SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

[s2s] add create student script (#7290)

Browse Source 
Co-authored-by: Suraj Patil <surajp815@gmail.com>
Co-authored-by: Sam Shleifer <sshleifer@gmail.com>
This commit is contained in:
Suraj Patil
2020-09-28 00:40:46 +05:30
committed by GitHub
parent e50a931c11
commit eab5f59682
9 changed files with 312 additions and 314 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
examples/seq2seq/train_distilbart_xsum.sh
View File

@@ -1,21 +1,20 @@
#!/usr/bin/env bash
export PYTHONPATH="../":"${PYTHONPATH}"
export BS=16
export GAS=2
python distillation.py \
--teacher facebook/bart-large-xsum --data_dir xsum \
--student_decoder_layers 6 --student_encoder_layers 12 \
--freeze_encoder --freeze_embeds \
--learning_rate=3e-4 \
--do_train \
--do_predict \
--fp16 \
--val_check_interval 0.1 --n_val 1000 \
--teacher facebook/bart-large-xsum --data_dir $XSUM_DIR \
--fp16 --fp16_opt_level=O1 \
--val_check_interval 0.1 --n_val 1000 --eval_beams 2 --length_penalty=0.5 \
--max_target_length=60 --val_max_target_length=60 --test_max_target_length=100 \
--student_decoder_layers 6 --student_encoder_layers 12 \
--freeze_encoder --freeze_embeds \
--model_name_or_path IGNORED \
--alpha_hid=3. --length_penalty=0.5 \
--train_batch_size=$BS --eval_batch_size=$BS --gradient_accumulation_steps=$GAS --num_train_epochs=6 \
--tokenizer_name facebook/bart-large \
--alpha_hid=3. \
--train_batch_size=16 --eval_batch_size=16 --gradient_accumulation_steps=2 \
--sortish_sampler \
--num_train_epochs=6 \
--warmup_steps 500 \
--output_dir distilbart_xsum_12_6 \
"$@"