new run

run.sh

@@ -5,11 +5,11 @@ python run_speech_recognition_ctc.py \
 --dataset_config_name="fr" \
 --output_dir="./" \
 --overwrite_output_dir \
---num_train_epochs="5" \
+--num_train_epochs="2" \
 --per_device_train_batch_size="64" \
 --per_device_eval_batch_size="64" \
 --gradient_accumulation_steps="1" \
---learning_rate="7e-5" \
+--learning_rate="1e-4" \
 --warmup_steps="1500" \
 --length_column_name="input_length" \
 --evaluation_strategy="steps" \
@@ -18,19 +18,19 @@ python run_speech_recognition_ctc.py \
 --eval_steps="500" \
 --logging_steps="100" \
 --layerdrop="0.0" \
---activation_dropout="0.1" \
---save_total_limit="3" \
+--activation_dropout="0.05" \
+--save_total_limit="2" \
 --freeze_feature_encoder \
 --feat_proj_dropout="0.0" \
 --mask_time_prob="0.75" \
 --mask_time_length="10" \
---mask_feature_prob="0.33" \
+--mask_feature_prob="0.4" \
 --mask_feature_length="10" \
 --gradient_checkpointing \
 --report_to="wandb" \
 --run_name="xls-r-300m-fr" \
---max_eval_samples="4000" \
---max_duration_in_seconds="5" \
+--max_eval_samples="6000" \
+--max_duration_in_seconds="9" \
 --use_auth_token \
 --fp16 \
 --group_by_length \