Training in progress, step 3000

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3d1467cc7d81b669faa608f96f0e48ecce3eb6d6824347594c97fcf770153c16
+oid sha256:b1c26c114980c4b75886387eff869a323b4cb16bc039f7dc7b0380a2737ca6e6
 size 1527847357

run.sh

@@ -1,18 +1,21 @@
 deepspeed run-ba.py \
 	--deepspeed="ds_config.json" \
 	--model_name_or_path="emilios/whisper-medium-el" \
-	--dataset_name="MLCommons/ml_spoken_words" \
-	--dataset_config_name="el_wav" \
+	--dataset_name="mozilla-foundation/common_voice_11_0,google/fleurs,MLCommons/ml_spoken_words" \
+	--dataset_config_name="el,el_gr,el_wav" \
 	--language="greek" \
-	--train_split_name="train+validation" \
-	--eval_split_name="test" \
+	--train_split_name="train+validation,train+validation,train+validation" \
+	--eval_split_name="test,-,-" \
 	--model_index_name="Whisper medium Greek El Greco" \
-	--text_column_name="keyword" \
+	--text_column_name="sentence,transcription,keyword" \
+	--streaming="False" \
+	--ignore_data_skip \
+	--resume_from_checkpoint="checkpoint-3000" \
 	--torch_compile="True" \
 	--torch_compile_mode="reduce-overhead" \
 	--torch_compile_mode="max-autotune" \
 	--logging_steps="25" \
-	--learning_rate="5e-6" \
+	--learning_rate="4e-6" \
 	--max_steps="5000" \
 	--output_dir="./" \
 	--per_device_train_batch_size="32" \
@@ -39,7 +42,6 @@ deepspeed run-ba.py \
 	--do_eval \
 	--predict_with_generate \
 	--do_normalize_eval \
-	--streaming="false" \
 	--use_auth_token \
 	--push_to_hub
 

runs/Dec21_16-27-53_129-146-176-120/events.out.tfevents.1671640138.129-146-176-120.405220.0

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f2a934041ca11bdc3b376be8151aa5626d848410d8e220cc849448be4d9a3815
-size 17481
+oid sha256:86c0ffc654bbe56e7927cf596d577f5b3114e2a85a6f270d58d20ce60554cb36
+size 24079