File size: 1,996 Bytes
12a385e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
#!/usr/bin/env bash

declare -a learning_rates=("1e-5" "3e-5" "1e-4" "3e-4" "1e-3")
declare -a batch_sizes=("8" "12" "14" "16")
declare -a gradient_accumulation_step_sizes=("2" "4" "8")

for learning_rate in "${learning_rates[@]}"; do
    for batch_size in "${batch_sizes[@]}"; do
        for gradient_accumulation_steps in "${gradient_accumulation_step_sizes[@]}"; do
	    python create_model.py
            CUDA_VISIBLE_DEVICES=0 python run_speech_recognition_seq2seq.py \
                --dataset_name="librispeech_asr" \
                --model_name_or_path="./" \
                --tokenizer_name="./" \
                --dataset_config_name="clean" \
                --train_split_name="train.100" \
                --eval_split_name="validation" \
                --output_dir="./" \
                --preprocessing_num_workers="1" \
                --length_column_name="input_length" \
                --overwrite_output_dir \
                --num_train_epochs="1" \
                --per_device_train_batch_size=$batch_size \
                --per_device_eval_batch_size=$batch_size \
                --gradient_accumulation_steps=$gradient_accumulation_steps \
                --generation_max_length="40" \
                --generation_num_beams="1" \
                --learning_rate=$learning_rate \
                --warmup_steps="500" \
                --evaluation_strategy="steps" \
                --text_column_name="text" \
                --save_steps="500" \
                --eval_steps="500" \
                --logging_steps="1" \
                --save_total_limit="1" \
                --freeze_feature_encoder \
                --gradient_checkpointing \
                --fp16 \
                --group_by_length \
                --predict_with_generate \
                --do_lower_case \
                --do_train \
                --do_eval \
		--report_to="wandb" \
                --push_to_hub \
                --use_auth_token
        done
    done
done