training / flax /long_form_transcription_scripts /run_tedlium_long_form_timestamps.sh
supawichwac's picture
Saving train state of step 50
55f3766 verified
#!/usr/bin/env bash
python run_long_form_transcription.py \
--model_name_or_path "sanchit-gandhi/large-32-2-ts-freeze-28k-wer-10-v4-8-10k-steps" \
--dataset_name "distil-whisper/tedlium-long-form+distil-whisper/tedlium-long-form" \
--dataset_config_name "all+all" \
--dataset_split_name "validation+test" \
--cache_dir "/home/sanchitgandhi/.cache" \
--dataset_cache_dir "/home/sanchitgandhi/.cache" \
--output_dir "./" \
--wandb_dir "/home/sanchitgandhi/.cache" \
--wandb_project "distil-whisper-long-form" \
--wandb_name "large-32-2-ts-freeze-28k-wer-10-v4-8-10k-steps-tedlium-timestamps" \
--per_device_eval_batch_size 32 \
--dtype "bfloat16" \
--report_to "wandb" \
--streaming \
--return_timestamps