Training in progress, step 10

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3532fb568b43f9a02717200d3a8f66647c25fb0349ab3eb5a7567b77f850b1c2
+oid sha256:67db960796c75f5af90d6ae38074eea1cad281bdbec28d10187feba4432720bf
 size 151098921

runs/Dec13_12-46-08_d7f040c448a8/1670935580.616109/events.out.tfevents.1670935580.d7f040c448a8.26499.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d0a2d84b4221c89c7f654e3b9f57b33bc79012fd62ef4c4724f3902c2ce77804
+size 5883

runs/Dec13_12-46-08_d7f040c448a8/events.out.tfevents.1670935580.d7f040c448a8.26499.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c2454e09e13c3ad647aeee178d68cb6420103d17f18c2d2dd06a68fb0e289d55
+size 4736

src/run_debug.sh

@@ -7,14 +7,14 @@ python src/run_speech_recognition_seq2seq_streaming.py \
 	--eval_split_name="validation" \
 	--model_index_name="Whisper Tiny Belarusian" \
 	\
-	--max_steps="300" \
+	--max_steps="100" \
 	--max_eval_samples="64" \
 	--output_dir="./" \
 	--per_device_train_batch_size="32" \
 	--per_device_eval_batch_size="32" \
 	--logging_steps="10" \
-	--learning_rate="1e-5" \
-	--warmup_steps="0" \
+	--learning_rate="1e-4" \
+	--warmup_steps="10" \
 	--evaluation_strategy="steps" \
 	--eval_steps="10" \
 	--save_strategy="steps" \
@@ -34,6 +34,7 @@ python src/run_speech_recognition_seq2seq_streaming.py \
 	\
 	--do_train \
 	--do_eval \
+	--overwrite_output_dir \
 	--ignore_data_skip \
 	--predict_with_generate \
 	--do_normalize_eval \

train.log

@@ -132,3 +132,4 @@
   eval_samples_per_second =      3.804
   eval_steps_per_second   =      0.119
   eval_wer                =    52.1978
+{'loss': 2.5622, 'learning_rate': 7e-05, 'epoch': 0.1}

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ba3e5752f0bcb0a1a12ea56a06813c45fc9e560ac738b4458d08b4141a6bb434
+oid sha256:c8a86b0841271f87ec4240b29b5ca1620f243ad5ad67d99f9919cf45237c2add
 size 3643