Dongwei Jiang commited on
Commit
48c82e5
1 Parent(s): 8a95a36

Add model files

Browse files
Files changed (29) hide show
  1. RESULTS.md +32 -0
  2. asr_train_raw_en_bpe500_sp/RESULTS.md +44 -0
  3. asr_train_raw_en_bpe500_sp/config.yaml +704 -0
  4. asr_train_raw_en_bpe500_sp/images/acc.png +0 -0
  5. asr_train_raw_en_bpe500_sp/images/backward_time.png +0 -0
  6. asr_train_raw_en_bpe500_sp/images/cer.png +0 -0
  7. asr_train_raw_en_bpe500_sp/images/cer_ctc.png +0 -0
  8. asr_train_raw_en_bpe500_sp/images/forward_time.png +0 -0
  9. asr_train_raw_en_bpe500_sp/images/gpu_max_cached_mem_GB.png +0 -0
  10. asr_train_raw_en_bpe500_sp/images/iter_time.png +0 -0
  11. asr_train_raw_en_bpe500_sp/images/loss.png +0 -0
  12. asr_train_raw_en_bpe500_sp/images/loss_att.png +0 -0
  13. asr_train_raw_en_bpe500_sp/images/loss_ctc.png +0 -0
  14. asr_train_raw_en_bpe500_sp/images/optim0_lr0.png +0 -0
  15. asr_train_raw_en_bpe500_sp/images/optim_step_time.png +0 -0
  16. asr_train_raw_en_bpe500_sp/images/train_time.png +0 -0
  17. asr_train_raw_en_bpe500_sp/images/wer.png +0 -0
  18. asr_train_raw_en_bpe500_sp/valid.acc.ave_10best.pth +3 -0
  19. lm_train_lm_en_bpe500/config.yaml +627 -0
  20. lm_train_lm_en_bpe500/images/backward_time.png +0 -0
  21. lm_train_lm_en_bpe500/images/forward_time.png +0 -0
  22. lm_train_lm_en_bpe500/images/gpu_max_cached_mem_GB.png +0 -0
  23. lm_train_lm_en_bpe500/images/iter_time.png +0 -0
  24. lm_train_lm_en_bpe500/images/loss.png +0 -0
  25. lm_train_lm_en_bpe500/images/optim0_lr0.png +0 -0
  26. lm_train_lm_en_bpe500/images/optim_step_time.png +0 -0
  27. lm_train_lm_en_bpe500/images/train_time.png +0 -0
  28. lm_train_lm_en_bpe500/perplexity_test/ppl +1 -0
  29. lm_train_lm_en_bpe500/valid.loss.ave_10best.pth +3 -0
RESULTS.md ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <!-- Generated by scripts/utils/show_asr_result.sh -->
2
+ # RESULTS
3
+ ## Environments
4
+ - date: `Mon Mar 27 04:02:03 EDT 2023`
5
+ - python version: `3.8.16 (default, Mar 2 2023, 03:21:46) [GCC 11.2.0]`
6
+ - espnet version: `espnet 202301`
7
+ - pytorch version: `pytorch 1.8.1`
8
+ - Git hash: `ff841366229d539eb74d23ac999cae7c0cc62cad`
9
+ - Commit date: `Mon Feb 20 12:23:15 2023 -0500`
10
+
11
+ ## exp/asr_train_raw_en_bpe500_sp
12
+ ### WER
13
+
14
+ |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
15
+ |---|---|---|---|---|---|---|---|---|
16
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave/dev|466|14671|94.0|2.7|3.3|0.7|6.6|65.9|
17
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave/test|1155|27500|93.9|2.7|3.4|0.7|6.8|61.1|
18
+
19
+ ### CER
20
+
21
+ |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
22
+ |---|---|---|---|---|---|---|---|---|
23
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave/dev|466|78259|96.6|0.6|2.8|0.6|4.0|65.9|
24
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave/test|1155|145066|96.6|0.6|2.8|0.6|4.1|61.1|
25
+
26
+ ### TER
27
+
28
+ |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
29
+ |---|---|---|---|---|---|---|---|---|
30
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave/dev|466|29364|95.5|1.9|2.7|0.5|5.1|65.9|
31
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave/test|1155|54206|95.5|1.7|2.7|0.6|5.1|61.1|
32
+
asr_train_raw_en_bpe500_sp/RESULTS.md ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <!-- Generated by scripts/utils/show_asr_result.sh -->
2
+ # RESULTS
3
+ ## Environments
4
+ - date: `Mon Mar 27 04:02:03 EDT 2023`
5
+ - python version: `3.8.16 (default, Mar 2 2023, 03:21:46) [GCC 11.2.0]`
6
+ - espnet version: `espnet 202301`
7
+ - pytorch version: `pytorch 1.8.1`
8
