marcoyang commited on
Commit
157c6c1
1 Parent(s): a40d894

upload train.sh and compute_ppl.sh

Browse files
Files changed (2) hide show
  1. compute_ppl.sh +12 -0
  2. train.sh +17 -0
compute_ppl.sh ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env bash
2
+
3
+ set -eou pipefail
4
+
5
+ export CUDA_VISIBLE_DEVICES="0"
6
+ export TORCH_DISTRIBUTED_DEBUG="INFO"
7
+
8
+ python ./transformer_lm/compute_perplexity.py \
9
+ --epoch 11 \
10
+ --avg 4 \
11
+ --exp-dir transformer_lm/exp_full_libri_16layer_8gpu \
12
+ --lm-data ./transformer_lm/libri_lm_training_bpe500/sorted_lm_data-valid.pt
train.sh ADDED
@@ -0,0 +1,17 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env bash
2
+
3
+ set -eou pipefail
4
+
5
+ export CUDA_VISIBLE_DEVICES="0,1,2,3,4,5,6,7"
6
+ export TORCH_DISTRIBUTED_DEBUG="INFO"
7
+
8
+ python ./transformer_lm/train.py \
9
+ --start-epoch 0 \
10
+ --world-size 8 \
11
+ --exp-dir transformer_lm/exp_full_libri_16layer_8gpu \
12
+ --num-epochs 20 \
13
+ --lm-data ./transformer_lm/libri_lm_training_bpe500/sorted-lm-data-libri-lm_maxlen200.pt \
14
+ --lm-data-valid ./transformer_lm/libri_lm_training_bpe500/sorted_lm_data-valid.pt \
15
+ --use-fp16 0 \
16
+ --num-layers 16 \
17
+ --batch-size 70