fixes

app.py

@@ -4,6 +4,9 @@ import numpy as np
 import torch
 import librosa
 import gradio as gr
+
+import matplotlib
+matplotlib.use("Agg")
 import matplotlib.pyplot as plt
 
 from huggingface_hub import hf_hub_download
@@ -71,16 +74,16 @@ def _fig_to_rgb_array(fig):
 
 @torch.inference_mode()
 def get_model():
-    # Cache model as a function attribute (simple + Space-friendly)
     if hasattr(get_model, "_model") and get_model._model is not None:
         return get_model._model
-
-    ckpt_path = hf_hub_download(repo_id=MODEL_REPO_ID, filename=CKPT_FILENAME)
-
-    model = VocosVocoderModule.load_from_checkpoint(ckpt_path, map_location="cpu")
-    model.eval()
-    get_model._model = model
-    return model
+    try:
+        ckpt_path = hf_hub_download(repo_id=MODEL_REPO_ID, filename=CKPT_FILENAME)
+        model = VocosVocoderModule.load_from_checkpoint(ckpt_path, map_location="cpu")
+        model.eval()
+        get_model._model = model
+        return model
+    except Exception as e:
+        raise RuntimeError(f"Model load failed: {e}")
 
 
 def mel_to_image(mel_80_t):
@@ -128,9 +131,13 @@ def run_reconstruct(audio_path, chunk_size):
     """
     if audio_path is None or str(audio_path).strip() == "":
         return None, None, None, None
+    
 
-    model = get_model()
-
+    try:
+        model = get_model()
+    except Exception as e:
+        return None, np.zeros((10,10,3), dtype=np.uint8), None, np.zeros((10,10,3), dtype=np.uint8)
+    
     # Load input audio
     x, _ = librosa.load(audio_path, sr=SAMPLE_RATE, mono=True)
 
@@ -202,5 +209,5 @@ with gr.Blocks() as demo:
     )
 
 demo.queue()
-demo.launch(server_name="0.0.0.0", server_port=7860)
+demo.launch(server_name="0.0.0.0", server_port=int(os.environ.get("PORT", 7860)))