+ - Git hash: `ff841366229d539eb74d23ac999cae7c0cc62cad`
9
+ - Commit date: `Mon Feb 20 12:23:15 2023 -0500`
10
+
11
+ ## exp/asr_train_raw_en_bpe500_sp
12
+ ### WER
13
+
14
+ |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
15
+ |---|---|---|---|---|---|---|---|---|
16
+ |decode_asr_model_valid.acc.ave/dev|466|14671|94.4|3.6|1.9|2.0|7.6|78.1|
17
+ |decode_asr_model_valid.acc.ave/test|1155|27500|93.9|3.9|2.1|1.9|7.9|72.6|
18
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave.bk_Mar_27_check_lm/dev|466|14671|94.0|2.7|3.3|0.7|6.6|65.9|
19
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave.bk_Mar_27_check_lm/test|1155|27500|93.9|2.7|3.4|0.7|6.8|61.1|
20
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave/dev|466|14671|94.6|3.4|2.1|1.3|6.7|70.6|
21
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave/test|1155|27500|94.2|3.5|2.3|1.2|7.0|66.4|
22
+
23
+ ### CER
24
+
25
+ |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
26
+ |---|---|---|---|---|---|---|---|---|
27
+ |decode_asr_model_valid.acc.ave/dev|466|78259|97.5|0.8|1.7|2.1|4.6|78.1|
28
+ |decode_asr_model_valid.acc.ave/test|1155|145066|97.2|1.0|1.8|2.1|4.9|72.6|
29
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave.bk_Mar_27_check_lm/dev|466|78259|96.6|0.6|2.8|0.6|4.0|65.9|
30
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave.bk_Mar_27_check_lm/test|1155|145066|96.6|0.6|2.8|0.6|4.1|61.1|
31
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave/dev|466|78259|97.4|0.8|1.8|1.3|3.9|70.6|
32
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave/test|1155|145066|97.2|0.8|1.9|1.3|4.1|66.4|
33
+
34
+ ### TER
35
+
36
+ |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
37
+ |---|---|---|---|---|---|---|---|---|
38
+ |decode_asr_model_valid.acc.ave/dev|466|29364|95.6|2.5|1.9|1.9|6.3|78.1|
39
+ |decode_asr_model_valid.acc.ave/test|1155|54206|95.5|2.5|2.0|2.1|6.5|72.6|
40
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave.bk_Mar_27_check_lm/dev|466|29364|95.5|1.9|2.7|0.5|5.1|65.9|
41
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave.bk_Mar_27_check_lm/test|1155|54206|95.5|1.7|2.7|0.6|5.1|61.1|
42
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave/dev|466|29364|95.7|2.3|2.0|1.2|5.4|70.6|
43
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave/test|1155|54206|95.7|2.2|2.1|1.3|5.6|66.4|
44
+
asr_train_raw_en_bpe500_sp/config.yaml ADDED
@@ -0,0 +1,704 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ config: conf/train.yaml
2
+ print_config: false
3
+ log_level: INFO
4
+ dry_run: false
5
+ iterator_type: sequence
6
+ output_dir: exp/asr_train_raw_en_bpe500_sp
7
+ ngpu: 1
8
+ seed: 2022
9
+ num_workers: 6
10
+ num_att_plot: 3
11
+ dist_backend: nccl
12
+ dist_init_method: env://
13
+ dist_world_size: 4
14
+ dist_rank: 0
15
+ local_rank: 0
16
+ dist_master_addr: localhost
17
+ dist_master_port: 46711
18
+ dist_launcher: null
19
+ multiprocessing_distributed: true
20
+ unused_parameters: false
21
+ sharded_ddp: false
22
+ cudnn_enabled: true
23
+ cudnn_benchmark: false
24
+ cudnn_deterministic: true
25
+ collect_stats: false
26
+ write_collected_feats: false
27
+ max_epoch: 50
28
+ patience: null
29
+ val_scheduler_criterion:
30
+ - valid
31
+ - loss
32
+ early_stopping_criterion:
33
+ - valid
34
+ - loss
35
+ - min
36
+ best_model_criterion:
37
+ - - valid
38
+ - acc
39
+ - max
40
+ keep_nbest_models: 10
41
+ nbest_averaging_interval: 0
42
+ grad_clip: 5.0
43
+ grad_clip_type: 2.0
44
+ grad_noise: false
45
+ accum_grad: 1
46
+ no_forward_run: false
47
+ resume: true
48
+ train_dtype: float32
49
+ use_amp: true
50
+ log_interval: null
51
+ use_matplotlib: true
52
