Stop training 3373 steps

run.sh

@@ -5,11 +5,11 @@ python run_speech_recognition_ctc.py \
 --preprocessing_num_workers="8" \
 --output_dir="./" \
 --overwrite_output_dir \
---num_train_epochs="5" \
+--num_train_epochs="150" \
 --per_device_train_batch_size="4" \
 --per_device_eval_batch_size="4" \
 --gradient_accumulation_steps="8" \
---learning_rate="7.5e-5" \
+--learning_rate="1e-6" \
 --warmup_steps="2000" \
 --length_column_name="input_length" \
 --evaluation_strategy="steps" \
@@ -17,13 +17,13 @@ python run_speech_recognition_ctc.py \
 --chars_to_ignore , ? . ! \- \; \: \" “ % ‘ ” � — ’ … – \
 --save_steps="500" \
 --eval_steps="500" \
---logging_steps="100" \
+--logging_steps="10" \
 --layerdrop="0.0" \
 --activation_dropout="0.1" \
---save_total_limit="3" \
+--save_total_limit="2" \
 --freeze_feature_encoder \
 --feat_proj_dropout="0.0" \
---mask_time_prob="0.75" \
+--mask_time_prob="0.3" \
 --mask_time_length="10" \
 --mask_feature_prob="0.25" \
 --mask_feature_length="64" \

run_bnb.sh

@@ -8,7 +8,7 @@ python run_speech_recognition_ctc_bnb.py \
 --num_train_epochs="150" \
 --per_device_train_batch_size="4" \
 --per_device_eval_batch_size="4" \
---gradient_accumulation_steps="8" \
+--gradient_accumulation_steps="12" \
 --learning_rate="1e-6" \
 --warmup_steps="2000" \
 --length_column_name="input_length" \

runs/Jan30_00-36-51_job-e3562c4a-f6a3-47dc-8131-7e578f1139a3/events.out.tfevents.1643503156.job-e3562c4a-f6a3-47dc-8131-7e578f1139a3.1780523.0

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9d797681ef19c9f4562c88f4f6006ceba29de27dfbc0c50851b7113fd1dd658b
-size 53651
+oid sha256:a1ce5d45d13b812ad24edefb2c595a47609c8c2281e1af6e5c59bf5edeea5c63
+size 59460