training / flax /finetuning_scripts /run_librispeech_dummy.sh
GalaktischeGurke's picture
Saving train state of step 5000
4008bf9 verified
raw
history blame
846 Bytes
#!/usr/bin/env bash
python run_finetuning.py \
--model_name_or_path "distil-whisper/tiny-random-whisper" \
--dataset_name "distil-whisper/librispeech_asr" \
--dataset_config_name "all" \
--train_split_name "train.clean.100[:1024]" \
--eval_split_name "validation.clean[:1024]" \
--cache_dir "/home/sanchitgandhi/cache" \
--dataset_cache_dir "/home/sanchitgandhi/cache" \
--wandb_dir "/home/sanchitgandhi/.cache" \
--text_column_name "text" \
--output_dir "./" \
--do_train \
--do_eval \
--per_device_train_batch_size 8 \
--per_device_eval_batch_size 4 \
--dtype "bfloat16" \
--num_train_epochs 2 \
--dataloader_num_workers 16 \
--freeze_encoder \
--wandb_project "distil-whisper-debug" \
--logging_steps 2 \
--use_scan \
--gradient_checkpointing \
--overwrite_output_dir \
--predict_with_generate