[SpeechEncoderDecoder] Fix from pretrained (huggingface#15043)

stevhliu · Jan 6, 2022 · e6eef10 · e6eef10
1 parent 68b95f0
commit e6eef10
Showing 1 changed file with 4 additions and 4 deletions.
diff --git a/src/transformers/models/speech_encoder_decoder/modeling_speech_encoder_decoder.py b/src/transformers/models/speech_encoder_decoder/modeling_speech_encoder_decoder.py
@@ -380,7 +380,7 @@ def from_encoder_decoder_pretrained(
                 )
 
             if "config" not in kwargs_encoder:
-                encoder_config = AutoConfig.from_pretrained(encoder_pretrained_model_name_or_path, **kwargs_encoder)
+                encoder_config = AutoConfig.from_pretrained(encoder_pretrained_model_name_or_path)
                 if encoder_config.is_decoder is True or encoder_config.add_cross_attention is True:
                     logger.info(
                         f"Initializing {encoder_pretrained_model_name_or_path} as a encoder model "
@@ -391,7 +391,7 @@ def from_encoder_decoder_pretrained(
 
                 kwargs_encoder["config"] = encoder_config
 
-            encoder = AutoModel.from_pretrained(encoder_pretrained_model_name_or_path, *model_args)
+            encoder = AutoModel.from_pretrained(encoder_pretrained_model_name_or_path, *model_args, **kwargs_encoder)
 
         decoder = kwargs_decoder.pop("model", None)
         if decoder is None:
@@ -402,7 +402,7 @@ def from_encoder_decoder_pretrained(
                 )
 
             if "config" not in kwargs_decoder:
-                decoder_config = AutoConfig.from_pretrained(decoder_pretrained_model_name_or_path, **kwargs_decoder)
+                decoder_config = AutoConfig.from_pretrained(decoder_pretrained_model_name_or_path)
                 if decoder_config.is_decoder is False or decoder_config.add_cross_attention is False:
                     logger.info(
                         f"Initializing {decoder_pretrained_model_name_or_path} as a decoder model. "
@@ -424,7 +424,7 @@ def from_encoder_decoder_pretrained(
                     "`decoder_config` to `.from_encoder_decoder_pretrained(...)`"
                 )
 
-            decoder = AutoModelForCausalLM.from_pretrained(decoder_pretrained_model_name_or_path)
+            decoder = AutoModelForCausalLM.from_pretrained(decoder_pretrained_model_name_or_path, **kwargs_decoder)
 
         # instantiate config with corresponding kwargs
         config = SpeechEncoderDecoderConfig.from_encoder_decoder_configs(encoder.config, decoder.config, **kwargs)