Fix Content-Length: Remove response_class, let FastAPI handle headers automatically, add defensive checks

app.py

@@ -485,23 +485,41 @@ with gr.Blocks(
 
 # Add API routes directly to Gradio's FastAPI app
 # These routes are added after Gradio Blocks context but before queue/launch
-from fastapi.responses import JSONResponse
+# Let FastAPI handle Content-Length automatically - return dicts, not Response objects
 from fastapi import Request
 
-@gradio_app.app.get("/health", response_class=JSONResponse)
+@gradio_app.app.get("/health")
 def api_health():
-    """API health check endpoint."""
-    return healthcheck()
+    """
+    API health check endpoint.
+    Returns JSON dict - FastAPI will serialize and set Content-Length correctly.
+    Never returns 204/304 - always returns 200 with JSON body.
+    """
+    result = healthcheck()
+    # Ensure we always return a dict (never empty/None) to avoid 204 issues
+    if not result:
+        result = {"status": "ok"}
+    return result
 
-@gradio_app.app.post("/v1/generate", response_class=JSONResponse)
+@gradio_app.app.post("/v1/generate")
 async def api_generate(request: Request):
-    """API generate endpoint."""
+    """
+    API generate endpoint.
+    Returns JSON dict - FastAPI will serialize and set Content-Length correctly.
+    Never returns 204/304 - always returns 200 with JSON body or 500 error.
+    """
     try:
         data = await request.json()
         payload = GeneratePayload(**data)
         result = generate_endpoint(payload)
+        # Return dict directly - FastAPI will serialize and set Content-Length correctly
+        # Ensure we always return a dict with content (never empty)
         return {"text": result.text}
+    except HTTPException:
+        # Re-raise HTTPException as-is - FastAPI handles Content-Length correctly
+        raise
     except Exception as exc:
+        # HTTPException will be handled by FastAPI with proper Content-Length
         raise HTTPException(status_code=500, detail=str(exc))
 
 # Call warm start