Training in progress, step 1000

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fe6c3c8428592f02092883c1cc0b29af6a816cb82a70d648ac0e5224b13512e7
+oid sha256:61134984f61eb3028cafec42304da7950a647b6d5cb334923b69f68f3c7cd5a5
 size 3055754841

run_mgb2.sh

@@ -6,7 +6,7 @@ python run_speech_recognition_seq2seq_mixed_mgb2.py \
 --train_split_name="train" \
 --eval_split_name="test" \
 --model_index_name="Whisper Small Arabic" \
---max_steps="20000" \
+--max_steps="10000" \
 --output_dir="./" \
 --per_device_train_batch_size="32" \
 --per_device_eval_batch_size="16" \

run_speech_recognition_seq2seq_mixed_mgb2.py

@@ -519,6 +519,8 @@ def main():
             "Make sure that `config.decoder_start_token_id` is correctly defined"
         )
 
+    max_label_length = model.config.max_length
+
     if model_args.freeze_feature_encoder:
         model.freeze_feature_encoder()
 
@@ -608,6 +610,12 @@ def main():
         input_columns=["input_length"],
     )
 
+    def filter_labels(labels):
+        """Filter label sequences longer than max length"""
+        return len(labels) < max_label_length
+
+    vectorized_datasets = vectorized_datasets.filter(filter_labels, input_columns=["labels"])
+    
     # 8. Load Metric
     metric = evaluate.load("wer")
     do_normalize_eval = data_args.do_normalize_eval

runs/Dec13_21-37-24_129-146-107-47/events.out.tfevents.1670967464.129-146-107-47.73685.0

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2b539ae9b19b47c25c56516612de0677cca124d183bd364ed80f33facf5de9ab
-size 10853
+oid sha256:6929a1e6ec09a1488c1260d4569f6949128fefdc1c46aca7ec8f6cc6b79ce565
+size 15877

runs/Dec14_09-02-25_129-146-107-47/1671008564.670491/events.out.tfevents.1671008564.129-146-107-47.118226.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9fafa28f043df7798f0d1728f6085d9212e70e0e2ee3441b3ccce19079da2457
+size 5870

runs/Dec14_09-02-25_129-146-107-47/events.out.tfevents.1671008564.129-146-107-47.118226.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4bef7bb4cdd63395e5293eb15188a1801d9dee7f495606eb2f6f54520ee2ccbc
+size 10853

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:660602fd31ec8bd0e852fc8fc2e58c5fce5c255ed3830a543bb8adcecd582e76
+oid sha256:a6ca2b5eecde3bf2c0aa481033e37958875739c493ad97e49782abc08fe32fe5
 size 3579