update

app.py

@@ -42,7 +42,7 @@ NUM_SAMPLE = 1
 # a = AutoModel.from_pretrained('bert-base-uncased')
 # b = AutoModel.from_pretrained('roberta-base')
 
-snapshot_download(repo_id="AndreasXi/MeanAudio", local_dir="./weights",allow_patterns=["*.pt", "*.pth"] )
+# snapshot_download(repo_id="AndreasXi/MeanAudio", local_dir="./weights",allow_patterns=["*.pt", "*.pth"] )
 # _clap_ckpt_path='./weights/music_speech_audioset_epoch_15_esc_89.98.pt'
 # laion_clap_model = laion_clap.CLAP_Module(enable_fusion=False, amodel='HTSAT-base').cuda().eval()
 
@@ -67,6 +67,11 @@ def generate_audio_gradio(
         raise ValueError(f"Unknown model variant: {variant}. Available: {list(all_model_cfg.keys())}")
 
     model_path = all_model_cfg[variant].model_path  # by default, this will use meanaudio_s_full.pth or fluxaudio_s_full.pth
+    if not model_path.exists():
+        log.info(f'Model not found at {model_path}')
+        log.info('Downloading models to "./weights/"...')
+        snapshot_download(repo_id="AndreasXi/MeanAudio", local_dir="./weights",allow_patterns=["*.pt", "*.pth"] )
+
     model = all_model_cfg[variant]
     seq_cfg = model.seq_cfg
     seq_cfg.duration = duration