File size: 1,329 Bytes
aae0a3f
 
0cf69a1
aae0a3f
 
 
 
 
 
 
 
 
0cf69a1
aae0a3f
 
 
0cf69a1
aae0a3f
 
 
 
0cf69a1
aae0a3f
82b1929
aae0a3f
 
 
 
 
 
 
 
 
 
 
0cf69a1
aae0a3f
 
3d81ebf
aae0a3f
 
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
#!/usr/bin/env bash
TOKENIZERS_PARALLELISM=false python3 run_distillation.py \
  --model_name_or_path "./first-save" \
  --teacher_model_name_or_path "NbAiLab/nb-whisper-large" \
  --train_dataset_name "NbAiLab/annotated_ncc_speech_styling_v2_vad3_distil_postLv2" \
  --train_dataset_config_name "" \
  --train_split_name "train" \
  --eval_dataset_name "NbAiLab/ncc_speech_v7" \
  --eval_dataset_config_name "" \
  --eval_split_name "validation_norwegian_fleurs" \
  --eval_steps 500 \
  --save_steps 5000 \
  --warmup_steps 0 \
  --learning_rate 0.0003 \
  --lr_scheduler_type "constant_with_warmup" \
  --logging_steps 500 \
  --save_total_limit 2 \
  --max_steps 200000 \
  --wer_threshold 10 \
  --per_device_train_batch_size 4\
  --per_device_eval_batch_size 4 \
  --dataloader_num_workers 32 \
  --dtype "bfloat16" \
  --output_dir "./nb-distil-whisper-large-flax7" \
  --do_train \
  --do_eval \
  --use_scan \
  --gradient_checkpointing \
  --overwrite_output_dir \
  --predict_with_generate \
  --freeze_encoder \
  --streaming \
  --use_auth_token \
  --report_to "wandb" \
  --wandb_project "nb-distil-whisper-large-fleurseval" \
  --wandb_name "flax_experiment2_bs4_v5_1e4_wer10_restart10k" \
  --save_code_to_wandb \
  --save_train_state \
  --hub_model_id "NbAiLab/nb-distil-whisper-large-flax7"\
  --push_to_hub