run full training

run.sh

@@ -14,8 +14,8 @@ python $1run_speech_recognition_seq2seq_streaming.py \
 	--per_device_train_batch_size="32" \
 	--per_device_eval_batch_size="16" \
 	--logging_steps="25" \
-	--learning_rate="1e-5" \
-	--warmup_steps="500" \
+	--learning_rate="3e-6" \
+	--warmup_ratio="0.1" \
 	--evaluation_strategy="steps" \
 	--eval_steps="1000" \
 	--save_strategy="steps" \
@@ -26,12 +26,13 @@ python $1run_speech_recognition_seq2seq_streaming.py \
 	--text_column_name="sentence,text,raw_transcription" \
 	--freeze_feature_encoder="False" \
 	--report_to="wandb" \
-	--save_total_limit="3" \
+	--save_total_limit="2" \
 	--metric_for_best_model="wer" \
 	--greater_is_better="False" \
 	--load_best_model_at_end \
 	--gradient_checkpointing \
 	--overwrite_output_dir \
+	--do_train \
 	--do_eval \
 	--fp16 \
 	--predict_with_generate \