update 3

run_image_captioning_flax_reduced.py

@@ -504,6 +504,10 @@ def main():
             "Decoder Config: Either a pretrained config or a model location for decoder is required."
         )
 
+    # GPT2 only has eos token but not pad token
+    if decoder_config.pad_token_id is None:
+        decoder_config.pad_token_id = decoder_config.eos_token_id
+
     config = VisionEncoderDecoderConfig.from_encoder_decoder_configs(encoder_config, decoder_config)
     model = FlaxVisionEncoderDecoderModel.from_encoder_decoder_pretrained(
         encoder_pretrained_model_name_or_path=model_args.encoder_model_name_or_path,
@@ -544,7 +548,7 @@ def main():
             "You are instantiating a new tokenizer from scratch. This is not supported by this script."
             "You can do it from another script, save it, and load it from here, using --tokenizer_name."
         )
-    tokenizer.pad_token = tokenizer.convert_ids_to_tokens(config.pad_token_id)
+    tokenizer.pad_token = tokenizer.convert_ids_to_tokens(config.decoder.pad_token_id)
 
     # Preprocessing the datasets.
     # We need to tokenize inputs and targets.