Yehor
/

wav2vec2-xls-r-300m-uk-with-lm

Automatic Speech Recognition

mozilla-foundation/common_voice_7_0

generated_from_trainer

Inference Endpoints

Model card Files Files and versions Community

Yehor commited on Jan 28, 2022

Commit

2c39323

•

1 Parent(s): 6c60626

Update run.sh

Files changed (1) hide show

run.sh +3 -3

run.sh CHANGED Viewed

@@ -2,14 +2,14 @@
 python run_speech_recognition_ctc.py \
 	--dataset_name="mozilla-foundation/common_voice_7_0" \
-	--model_name_or_path="facebook/wav2vec2-xls-r-1b" \
 	--dataset_config_name="uk" \
 	--output_dir="./cv-uk-ft" \
 	--num_train_epochs="100" \
 	--per_device_train_batch_size="8" \
 	--per_device_eval_batch_size="8" \
 	--gradient_accumulation_steps="20" \
-	--learning_rate="5e-5" \
 	--warmup_steps="500" \
 	--length_column_name="input_length" \
 	--evaluation_strategy="steps" \
@@ -27,7 +27,7 @@ python run_speech_recognition_ctc.py \
 	--mask_feature_prob="0.25" \
 	--mask_feature_length="64" \
 	--freeze_feature_encoder \
-	--chars_to_ignore « » ы — – ՚ ’ … , ? . ! - \; \: \" “ % ‘ \` ” � \
 	--fp16 \
 	--group_by_length \
 	--use_auth_token \

 python run_speech_recognition_ctc.py \
 	--dataset_name="mozilla-foundation/common_voice_7_0" \
+	--model_name_or_path="facebook/wav2vec2-xls-r-300m" \
 	--dataset_config_name="uk" \
 	--output_dir="./cv-uk-ft" \
 	--num_train_epochs="100" \
 	--per_device_train_batch_size="8" \
 	--per_device_eval_batch_size="8" \
 	--gradient_accumulation_steps="20" \
+	--learning_rate="3e-4" \
 	--warmup_steps="500" \
 	--length_column_name="input_length" \
 	--evaluation_strategy="steps" \
 	--mask_feature_prob="0.25" \
 	--mask_feature_length="64" \
 	--freeze_feature_encoder \
+	--chars_to_ignore — , ? . ! - \; \: \" “ % ‘ ” � \
 	--fp16 \
 	--group_by_length \
 	--use_auth_token \