revery cursor fallback strategies on zerogpu decorator

README.md

@@ -64,5 +64,5 @@ python download_bundles.py --inspect-only --infer   # inspect existing downloads
 
 ## Deployment
 
-- **Hugging Face Spaces:** Set hardware to **Zero GPU** (or standard GPU). The app uses `@spaces.GPU` when available. If Zero GPU workers fail with *"No CUDA GPUs are available"* (e.g. no free GPU in the pool), set the **CPU_FALLBACK** secret or environment variable to `1` so inference runs on CPU in-process instead of in GPU workers.
+- **Hugging Face Spaces:** Set hardware to **Zero GPU** (or standard GPU). Decorate GPU-dependent functions with `@spaces.GPU` per [Spaces ZeroGPU docs](https://huggingface.co/docs/hub/spaces-zerogpu).
 - **Local:** Run `python app.py`; requires a GPU with ~10 GB VRAM for all 6 models.

app.py

@@ -18,18 +18,7 @@ from ui_strings import get_strings
 STARTUP_EXAMPLE_PROMPT = "dites moi en plus sur les jardins japonnais a paris :"
 STARTUP_EXAMPLE_SYSTEM = ""
 
-# Optional: use @spaces.GPU for ZeroGPU deployment.
-# Set CPU_FALLBACK=1 (or any non-empty value) to skip ZeroGPU and run inference on CPU
-# when no GPU is available (avoids "RuntimeError: No CUDA GPUs are available" in workers).
-import os
-
-try:
-    import spaces
-
-    _use_cpu_fallback = os.environ.get("CPU_FALLBACK", "").strip().lower() in ("1", "true", "yes")
-    GPU_DECORATOR = (lambda f: f) if _use_cpu_fallback else spaces.GPU
-except ImportError:
-    GPU_DECORATOR = lambda f: f  # no-op when not on Spaces
+import spaces
 
 def build_params_by_model(
     temp_baguettotron: float,
@@ -64,7 +53,7 @@ def build_params_by_model(
     return params_by_model
 
 
-@GPU_DECORATOR
+@spaces.GPU
 def generate_all(
     prompt: str,
     system_prompt: str,