fix: torch_dtype->dtype, robust app launch for HF Spaces

app.py

@@ -17,6 +17,7 @@ from __future__ import annotations
 
 import json
 import logging
+import os
 from pathlib import Path
 
 # Ensure workspace exists
@@ -54,49 +55,59 @@ def main():
             else:
                 logger.info(f"MCP server '{name}' connected")
 
-    # Create the Gradio app
+    # Create the Gradio app with custom HTML
     from sonicoder.server import get_app, api_get_config
-    application = get_app()
 
-    # Inject runtime config into index.html and serve via custom route
+    # Build config JSON for frontend injection
     config_json = api_get_config()
     config_data = json.loads(config_json)
     config_str = json.dumps(config_data, ensure_ascii=False)
 
+    # Read index.html and inject config
     index_path = Path(__file__).parent / "static" / "index.html"
+    custom_html = None
     if index_path.exists():
         html = index_path.read_text(encoding="utf-8")
         html = html.replace(
             "__RUNTIME_CONFIG__ || {}",
             config_str + " || {}"
         )
+        custom_html = html
+        logger.info("Custom HTML prepared")
+
+    application = get_app()
 
-        # Serve custom index.html as the root page
-        from fastapi import Response
-        from fastapi.responses import HTMLResponse
+    # Register custom root route after app is built
+    if custom_html:
+        try:
+            from fastapi import Response
+            from fastapi.responses import HTMLResponse
 
-        # Gradio 4.x and 5.x: access the underlying FastAPI/Starlette app
-        underlying = getattr(application, "app", None) or getattr(application, "_queue", None)
-        if underlying is None:
-            # Fallback: try the internal app attribute
-            for attr_name in ("app", "server_app", "root_path"):
-                underlying = getattr(application, attr_name, None)
+            # Try multiple ways to access the underlying app in Gradio 6.x
+            underlying = None
+            for attr in ("app", "server_app", "_app", "app_obj"):
+                underlying = getattr(application, attr, None)
                 if underlying is not None:
                     break
 
-        if underlying is not None:
-            @underlying.get("/", response_class=HTMLResponse, include_in_schema=False)
-            async def custom_root():
-                return Response(content=html, media_type="text/html")
-            logger.info("Custom root route registered")
-        else:
-            logger.warning("Could not find underlying app for custom root route")
+            if underlying is not None and hasattr(underlying, "get"):
+                @underlying.get("/", response_class=HTMLResponse, include_in_schema=False)
+                async def serve_index():
+                    return Response(content=custom_html, media_type="text/html")
+                logger.info("Custom root route registered successfully")
+            else:
+                logger.warning("Could not register custom root route — using default Gradio UI")
+        except Exception as e:
+            logger.warning(f"Custom root route registration failed: {e}")
+
+    # Launch — HF Spaces sets PORT env var
+    port = int(os.environ.get("PORT", 7860))
+    logger.info(f"Launching web interface on port {port}")
 
-    logger.info("Launching web interface on http://0.0.0.0:7860")
     application.launch(
         show_error=True,
         server_name="0.0.0.0",
-        server_port=7860,
+        server_port=port,
         share=False,
     )
 

sonicoder/model/loader.py

@@ -89,7 +89,7 @@ def load_model(model_key: str) -> None:
             model_kwargs = {
                 "trust_remote_code": config_entry.trust_remote_code,
                 "low_cpu_mem_usage": True,
-                "torch_dtype": dtype,
+                "dtype": dtype,
             }
 
             if device == "cuda":