+ use_tensorboard: true
53
+ create_graph_in_tensorboard: false
54
+ use_wandb: false
55
+ wandb_project: null
56
+ wandb_id: null
57
+ wandb_entity: null
58
+ wandb_name: null
59
+ wandb_model_log_interval: -1
60
+ detect_anomaly: false
61
+ pretrain_path: null
62
+ init_param: []
63
+ ignore_init_mismatch: false
64
+ freeze_param: []
65
+ num_iters_per_epoch: null
66
+ batch_size: 20
67
+ valid_batch_size: null
68
+ batch_bins: 50000000
69
+ valid_batch_bins: null
70
+ train_shape_file:
71
+ - exp/asr_stats_raw_en_bpe500_sp/train/speech_shape
72
+ - exp/asr_stats_raw_en_bpe500_sp/train/text_shape.bpe
73
+ valid_shape_file:
74
+ - exp/asr_stats_raw_en_bpe500_sp/valid/speech_shape
75
+ - exp/asr_stats_raw_en_bpe500_sp/valid/text_shape.bpe
76
+ batch_type: numel
77
+ valid_batch_type: null
78
+ fold_length:
79
+ - 80000
80
+ - 150
81
+ sort_in_batch: descending
82
+ sort_batch: descending
83
+ multiple_iterator: false
84
+ chunk_length: 500
85
+ chunk_shift_ratio: 0.5
86
+ num_cache_chunks: 1024
87
+ train_data_path_and_name_and_type:
88
+ - - dump/raw/train_sp/wav.scp
89
+ - speech
90
+ - kaldi_ark
91
+ - - dump/raw/train_sp/text
92
+ - text
93
+ - text
94
+ valid_data_path_and_name_and_type:
95
+ - - dump/raw/dev/wav.scp
96
+ - speech
97
+ - kaldi_ark
98
+ - - dump/raw/dev/text
99
+ - text
100
+ - text
101
+ allow_variable_data_keys: false
102
+ max_cache_size: 0.0
103
+ max_cache_fd: 32
104
+ valid_max_cache_size: null
105
+ exclude_weight_decay: false
106
+ exclude_weight_decay_conf: {}
107
+ optim: adam
108
+ optim_conf:
109
+ lr: 0.002
110
+ weight_decay: 1.0e-06
111
+ scheduler: warmuplr
112
+ scheduler_conf:
113
+ warmup_steps: 15000
114
+ token_list:
115
+ - <blank>
116
+ - <unk>
117
+ - '[unk]'
118
+ - ▁
119
+ - s
120
+ - ▁the
121
+ - t
122
+ - ▁and
123
+ - e
124
+ - ▁a
125
+ - ▁to
126
+ - d
127
+ - ▁of
128
+ - ''''
129
+ - n
130
+ - ing
131
+ - ▁in
132
+ - ▁that
133
+ - re
134
+ - ▁i
135
+ - c
136
+ - o
137
+ - u
138
+ - ▁we
139
+ - y
140
+ - a
141
+ - ed
142
+ - ▁it
143
+ - ▁you
144
+ - i
145
+ - m
146
+ - ▁is
147
+ - er
148
+ - p
149
+ - g
150
+ - w
151
+ - al
152
+ - ▁this
153
+ - ▁so
154
+ - f
155
+ - le
156
+ - b
157
+ - ar
158
+ - ▁f
159
+ - k
160
+ - ▁c
161
+ - r
162
+ - in
163
+ - or
164
+ - ▁for
165
+ - ▁be
166
+ - ve
167
+ - ▁was
168
+ - te
169
+ - th
170
+ - ▁do
171
+ - es
172
+ - ly
173
+ - ▁they
174
+ - ro
175
+ - ▁are
176
+ - ▁with
177
+ - ▁have
178
+ - an
179
+ - v
180
+ - ch
181
+ - ▁on
182
+ - se
183
+ - lo
184
+ - ▁but
185
+ - en
186
+ - ri
187
+ - li
188
+ - ▁what
189
+ - it
190
+ - ic
191
+ - ▁can
192
+ - l
193
+ - ur
194
+ - ce
195
+ - ent
196
+ - ▁me
197
+ - ▁b
198
+ - ▁ma
199
+ - ▁he
200
+ - ra
201
+ - ▁de
202
+ - ll
203
+ - at
204
+ - ▁about
205
+ - ▁one
206
+ - ▁not
207
+ - ne
208
+ - ▁all
209
+ - ▁my
210
+ - ter
211
+ - el
212
+ - il
213
+ - ▁there
214
+ - 'on'
215
+ - ad
216
+ - ▁mo
217
+ - ol
218
+ - ation
219
+ - nd
220
+ - ▁like
221
+ - ▁people
222
+ - po
223
+ - ▁at
224
+ - ▁us
225
+ - us
226
+ - ▁g
227
+ - ci
228
+ - ▁our
229
+ - h
230
+ - pe
231
+ - ▁as
232
+ - ▁from
233
+ - vi
234
+ - ▁if
235
+ - as
236
+ - ▁ex
237
+ - ▁con
238
+ - ▁an
239
+ - ver
240
+ - ▁out
241
+ - ▁just
242
+ - un
243
+ - ▁see
244
+ - la
245
+ - ▁di
246
+ - ▁when
247
+ - ▁now
248
+ - ▁p
249
+ - ha
250
+ - ▁who
251
+ - ck
252
+ - ▁these
253
+ - ▁because
254
+ - ▁or
255
+ - ▁know
256
+ - ion
257
+ - ir
258
+ - ▁co
259
+ - ▁up
260
+ - ▁pa
261
+ - ment
262
+ - ▁think
263
+ - ge
264
+ - ▁how
265
+ - ide
266
+ - ▁by
267
+ - ul
268
+ - ity
269
+ - ▁go
270
+ - ▁get
271
+ - ▁ho
272
+ - ive
273
+ - ▁very
274
+ - ate
275
+ - ng
276
+ - ▁no
277
+ - ▁had
278
+ - ac
279
+ - ▁bo
280
