test

run.sh

@@ -1,5 +1,4 @@
-deepspeed --master_port 29501 run_speech_recognition_seq2seq_streaming.py \
-	--deepspeed="ds_config.json" \
+python run_speech_recognition_seq2seq_streaming.py \
 	--model_name_or_path="openai/whisper-small" \
 	--dataset_name="NbAiLab/NCC_S3_NRK" \
 	--language="Norwegian" \
@@ -8,7 +7,7 @@ deepspeed --master_port 29501 run_speech_recognition_seq2seq_streaming.py \
 	--model_index_name="Whisper Small Norwegian" \
 	--max_steps="100000" \
 	--output_dir="./" \
-	--per_device_train_batch_size="128" \
+	--per_device_train_batch_size="64" \
 	--per_device_eval_batch_size="32" \
 	--gradient_accumulation_steps="2" \
 	--logging_steps="25" \

run_deepspeed.sh

@@ -0,0 +1,39 @@
+deepspeed --master_port 29501 run_speech_recognition_seq2seq_streaming.py \
+	--deepspeed="ds_config.json" \
+	--model_name_or_path="openai/whisper-small" \
+	--dataset_name="NbAiLab/NCC_S3_NRK" \
+	--language="Norwegian" \
+	--train_split_name="train" \
+	--eval_split_name="validation" \
+	--model_index_name="Whisper Small Norwegian" \
+	--max_steps="100000" \
+	--output_dir="./" \
+	--per_device_train_batch_size="128" \
+	--per_device_eval_batch_size="32" \
+	--gradient_accumulation_steps="2" \
+	--logging_steps="25" \
+	--learning_rate="3e-6" \
+	--warmup_steps="1000" \
+	--evaluation_strategy="steps" \
+	--eval_steps="1000" \
+	--save_strategy="steps" \
+	--save_steps="1000" \
+	--generation_max_length="225" \
+	--length_column_name="duration" \
+	--max_duration_in_seconds="30" \
+	--text_column_name="text" \
+	--freeze_feature_encoder="False" \
+	--report_to="tensorboard" \
+	--metric_for_best_model="wer" \
+	--greater_is_better="False" \
+	--load_best_model_at_end \
+	--gradient_checkpointing \
+	--fp16 \
+	--overwrite_output_dir \
+	--do_train \
+	--do_eval \
+	--predict_with_generate \
+	--do_normalize_eval \
+	--use_auth_token \
+	--push_to_hub \
+	--hub_model_id="NbAiLab/whisper-small-nob"