training / flax /finetuning_scripts /run_librispeech_eval.sh
GalaktischeGurke's picture
Saving train state of step 5000
4008bf9 verified
raw
history blame
749 Bytes
#!/usr/bin/env bash
python run_eval.py \
--model_name_or_path "./" \
--dataset_name "distil-whisper/librispeech_asr" \
--dataset_config_name "all" \
--test_split_name "validation.clean+validation.other+test.clean+test.other" \
--text_column_name "text" \
--cache_dir "/home/sanchitgandhi/cache" \
--dataset_cache_dir "/home/sanchitgandhi/cache" \
--output_dir "./" \
--wandb_name "large-32-2-pl-freeze-librispeech-eval" \
--wandb_dir "/home/sanchitgandhi/.cache" \
--wandb_project "distil-whisper-librispeech" \
--per_device_eval_batch_size 128 \
--dtype "bfloat16" \
--do_predict \
--preprocessing_num_workers 16 \
--dataloader_num_workers 8 \
--load_with_scan \
--predict_with_generate \
--report_to "wandb"