1#!/usr/bin/env bash
2CUDA_VISIBLE_DEVICES=0 python run_speech_recognition_whisper.py \
3 --model_name_or_path="medium.en" \
4 --dataset_name="esc-benchmark/esc-datasets" \
5 --dataset_config_name="tedlium" \
6 --max_steps="2500" \
7 --output_dir="./" \
8 --run_name="whisper-tedlium" \
9 --wandb_project="whisper" \
10 --per_device_train_batch_size="64" \
11 --per_device_eval_batch_size="16" \
12 --logging_steps="25" \
13 --learning_rate="1e-4" \
14 --warmup_steps="500" \
15 --report_to="wandb" \
16 --preprocessing_num_workers="16" \
17 --evaluation_strategy="steps" \
18 --eval_steps="500" \
19 --save_strategy="steps" \
20 --save_steps="500" \
21 --generation_max_length="224" \
22 --length_column_name="input_lengths" \
23 --gradient_checkpointing \
24 --group_by_length \
25 --freeze_encoder \
26 --fp16 \
27 --overwrite_output_dir \
28 --do_train \
29 --do_eval \
30 --do_predict \
31 --predict_with_generate \
32 --use_auth_token
33
