Update modelling_musilingo.py

modelling_musilingo.py

@@ -23,7 +23,7 @@ from .configuration_musilingo import MusiLingoConfig, PATH
 import timm.models.hub as timm_hub
 
 
-from transformers import LlamaTokenizer, Wav2Vec2FeatureExtractor, AutoModel
+from transformers import LlamaTokenizer, Wav2Vec2FeatureExtractor, AutoModel, AutoConfig
 from transformers.activations import ACT2FN
 from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
 from transformers.modeling_utils import PreTrainedModel
@@ -2032,7 +2032,9 @@ class MusiLingo(BaseModel):
         self.low_resource = low_resource
 
         print('Loading Audio Encoder')
-        self.audio_encoder = AutoModel.from_pretrained(mert_model, trust_remote_code=True)
+        mert_modelconfig = AutoConfig.from_pretrained(mert_model, trust_remote_code=True)
+        mert_modelconfig.conv_pos_batch_norm = False
+        self.audio_encoder = AutoModel.from_pretrained(mert_model, config=mert_modelconfig,trust_remote_code=True)
         # loading the corresponding preprocessor config
         self.processor = Wav2Vec2FeatureExtractor.from_pretrained(mert_model, trust_remote_code=True)