+ - ry
281
+ - ▁more
282
+ - ▁them
283
+ - ▁some
284
+ - mi
285
+ - ▁time
286
+ - ▁your
287
+ - me
288
+ - ▁going
289
+ - op
290
+ - am
291
+ - per
292
+ - et
293
+ - ▁would
294
+ - ru
295
+ - ure
296
+ - ti
297
+ - ist
298
+ - ▁their
299
+ - x
300
+ - ▁were
301
+ - ▁look
302
+ - ▁pro
303
+ - ▁which
304
+ - ▁work
305
+ - tion
306
+ - est
307
+ - ty
308
+ - im
309
+ - z
310
+ - ta
311
+ - ▁want
312
+ - ▁two
313
+ - age
314
+ - ▁really
315
+ - om
316
+ - ma
317
+ - ers
318
+ - ting
319
+ - ▁world
320
+ - co
321
+ - ▁way
322
+ - ▁don
323
+ - wa
324
+ - hi
325
+ - tra
326
+ - ▁la
327
+ - ▁here
328
+ - able
329
+ - lu
330
+ - ▁other
331
+ - mo
332
+ - ies
333
+ - ▁has
334
+ - ▁could
335
+ - j
336
+ - ▁make
337
+ - ally
338
+ - ▁sta
339
+ - ten
340
+ - ▁will
341
+ - ▁un
342
+ - ig
343
+ - ▁where
344
+ - ▁into
345
+ - ke
346
+ - ▁than
347
+ - ▁comp
348
+ - ▁actually
349
+ - tic
350
+ - sh
351
+ - ▁did
352
+ - tor
353
+ - fa
354
+ - ical
355
+ - ▁she
356
+ - ▁years
357
+ - ▁say
358
+ - one
359
+ - ted
360
+ - ▁things
361
+ - ph
362
+ - ▁new
363
+ - ▁pre
364
+ - ▁any
365
+ - ▁thousand
366
+ - ▁been
367
+ - ▁inter
368
+ - ▁his
369
+ - ▁com
370
+ - ▁need
371
+ - nce
372
+ - ▁right
373
+ - ▁take
374
+ - ▁even
375
+ - ▁over
376
+ - ▁start
377
+ - ▁hundred
378
+ - min
379
+ - ▁sp
380
+ - ▁those
381
+ - ▁car
382
+ - ▁then
383
+ - mp
384
+ - ap
385
+ - ▁first
386
+ - les
387
+ - ize
388
+ - ▁every
389
+ - ba
390
+ - ▁something
391
+ - ▁well
392
+ - ard
393
+ - ▁str
394
+ - ▁back
395
+ - und
396
+ - ia
397
+ - pl
398
+ - ki
399
+ - ho
400
+ - ▁call
401
+ - ▁most
402
+ - ▁also
403
+ - bi
404
+ - ▁thing
405
+ - ▁life
406
+ - um
407
+ - ▁said
408
+ - ▁kind
409
+ - ▁lot
410
+ - ▁much
411
+ - va
412
+ - ▁ra
413
+ - ▁little
414
+ - ▁dr
415
+ - ▁got
416
+ - ▁come
417
+ - ful
418
+ - ▁talk
419
+ - ▁part
420
+ - ▁day
421
+ - ant
422
+ - ction
423
+ - ▁happen
424
+ - ▁only
425
+ - ▁many
426
+ - ▁wo
427
+ - pri
428
+ - ▁her
429
+ - ▁br
430
+ - qui
431
+ - ▁mean
432
+ - ▁three
433
+ - iv
434
+ - ▁different
435
+ - ugh
436
+ - ain
437
+ - ▁human
438
+ - ance
439
+ - ▁change
440
+ - ▁let
441
+ - ▁real
442
+ - ▁show
443
+ - ▁good
444
+ - ▁around
445
+ - ▁through
446
+ - ▁jo
447
+ - bu
448
+ - ▁down
449
+ - ight
450
+ - ga
451
+ - ▁why
452
+ - ▁live
453
+ - ff
454
+ - ▁tell
455
+ - ▁put
456
+ - ▁idea
457
+ - port
458
+ - ▁same
459
+ - ▁give
460
+ - ated
461
+ - ish
462
+ - ible
463
+ - ▁though
464
+ - ious
465
+ - ▁problem
466
+ - ▁five
467
+ - par
468
+ - ▁fact
469
+ - ▁cha
470
+ - ition
471
+ - ▁year
472
+ - ▁big
473
+ - ▁plan
474
+ - ▁great
475
+ - ▁find
476
+ - ▁four
477
+ - ▁app
478
+ - ▁after
479
+ - ▁system
480
+ - ▁place
481
+ - ▁em
482
+ - ▁build
483
+ - ▁percent
484
+ - ▁again
485
+ - ▁point
486
+ - ▁learn
487
+ - ▁own
488
+ - ▁long
489
+ - ▁made
490
+ - ▁today
491
+ - ▁nine
492
+ - ities
493
+ - ▁gene
494
+ - ▁six
495
+ - ▁question
496
+ - light
497
+ - ▁should
498
+ - ▁came
499
+ - ▁feel
500
+ - ▁turn
501
+ - ▁person
502
+ - ▁end
503
+ - ▁hu
504
+ - ▁design
505
+ - ▁help
506
+ - ▁brain
507
+ - ▁last
508
+ - ▁create
509
+ - ▁important
510
+ - ▁before
511
+ - ▁high
512
+ - ▁never
513
+ - ▁trans
514
+ - ▁another
515
+ - ▁him
516
+ - ▁eight
517
+ - ▁might
518
+ - ▁understand
519
+ - ▁power
520
+ - ▁better
521
+ - q
522
+ - ▁found
523
+ - ▁play
524
+ - ▁twenty
525
+ - ▁still
526
+ - ▁school
527
+ - ▁each
528
+ - ▁seven
529
+ - ▁together
530
+ - ▁few
531
+ - ▁hand
532
+ - ▁example
533
+ - que
534
+ - ▁next
535
+ - ▁million
536
+ - ▁story
537
+ - ▁women
538
+ - ▁under
539
+ - ▁number
540
+ - ▁course
541
+ - ▁water
542
+ - ▁ago
543
+ - ▁grow
544
+ - ▁between
545
+ - ▁develop
546
+ - ▁america
547
+ - ▁sort
548
