ins, 7911 del, 4284 sub ]
233
+ 2022-06-27 19:04:35,500 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.6.txt
234
+ 2022-06-27 19:04:35,542 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.7.txt
235
+ 2022-06-27 19:04:35,632 INFO [utils.py:418] [dev-lm_scale_0.7] %WER 22.99% [13830 / 60169, 113 ins, 9743 del, 3974 sub ]
236
+ 2022-06-27 19:04:35,863 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.7.txt
237
+ 2022-06-27 19:04:35,902 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.8.txt
238
+ 2022-06-27 19:04:35,993 INFO [utils.py:418] [dev-lm_scale_0.8] %WER 25.88% [15571 / 60169, 91 ins, 11766 del, 3714 sub ]
239
+ 2022-06-27 19:04:36,344 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.8.txt
240
+ 2022-06-27 19:04:36,387 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_0.9.txt
241
+ 2022-06-27 19:04:36,477 INFO [utils.py:418] [dev-lm_scale_0.9] %WER 29.42% [17700 / 60169, 78 ins, 14199 del, 3423 sub ]
242
+ 2022-06-27 19:04:36,711 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_0.9.txt
243
+ 2022-06-27 19:04:36,752 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.0.txt
244
+ 2022-06-27 19:04:36,842 INFO [utils.py:418] [dev-lm_scale_1.0] %WER 33.12% [19926 / 60169, 69 ins, 16663 del, 3194 sub ]
245
+ 2022-06-27 19:04:37,077 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.0.txt
246
+ 2022-06-27 19:04:37,117 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.1.txt
247
+ 2022-06-27 19:04:37,207 INFO [utils.py:418] [dev-lm_scale_1.1] %WER 37.31% [22452 / 60169, 62 ins, 19379 del, 3011 sub ]
248
+ 2022-06-27 19:04:37,443 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.1.txt
249
+ 2022-06-27 19:04:37,482 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.2.txt
250
+ 2022-06-27 19:04:37,692 INFO [utils.py:418] [dev-lm_scale_1.2] %WER 41.59% [25025 / 60169, 51 ins, 22214 del, 2760 sub ]
251
+ 2022-06-27 19:04:37,932 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.2.txt
252
+ 2022-06-27 19:04:37,971 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.3.txt
253
+ 2022-06-27 19:04:38,057 INFO [utils.py:418] [dev-lm_scale_1.3] %WER 45.83% [27574 / 60169, 45 ins, 24966 del, 2563 sub ]
254
+ 2022-06-27 19:04:38,282 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.3.txt
255
+ 2022-06-27 19:04:38,351 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.4.txt
256
+ 2022-06-27 19:04:38,443 INFO [utils.py:418] [dev-lm_scale_1.4] %WER 49.77% [29947 / 60169, 41 ins, 27519 del, 2387 sub ]
257
+ 2022-06-27 19:04:38,682 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.4.txt
258
+ 2022-06-27 19:04:38,762 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.5.txt
259
+ 2022-06-27 19:04:38,844 INFO [utils.py:418] [dev-lm_scale_1.5] %WER 53.28% [32057 / 60169, 34 ins, 29848 del, 2175 sub ]
260
+ 2022-06-27 19:04:39,063 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.5.txt
261
+ 2022-06-27 19:04:39,099 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.6.txt
262
+ 2022-06-27 19:04:39,299 INFO [utils.py:418] [dev-lm_scale_1.6] %WER 56.26% [33853 / 60169, 33 ins, 31754 del, 2066 sub ]
263
+ 2022-06-27 19:04:39,517 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.6.txt
264
+ 2022-06-27 19:04:39,551 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.7.txt
265
+ 2022-06-27 19:04:39,631 INFO [utils.py:418] [dev-lm_scale_1.7] %WER 58.66% [35296 / 60169, 30 ins, 33295 del, 1971 sub ]
266
+ 2022-06-27 19:04:39,851 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.7.txt
267
+ 2022-06-27 19:04:39,886 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.8.txt
268
+ 2022-06-27 19:04:39,966 INFO [utils.py:418] [dev-lm_scale_1.8] %WER 60.66% [36496 / 60169, 31 ins, 34535 del, 1930 sub ]
269
+ 2022-06-27 19:04:40,187 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.8.txt
270
+ 2022-06-27 19:04:40,222 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_1.9.txt
271
+ 2022-06-27 19:04:40,305 INFO [utils.py:418] [dev-lm_scale_1.9] %WER 62.48% [37591 / 60169, 28 ins, 35674 del, 1889 sub ]
272
+ 2022-06-27 19:04:40,664 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_1.9.txt
273
+ 2022-06-27 19:04:40,700 INFO [decode.py:504] The transcripts are stored in conformer_ctc/exp_5000_att0.8/recogs-dev-lm_scale_2.0.txt
274
+ 2022-06-27 19:04:40,786 INFO [utils.py:418] [dev-lm_scale_2.0] %WER 63.74% [38350 / 60169, 27 ins, 36440 del, 1883 sub ]
275
+ 2022-06-27 19:04:41,020 INFO [decode.py:516] Wrote detailed error stats to conformer_ctc/exp_5000_att0.8/errs-dev-lm_scale_2.0.txt
276
+ 2022-06-27 19:04:41,025 INFO [decode.py:532]
277
+ For dev, WER of different settings are:
278
+ lm_scale_0.1 15.62 best for dev
279
+ lm_scale_0.2 15.85
280
+ lm_scale_0.3 16.38
281
+ lm_scale_0.4 17.22
282
+ lm_scale_0.5 18.58
283
+ lm_scale_0.6 20.5
284
+ lm_scale_0.7 22.99
285
+ lm_scale_0.8 25.88
286
+ lm_scale_0.9 29.42
287
+ lm_scale_1.0 33.12
288
+ lm_scale_1.1 37.31
289
+ lm_scale_1.2 41.59
290
+ lm_scale_1.3 45.83
291
+ lm_scale_1.4 49.77
292
+ lm_scale_1.5 53.28
293
+ lm_scale_1.6 56.26
294
+ lm_scale_1.7 58.66
295
+ lm_scale_1.8 60.66
296
+ lm_scale_1.9 62.48
297
+ lm_scale_2.0 63.74
298
+
299
+ 2022-06-27 19:04:41,025 INFO [decode.py:695] Done!