try

meanaudio/model/utils/features_utils.py

@@ -63,8 +63,8 @@ class FeaturesUtils(nn.Module):
                 self.text_encoder = T5EncoderModel.from_pretrained('google/flan-t5-large').eval()
 
             elif encoder_name == 't5_clap' or encoder_name == 't5_clap_cat':
-                self.tokenizer = AutoTokenizer.from_pretrained('google/flan-t5-large')
-                self.text_encoder = T5EncoderModel.from_pretrained('google/flan-t5-large').eval()
+                self.tokenizer = AutoTokenizer.from_pretrained('google/flan-t5-large',revision="main")
+                self.text_encoder = T5EncoderModel.from_pretrained('google/flan-t5-large',revision="main").eval()
                 self.laion_clap_model = laion_clap.CLAP_Module(enable_fusion=False, amodel='HTSAT-base').eval()
                 self._clap_ckpt_path = "./weights/music_speech_audioset_epoch_15_esc_89.98.pt"  
                 self.laion_clap_model.load_ckpt(self._clap_ckpt_path, verbose=False)