+ - ▁technology
549
+ - ▁believe
550
+ - ▁second
551
+ - ▁small
552
+ - ▁maybe
553
+ - ▁become
554
+ - press
555
+ - ▁health
556
+ - ▁space
557
+ - ▁word
558
+ - ▁hard
559
+ - ▁children
560
+ - ▁organ
561
+ - ▁always
562
+ - ▁country
563
+ - ▁reason
564
+ - ▁experience
565
+ - ▁large
566
+ - ▁everything
567
+ - ▁friend
568
+ - ▁project
569
+ - ▁computer
570
+ - ▁fifty
571
+ - ▁money
572
+ - ▁information
573
+ - graph
574
+ - ▁walk
575
+ - ization
576
+ - ▁africa
577
+ - ▁picture
578
+ - ▁process
579
+ - ▁teach
580
+ - ▁enough
581
+ - ▁elect
582
+ - ▁thirty
583
+ - '0'
584
+ - '1'
585
+ - '2'
586
+ - '9'
587
+ - '3'
588
+ - '5'
589
+ - '8'
590
+ - '4'
591
+ - '7'
592
+ - '6'
593
+ - '&'
594
+ - +
595
+ - '#'
596
+ - '@'
597
+ - '*'
598
+ - \
599
+ - ^
600
+ - R
601
+ - _
602
+ - '-'
603
+ - '%'
604
+ - '='
605
+ - $
606
+ - M
607
+ - ā
608
+ - ']'
609
+ - E
610
+ - U
611
+ - A
612
+ - G
613
+ - '['
614
+ - <sos/eos>
615
+ init: null
616
+ input_size: null
617
+ ctc_conf:
618
+ dropout_rate: 0.0
619
+ ctc_type: builtin
620
+ reduce: true
621
+ ignore_nan_grad: null
622
+ zero_infinity: true
623
+ joint_net_conf: null
624
+ use_preprocessor: true
625
+ token_type: bpe
626
+ bpemodel: data/en_token_list/bpe_unigram500/bpe.model
627
+ non_linguistic_symbols: null
628
+ cleaner: null
629
+ g2p: null
630
+ speech_volume_normalize: null
631
+ rir_scp: null
632
+ rir_apply_prob: 1.0
633
+ noise_scp: null
634
+ noise_apply_prob: 1.0
635
+ noise_db_range: '13_15'
636
+ short_noise_thres: 0.5
637
+ aux_ctc_tasks: []
638
+ frontend: default
639
+ frontend_conf:
640
+ n_fft: 512
641
+ win_length: 400
642
+ hop_length: 160
643
+ fs: 16k
644
+ specaug: specaug
645
+ specaug_conf:
646
+ apply_time_warp: true
647
+ time_warp_window: 5
648
+ time_warp_mode: bicubic
649
+ apply_freq_mask: true
650
+ freq_mask_width_range:
651
+ - 0
652
+ - 27
653
+ num_freq_mask: 2
654
+ apply_time_mask: true
655
+ time_mask_width_ratio_range:
656
+ - 0.0
657
+ - 0.05
658
+ num_time_mask: 5
659
+ normalize: global_mvn
660
+ normalize_conf:
661
+ stats_file: exp/asr_stats_raw_en_bpe500_sp/train/feats_stats.npz
662
+ model: espnet
663
+ model_conf:
664
+ ctc_weight: 0.3
665
+ lsm_weight: 0.1
666
+ length_normalized_loss: false
667
+ preencoder: null
668
+ preencoder_conf: {}
669
+ encoder: conformer
670
+ encoder_conf:
671
+ output_size: 256
672
+ attention_heads: 4
673
+ linear_units: 1024
674
+ num_blocks: 12
675
+ dropout_rate: 0.1
676
+ positional_dropout_rate: 0.1
677
+ attention_dropout_rate: 0.1
678
+ input_layer: conv2d
679
+ normalize_before: true
680
+ macaron_style: true
681
+ rel_pos_type: latest
682
+ pos_enc_layer_type: rel_pos
683
+ selfattention_layer_type: rel_selfattn
684
+ activation_type: swish
685
+ use_cnn_module: true
686
+ cnn_module_kernel: 31
687
+ postencoder: null
688
+ postencoder_conf: {}
689
+ decoder: transformer
690
+ decoder_conf:
691
+ attention_heads: 4
692
+ linear_units: 2048
693
+ num_blocks: 6
694
+ dropout_rate: 0.1
695
+ positional_dropout_rate: 0.1
696
+ self_attention_dropout_rate: 0.1
697
+ src_attention_dropout_rate: 0.1
698
+ preprocessor: default
699
+ preprocessor_conf: {}
700
+ required:
701
+ - output_dir
702
+ - token_list
703
+ version: '202301'
704
+ distributed: true
asr_train_raw_en_bpe500_sp/images/acc.png ADDED
asr_train_raw_en_bpe500_sp/images/backward_time.png ADDED
asr_train_raw_en_bpe500_sp/images/cer.png ADDED
asr_train_raw_en_bpe500_sp/images/cer_ctc.png ADDED
asr_train_raw_en_bpe500_sp/images/forward_time.png ADDED
asr_train_raw_en_bpe500_sp/images/gpu_max_cached_mem_GB.png ADDED
asr_train_raw_en_bpe500_sp/images/iter_time.png ADDED
