Upload folder using huggingface_hub

README.md

@@ -18,9 +18,11 @@ A CPU-based Hugging Face Space for generating music using [Stability AI's Stable
 - **Model**: `small-music` (433M parameters)
 - **Duration**: 1–120 seconds
 - **Sample rate**: 44.1 kHz stereo
-- **Runs on CPU** (no GPU required)
+- **Runs on CPU** (only a 1s ZeroGPU touch at startup; no GPU quota used per request)
 - Gradio 6.3.0 interface
 
+This Space is configured for CPU inference. ZeroGPU Spaces fail to boot unless at least one `spaces.GPU` function is called, so `app.py` calls one minimal `@spaces.GPU(duration=1)` function a single time at startup. Keep every other function (especially generation) undecorated so no GPU allocation happens per request.
+
 ## Usage
 
 1. Enter a text prompt describing the music you want

app.py

@@ -13,21 +13,22 @@ if hf_token:
     login(token=hf_token)
 
 
-# Required for ZeroGPU Spaces - must have at least one @spaces.GPU function
+# ZeroGPU Spaces refuse to start if no @spaces.GPU function is ever called, but
+# every call into a GPU function burns quota. So call this exactly ONCE at boot
+# with the smallest budget (1s), and keep all real generation on CPU below
+# (NOT decorated). That way quota is only touched here, never per request.
 @spaces.GPU(duration=1)
-def _gpu_startup_check():
-    return "GPU check passed"
+def _gpu_startup_touch():
+    return "ok"
 
 
-# Run the check at startup
-_gpu_startup_check()
+_gpu_startup_touch()
 
 
 # Model cache
 MODEL_CACHE = {}
 
 
-@spaces.GPU(duration=1)
 def load_model(model_name):
     """Load model on demand and cache it."""
     if model_name not in MODEL_CACHE: