Restore working structure with FastAPI + Gradio mount

app.py

@@ -7,16 +7,16 @@ import torch
 import torchaudio
 import gradio as gr
 import spaces
-
-for _stream in (sys.stdout, sys.stderr):
-    if hasattr(_stream, "reconfigure"):
-        _stream.reconfigure(encoding="utf-8", errors="backslashreplace")
-
+from fastapi import FastAPI
 from fastapi.responses import JSONResponse
 from einops import rearrange
 from huggingface_hub import login
 from stable_audio_3 import StableAudioModel
 
+for _stream in (sys.stdout, sys.stderr):
+    if hasattr(_stream, "reconfigure"):
+        _stream.reconfigure(encoding="utf-8", errors="backslashreplace")
+
 hf_token = os.environ.get("HF_TOKEN")
 if hf_token:
     login(token=hf_token)
@@ -45,6 +45,10 @@ def _get_ram_bytes():
     return None
 
 
+# ---------------------------------------------------------------------------
+# API metadata
+# ---------------------------------------------------------------------------
+
 API_RESOURCES = {
     "audio_generation": {
         "name": "Audio generation",
@@ -93,6 +97,10 @@ def _get_runtime_specs():
     }
 
 
+# ---------------------------------------------------------------------------
+# Model cache
+# ---------------------------------------------------------------------------
+
 MODEL_CACHE = {}
 
 
@@ -126,6 +134,27 @@ def generate_audio(prompt, duration, steps, cfg_scale, seed, model_name):
     return output_path
 
 
+# ---------------------------------------------------------------------------
+# FastAPI application with discovery routes
+# ---------------------------------------------------------------------------
+
+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
+# ---------------------------------------------------------------------------
+
 with gr.Blocks(title="Respite API") as demo:
     gr.Markdown("# Respite API - Music & SFX Generation")
     gr.Markdown(
@@ -160,5 +189,11 @@ with gr.Blocks(title="Respite API") as demo:
     )
 
 
-# This is the only thing that matters for the HF Gradio runner.
+# Mount Gradio under the FastAPI app and export as `app`.
+# HF Spaces Gradio runner finds the Blocks `demo` and calls demo.launch().
+# The `app` variable is here for direct uvicorn execution.
 demo.queue(max_size=4, default_concurrency_limit=1)
+app = gr.mount_gradio_app(api, demo, path="/")
+
+if __name__ == "__main__":
+    demo.launch()