asr_train_raw_en_bpe500_sp/images/loss.png ADDED
asr_train_raw_en_bpe500_sp/images/loss_att.png ADDED
asr_train_raw_en_bpe500_sp/images/loss_ctc.png ADDED
asr_train_raw_en_bpe500_sp/images/optim0_lr0.png ADDED
asr_train_raw_en_bpe500_sp/images/optim_step_time.png ADDED
asr_train_raw_en_bpe500_sp/images/train_time.png ADDED
asr_train_raw_en_bpe500_sp/images/wer.png ADDED
asr_train_raw_en_bpe500_sp/valid.acc.ave_10best.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f5389ba035589c49132d662c7cc9f9fdd6b0e574a566f0141c1ca8d231f61341
3
+ size 123413191
lm_train_lm_en_bpe500/config.yaml ADDED
@@ -0,0 +1,627 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ config: conf/train_lm.yaml
2
+ print_config: false
3
+ log_level: INFO
4
+ dry_run: false
5
+ iterator_type: sequence
6
+ output_dir: exp/lm_train_lm_en_bpe500
7
+ ngpu: 1
8
+ seed: 0
9
+ num_workers: 1
10
+ num_att_plot: 3
11
+ dist_backend: nccl
12
+ dist_init_method: env://
13
+ dist_world_size: 4
14
+ dist_rank: 0
15
+ local_rank: 0
16
+ dist_master_addr: localhost
17
+ dist_master_port: 60077
18
+ dist_launcher: null
19
+ multiprocessing_distributed: true
20
+ unused_parameters: false
21
+ sharded_ddp: false
22
+ cudnn_enabled: true
23
+ cudnn_benchmark: false
24
+ cudnn_deterministic: true
25
+ collect_stats: false
26
+ write_collected_feats: false
27
+ max_epoch: 50
28
+ patience: null
29
+ val_scheduler_criterion:
30
+ - valid
31
+ - loss
32
+ early_stopping_criterion:
33
+ - valid
34
+ - loss
35
+ - min
36
+ best_model_criterion:
37
+ - - valid
38
+ - loss
39
+ - min
40
+ keep_nbest_models: 10
41
+ nbest_averaging_interval: 0
42
+ grad_clip: 5.0
43
+ grad_clip_type: 2.0
44
+ grad_noise: false
45
+ accum_grad: 1
46
+ no_forward_run: false
47
+ resume: true
48
+ train_dtype: float32
49
+ use_amp: false
50
+ log_interval: null
51
+ use_matplotlib: true
52
+ use_tensorboard: true
53
+ create_graph_in_tensorboard: false
54
+ use_wandb: false
55
+ wandb_project: null
56
+ wandb_id: null
57
+ wandb_entity: null
58
+ wandb_name: null
59
+ wandb_model_log_interval: -1
60
+ detect_anomaly: false
61
+ pretrain_path: null
62
+ init_param: []
63
+ ignore_init_mismatch: false
64
+ freeze_param: []
65
+ num_iters_per_epoch: null
66
+ batch_size: 20
67
+ valid_batch_size: null
68
+ batch_bins: 20000000
69
+ valid_batch_bins: null
70
+ train_shape_file:
71
+ - exp/lm_stats_en_bpe500/train/text_shape.bpe
72
+ valid_shape_file:
73
+ - exp/lm_stats_en_bpe500/valid/text_shape.bpe
74
+ batch_type: numel
75
+ valid_batch_type: null
76
+ fold_length:
77
+ - 150
78
+ sort_in_batch: descending
79
+ sort_batch: descending
80
+ multiple_iterator: false
81
+ chunk_length: 500
82
+ chunk_shift_ratio: 0.5
83
+ num_cache_chunks: 1024
84
+ train_data_path_and_name_and_type:
85
+ - - dump/raw/lm_train.txt
86
+ - text
87
+ - text
88
+ valid_data_path_and_name_and_type:
89
+ - - dump/raw/dev/text
90
+ - text
91
+ - text
92
+ allow_variable_data_keys: false
93
+ max_cache_size: 0.0
94
+ max_cache_fd: 32
95
+ valid_max_cache_size: null
96
+ exclude_weight_decay: false
97
+ exclude_weight_decay_conf: {}
98
+ optim: adam
99
+ optim_conf:
100
+ lr: 0.001
101
+ scheduler: warmuplr
102
+ scheduler_conf:
103
+ warmup_steps: 25000
104
+ token_list:
105
+ - <blank>
106
+ - <unk>
107
+ - '[unk]'
108
+ - ▁
109
+ - s
110
+ - ▁the
111
+ - t
112
+ - ▁and
113
+ - e
114
+ - ▁a
115
+ - ▁to
116
+ - d
117
+ - ▁of
118
+ - ''''
119
+ - n
120
+ - ing
121
+ - ▁in
122
+ - ▁that
123
+ - re
124
+ - ▁i
125
+ - c
126
+ - o
127
+ - u
128
+ - ▁we
129
+ - y
130
+ - a
131
+ - ed
132
+ - ▁it
133
+ - ▁you
134
+ - i
135
+ - m
136
+ - ▁is
137
+ - er
138
+ - p
139
+ - g
140
+ - w
141
+ - al
142
+ - ▁this
143
