training / flax /evaluation_scripts /run_librispeech_eval_dummy.sh
nullonesix's picture
Saving train state of step 1
a1be16b verified
raw
history blame
714 Bytes
#!/usr/bin/env bash
python run_eval.py \
--model_name_or_path "openai/whisper-large-v2" \
--dataset_name "gigaspeech-l+gigaspeech-l" \
--dataset_config_name "l+l" \
--dataset_split_name "train+validation" \
--text_column_name "text" \
--cache_dir "/home/sanchitgandhi/.cache" \
--dataset_cache_dir "/home/sanchitgandhi/.cache" \
--output_dir "./" \
--wandb_dir "/home/sanchitgandhi/.cache" \
--wandb_project "distil-whisper-label" \
--wandb_name "whisper-large-v2-gigaspeech-l-with-audio" \
--per_device_eval_batch_size 64 \
--dtype "bfloat16" \
--dataloader_num_workers 0 \
--report_to "wandb" \
--streaming \
--max_eval_samples 1024 \
--predict_with_generate \
--log_audio