prog update

app.py

@@ -5,6 +5,14 @@ import httpx
 import json
 import base64
 from io import BytesIO
+
+# Fix for Python 3.13 audioop removal
+try:
+    import audioop
+except ImportError:
+    import audioop_lts as audioop
+    sys.modules["audioop"] = audioop
+
 from fastapi import Request
 from fastapi.responses import StreamingResponse, JSONResponse
 import uvicorn
@@ -13,7 +21,7 @@ from openai import OpenAI
 
 # --- CONFIGURATION ---
 MODEL_ID = "numind/NuMarkdown-8B-Thinking"
-GPU_UTILIZATION = 0.85 
+GPU_UTILIZATION = 0.95
 MAX_MODEL_LEN = 16384 
 VLLM_PORT = 8000
 HF_PORT = 7860
@@ -87,10 +95,15 @@ with gr.Blocks(title="NuMarkdown API") as demo:
 
 # --- STEP 3: ATTACH PROXY TO GRADIO'S APP ---
 # We enable the queue for long tasks
+# 1. FIX ATTRIBUTE ERROR: Patch missing attributes onto the demo object
+demo.max_file_size = 100 * 1024 * 1024 # 100MB
 demo.queue()
 
 # We get the FastAPI instance from Gradio
 app = demo.app
+# 3. Mount Gradio to FastAPI
+# Using path="" and assigning to the app ensures assets are at the root
+app = gr.mount_gradio_app(app, demo, path="/")
 
 # We add the external API proxy directly to this app
 @app.api_route("/v1/{path:path}", methods=["GET", "POST", "PUT", "DELETE"])