+ - ▁so
144
+ - f
145
+ - le
146
+ - b
147
+ - ar
148
+ - ▁f
149
+ - k
150
+ - ▁c
151
+ - r
152
+ - in
153
+ - or
154
+ - ▁for
155
+ - ▁be
156
+ - ve
157
+ - ▁was
158
+ - te
159
+ - th
160
+ - ▁do
161
+ - es
162
+ - ly
163
+ - ▁they
164
+ - ro
165
+ - ▁are
166
+ - ▁with
167
+ - ▁have
168
+ - an
169
+ - v
170
+ - ch
171
+ - ▁on
172
+ - se
173
+ - lo
174
+ - ▁but
175
+ - en
176
+ - ri
177
+ - li
178
+ - ▁what
179
+ - it
180
+ - ic
181
+ - ▁can
182
+ - l
183
+ - ur
184
+ - ce
185
+ - ent
186
+ - ▁me
187
+ - ▁b
188
+ - ▁ma
189
+ - ▁he
190
+ - ra
191
+ - ▁de
192
+ - ll
193
+ - at
194
+ - ▁about
195
+ - ▁one
196
+ - ▁not
197
+ - ne
198
+ - ▁all
199
+ - ▁my
200
+ - ter
201
+ - el
202
+ - il
203
+ - ▁there
204
+ - 'on'
205
+ - ad
206
+ - ▁mo
207
+ - ol
208
+ - ation
209
+ - nd
210
+ - ▁like
211
+ - ▁people
212
+ - po
213
+ - ▁at
214
+ - ▁us
215
+ - us
216
+ - ▁g
217
+ - ci
218
+ - ▁our
219
+ - h
220
+ - pe
221
+ - ▁as
222
+ - ▁from
223
+ - vi
224
+ - ▁if
225
+ - as
226
+ - ▁ex
227
+ - ▁con
228
+ - ▁an
229
+ - ver
230
+ - ▁out
231
+ - ▁just
232
+ - un
233
+ - ▁see
234
+ - la
235
+ - ▁di
236
+ - ▁when
237
+ - ▁now
238
+ - ▁p
239
+ - ha
240
+ - ▁who
241
+ - ck
242
+ - ▁these
243
+ - ▁because
244
+ - ▁or
245
+ - ▁know
246
+ - ion
247
+ - ir
248
+ - ▁co
249
+ - ▁up
250
+ - ▁pa
251
+ - ment
252
+ - ▁think
253
+ - ge
254
+ - ▁how
255
+ - ide
256
+ - ▁by
257
+ - ul
258
+ - ity
259
+ - ▁go
260
+ - ▁get
261
+ - ▁ho
262
+ - ive
263
+ - ▁very
264
+ - ate
265
+ - ng
266
+ - ▁no
267
+ - ▁had
268
+ - ac
269
+ - ▁bo
270
+ - ry
271
+ - ▁more
272
+ - ▁them
273
+ - ▁some
274
+ - mi
275
+ - ▁time
276
+ - ▁your
277
+ - me
278
+ - ▁going
279
+ - op
280
+ - am
281
+ - per
282
+ - et
283
+ - ▁would
284
+ - ru
285
+ - ure
286
+ - ti
287
+ - ist
288
+ - ▁their
289
+ - x
290
+ - ▁were
291
+ - ▁look
292
+ - ▁pro
293
+ - ▁which
294
+ - ▁work
295
+ - tion
296
+ - est
297
+ - ty
298
+ - im
299
+ - z
300
+ - ta
301
+ - ▁want
302
+ - ▁two
303
+ - age
304
+ - ▁really
305
+ - om
306
+ - ma
307
+ - ers
308
+ - ting
309
+ - ▁world
310
+ - co
311
+ - ▁way
312
+ - ▁don
313
+ - wa
314
+ - hi
315
+ - tra
316
+ - ▁la
317
+ - ▁here
318
+ - able
319
+ - lu
320
+ - ▁other
321
+ - mo
322
+ - ies
323
+ - ▁has
324
+ - ▁could
325
+ - j
326
+ - ▁make
327
+ - ally
328
+ - ▁sta
329
+ - ten
330
+ - ▁will
331
+ - ▁un
332
+ - ig
333
+ - ▁where
334
+ - ▁into
335
+ - ke
336
+ - ▁than
337
+ - ▁comp
338
+ - ▁actually
339
+ - tic
340
+ - sh
341
+ - ▁did
342
+ - tor
343
+ - fa
344
+ - ical
345
+ - ▁she
346
+ - ▁years
347
+ - ▁say
348
+ - one
349
+ - ted
350
+ - ▁things
351
+ - ph
352
+ - ▁new
353
+ - ▁pre
354
+ - ▁any
355
+ - ▁thousand
356
+ - ▁been
357
+ - ▁inter
358
+ - ▁his
359
+ - ▁com
360
+ - ▁need
361
+ - nce
362
+ - ▁right
363
+ - ▁take
364
+ - ▁even
365
+ - ▁over
366
+ - ▁start
367
+ - ▁hundred
368
+ - min
369
+ - ▁sp
370
+ - ▁those
371
+ - ▁car
372
+ - ▁then
373
+ - mp
374
+ - ap
375
+ - ▁first
376
+ - les
377
+ - ize
378
+ - ▁every
379
+ - ba
380
+ - ▁something
381
+ - ▁well
382
+ - ard
383
+ - ▁str
384
+ - ▁back
385
+ - und
386
+ - ia
387
+ - pl
388
+ - ki
389
+ - ho
390
+ - ▁call
391
+ - ▁most
392
+ - ▁also
393
+ - bi
394
+ - ▁thing
395
+ - ▁life
396
+ - um
397
+ - ▁said
398
+ - ▁kind
399
+ - ▁lot
400
+ - ▁much
401
+ - va
402
+ - ▁ra
403
+ - ▁little
404
+ - ▁dr
405
+ - ▁got
406
+ - ▁come
407
+ - ful
408
+ - ▁talk
409
+ - ▁part
410
+ - ▁day
411
+ - ant
412
+ - ction
413
+ - ▁happen
414
+ - ▁only
415
+ - ▁many
416
+ - ▁wo
417
+ - pri
418
+ - ▁her
419
+ - ▁br
420
+ - qui
421
+ - ▁mean
422
+ - ▁three
423
+ - iv
424
+ - ▁different
425
+ - ugh
426
+ - ain
427
+ - ▁human
428
+ - ance
429
+ - ▁change
430
+ - ▁let
431
+ - ▁real
432
+ - ▁show
433
+ - ▁good
434
+ - ▁around
435
+ - ▁through
436
+ - ▁jo
437
+ - bu
438
+ - ▁down
439
+ - ight
440
+ - ga
441
+ - ▁why
442
+ - ▁live
443
+ - ff
444
+ - ▁tell
445
+ - ▁put
446
+ - ▁idea
447
+ - port
448
+ - ▁same
449
+ - ▁give
450
+ - ated
451
+ - ish
452
+ - ible
453
+ - ▁though
454
+ - ious
455
+ - ▁problem
456
+ - ▁five
457
+ - par
458
+ - ▁fact
459
+ - ▁cha
460
+ - ition
461
+ - ▁year
462
+ - ▁big
463
+ - ▁plan
464
+ - ▁great
465
+ - ▁find
466
+ - ▁four
467
+ - ▁app
468
+ - ▁after
469
+ - ▁system
470
+ - ▁place
471
+ - ▁em
472
+ - ▁build
473
+ - ▁percent
474
+ - ▁again
475
+ - ▁point
476
+ - ▁learn
477
+ - ▁own
478
+ - ▁long
479
+ - ▁made
480
+ - ▁today
481
+ - ▁nine
482
+ - ities
483
+ - ▁gene
484
+ - ▁six
485
+ - ▁question
486
+ - light
487
+ - ▁should
488
+ - ▁came
489
+ - ▁feel
490
+ - ▁turn
491
+ - ▁person
492
+ - ▁end
493
+ - ▁hu
494
+ - ▁design
495
+ - ▁help
496
+ - ▁brain
497
+ - ▁last
498
+ - ▁create
499
+ - ▁important
500
+ - ▁before
501
+ - ▁high
502
+ - ▁never
503
+ - ▁trans
504
+ - ▁another
505
+ - ▁him
506
+ - ▁eight
507
+ - ▁might
508
+ - ▁understand
509
+ - ▁power
510
+ - ▁better
511
+ - q
512
+ - ▁found
513
+ - ▁play
514
+ - ▁twenty
515
+ - ▁still
516
+ - ▁school
517
+ - ▁each
518
+ - ▁seven
519
+ - ▁together
520
+ - ▁few
521
+ - ▁hand
522
+ - ▁example
523
+ - que
524
+ - ▁next
525
+ - ▁million
526
+ - ▁story
527
+ - ▁women
528
+ - ▁under
529
+ - ▁number
530
+ - ▁course
531
+ - ▁water
532
+ - ▁ago
533
+ - ▁grow
534
+ - ▁between
535
+ - ▁develop
536
+ - ▁america
537
+ - ▁sort
538
+ - ▁technology
539
+ - ▁believe
540
+ - ▁second
541
+ - ▁small
542
+ - ▁maybe
543
+ - ▁become
544
+ - press
545
+ - ▁health
546
+ - ▁space
547
+ - ▁word
548
+ - ▁hard
549
+ - ▁children
550
+ - ▁organ
551
+ - ▁always
552
+ - ▁country
553
+ - ▁reason
554
+ - ▁experience
555
+ - ▁large
556
+ - ▁everything
557
+ - ▁friend
558
+ - ▁project
559
+ - ▁computer
560
+ - ▁fifty
561
+ - ▁money
562
+ - ▁information
563
+ - graph
564
+ - ▁walk
565
+ - ization
566
+ - ▁africa
567
+ - ▁picture
568
+ - ▁process
569
+ - ▁teach
570
+ - ▁enough
571
+ - ▁elect
572
+ - ▁thirty
573
+ - '0'
574
+ - '1'
575
+ - '2'
576
+ - '9'
577
+ - '3'
578
+ - '5'
579
+ - '8'
580
+ - '4'
581
+ - '7'
582
+ - '6'
583
+ - '&'
584
+ - +
585
+ - '#'
586
+ - '@'
587
+ - '*'
588
+ - \
589
+ - ^
590
+ - R
591
+ - _
592
+ - '-'
593
+ - '%'
594
+ - '='
595
+ - $
596
+ - M
597
+ - ā
598
+ - ']'
599
+ - E
600
+ - U
601
+ - A
602
+ - G
603
+ - '['
604
+ - <sos/eos>
605
+ init: null
606
+ model_conf:
607
+ ignore_id: 0
608
+ use_preprocessor: true
609
+ token_type: bpe
610
+ bpemodel: data/en_token_list/bpe_unigram500/bpe.model
611
+ non_linguistic_symbols: null
612
+ cleaner: null
613
+ g2p: null
614
+ lm: transformer
615
+ lm_conf:
616
+ pos_enc: null
617
+ embed_unit: 128
618
+ att_unit: 512
619
+ head: 8
620
+ unit: 2048
621
+ layer: 15
622
+ dropout_rate: 0.1
623
+ required:
624
+ - output_dir
625
+ - token_list
626
+ version: '202301'
627
+ distributed: true
lm_train_lm_en_bpe500/images/backward_time.png ADDED
lm_train_lm_en_bpe500/images/forward_time.png ADDED
lm_train_lm_en_bpe500/images/gpu_max_cached_mem_GB.png ADDED
lm_train_lm_en_bpe500/images/iter_time.png ADDED
lm_train_lm_en_bpe500/images/loss.png ADDED
lm_train_lm_en_bpe500/images/optim0_lr0.png ADDED
lm_train_lm_en_bpe500/images/optim_step_time.png ADDED
lm_train_lm_en_bpe500/images/train_time.png ADDED
lm_train_lm_en_bpe500/perplexity_test/ppl ADDED
@@ -0,0 +1 @@
 
 
1
+ 10.155182315138273
lm_train_lm_en_bpe500/valid.loss.ave_10best.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ec899e08f7764595b4bf1d45cd88d21156c92c7cb9885071e53d118232970df0
3
+ size 190793638