test
Browse files- run_restore.sh +38 -0
run_restore.sh
ADDED
@@ -0,0 +1,38 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
python run_speech_recognition_seq2seq_streaming.py \
|
2 |
+
--model_name_or_path="pere/whisper-small-nob-clr" \
|
3 |
+
--dataset_name="NbAiLab/NCC_S" \
|
4 |
+
--language="Norwegian" \
|
5 |
+
--train_split_name="train" \
|
6 |
+
--eval_split_name="validation" \
|
7 |
+
--model_index_name="Whisper Base Norwegian" \
|
8 |
+
--max_steps="8000" \
|
9 |
+
--output_dir="./" \
|
10 |
+
--per_device_train_batch_size="64" \
|
11 |
+
--per_device_eval_batch_size="32" \
|
12 |
+
--gradient_accumulation_steps="2" \
|
13 |
+
--lr_scheduler_type="constant_with_warmup" \
|
14 |
+
--logging_steps="50" \
|
15 |
+
--learning_rate="1e-5" \
|
16 |
+
--warmup_steps="0" \
|
17 |
+
--evaluation_strategy="steps" \
|
18 |
+
--eval_steps="1000" \
|
19 |
+
--save_strategy="steps" \
|
20 |
+
--save_steps="1000" \
|
21 |
+
--generation_max_length="225" \
|
22 |
+
--length_column_name="duration" \
|
23 |
+
--max_duration_in_seconds="30" \
|
24 |
+
--text_column_name="text" \
|
25 |
+
--freeze_feature_encoder="False" \
|
26 |
+
--report_to="tensorboard" \
|
27 |
+
--metric_for_best_model="wer" \
|
28 |
+
--greater_is_better="False" \
|
29 |
+
--load_best_model_at_end \
|
30 |
+
--gradient_checkpointing \
|
31 |
+
--fp16 \
|
32 |
+
--overwrite_output_dir \
|
33 |
+
--do_train \
|
34 |
+
--do_eval \
|
35 |
+
--predict_with_generate \
|
36 |
+
--do_normalize_eval \
|
37 |
+
--use_auth_token \
|
38 |
+
--push_to_hub
|