Switch to Docker SDK for full ASGI control

app.py

@@ -7,6 +7,11 @@ import torch
 import torchaudio
 import gradio as gr
 import spaces
+from fastapi import FastAPI
+from fastapi.responses import JSONResponse
+from einops import rearrange
+from huggingface_hub import login
+from stable_audio_3 import StableAudioModel
 
 # Hugging Face Spaces should use UTF-8; explicitly configure streams so model
 # libraries cannot inherit a platform-specific charmap encoding.
@@ -14,11 +19,6 @@ for _stream in (sys.stdout, sys.stderr):
     if hasattr(_stream, "reconfigure"):
         _stream.reconfigure(encoding="utf-8", errors="backslashreplace")
 
-from fastapi.responses import JSONResponse
-from einops import rearrange
-from huggingface_hub import login
-from stable_audio_3 import StableAudioModel
-
 # Authenticate with gated model when HF_TOKEN secret is present
 hf_token = os.environ.get("HF_TOKEN")
 if hf_token:
@@ -156,6 +156,23 @@ def generate_audio(prompt, duration, steps, cfg_scale, seed, model_name):
     return output_path
 
 
+# ---------------------------------------------------------------------------
+# FastAPI application
+# ---------------------------------------------------------------------------
+
+api = FastAPI(title=API_SPECS["name"], version=API_SPECS["version"])
+
+
+@api.get("/api/resources")
+def resources():
+    return JSONResponse({"resources": API_RESOURCES})
+
+
+@api.get("/api/specs")
+def specs():
+    return JSONResponse({**API_SPECS, "runtime": _get_runtime_specs()})
+
+
 # ---------------------------------------------------------------------------
 # Gradio UI
 # ---------------------------------------------------------------------------
@@ -200,20 +217,9 @@ with gr.Blocks(title="Respite API") as demo:
         outputs=audio_output,
     )
 
-
-# ---------------------------------------------------------------------------
-# Mount API discovery routes directly onto Gradio's FastAPI app.
-# This avoids conflicts with Gradio's internal /api proxy.
-# ---------------------------------------------------------------------------
-
 demo.queue(max_size=4, default_concurrency_limit=1)
 
-
-@demo.app.get("/api/resources")
-def resources():
-    return JSONResponse({"resources": API_RESOURCES})
-
-
-@demo.app.get("/api/specs")
-def specs():
-    return JSONResponse({**API_SPECS, "runtime": _get_runtime_specs()})
+# Mount Gradio UI under the FastAPI app.
+# Routes are defined on `api` first, then Gradio is mounted at "/".
+# FastAPI evaluates routes top-down, so /api/* routes are served directly.
+app = gr.mount_gradio_app(api, demo, path="/")