Training in progress, step 1000

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b0ebc286c87e2a415902d8a07c829f63b2f78cf488c217570b28c22a72408ae1
+oid sha256:63e0c0e93a2f2b61938f7a8adaf7456e978d2c89d12f937657a7c2dacb43e79b
 size 290403936

run.sh

@@ -4,15 +4,15 @@ python run_speech_recognition_seq2seq_streaming.py \
 	--dataset_name="asierhv/composite_corpus_eu_v2.1" \
 	--language="basque" \
 	--train_split_name="train" \
-	--eval_split_name="dev_parl+test_parl+test_cv+test_oslr" \
+	--eval_split_name="dev" \
 	--model_index_name="Whisper Base Basque" \
-	--max_steps="8000" \
+	--max_steps="10000" \
 	--output_dir="./" \
 	--per_device_train_batch_size="32" \
 	--per_device_eval_batch_size="16" \
 	--gradient_accumulation_steps="1" \
 	--logging_steps="25" \
-	--learning_rate="2e-5" \
+	--learning_rate="2.5e-5" \
 	--warmup_steps="500" \
 	--evaluation_strategy="steps" \
 	--eval_steps="1000" \
@@ -39,4 +39,4 @@ python run_speech_recognition_seq2seq_streaming.py \
 	--use_auth_token \
 	--push_to_hub \
 	--report_to "wandb" \
-	--run_name "whisper-base-eu"
+	--run_name "whisper-base-eu-25.02-r1"

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:919dfb9130e053a4aa40a748eefa2ccb8b73c9660013bdf2710bd7281862ae85
+oid sha256:9af7a40aa7a00bb6012dd0f11d83508aa945a32d238fb5ce6165110fcc8d49e9
 size 5432