training / flax /distillation_scripts /run_librispeech.sh
GalaktischeGurke's picture
Saving train state of step 5000
4008bf9 verified
raw
history blame
1.14 kB
#!/usr/bin/env bash
TCMALLOC_LARGE_ALLOC_REPORT_THRESHOLD=10000000000 python run_distillation.py \
--model_name_or_path "distil-whisper/large-32-2" \
--teacher_model_name_or_path "openai/whisper-large-v2" \
--dataset_name "distil-whisper/librispeech_asr" \
--dataset_config_name "all" \
--train_split_name "train.clean.100+train.clean.360+train.other.500" \
--eval_split_name "validation.clean" \
--text_column_name "whisper_transcript" \
--cache_dir "/home/sanchitgandhi/cache" \
--dataset_cache_dir "/home/sanchitgandhi/cache" \
--output_dir "./" \
--wandb_name "large-32-2-ts-librispeech" \
--wandb_dir "/home/sanchitgandhi/.cache" \
--wandb_project "distil-whisper-librispeech" \
--per_device_train_batch_size 32 \
--per_device_eval_batch_size 16 \
--dtype "bfloat16" \
--learning_rate 1e-4 \
--warmup_steps 500 \
--temperature 2.0 \
--do_train \
--do_eval \
--num_train_epochs 10 \
--preprocessing_num_workers 16 \
--dataloader_num_workers 8 \
--logging_steps 25 \
--use_scan \
--gradient_checkpointing \
--overwrite_output_dir \
--predict_with_generate \
--push_to_hub