refactor: remove dead ZeroGPU shim (Docker Spaces don't support ZeroGPU)

CLAUDE.md

@@ -22,7 +22,6 @@ Use the existing local git config; do not override `user.name` / `user.email` an
 
 ## Working preferences
 
-- Multi-platform code: must run on macOS (MPS), Linux (CUDA/CPU), Windows (CUDA/CPU), and HF Spaces (Free CPU).
-- ZeroGPU-ready: keep the `@spaces.GPU` decorator path working as a no-op locally.
+- Multi-platform code: must run on macOS (MPS), Linux (CUDA/CPU), Windows (CUDA/CPU), and HF Spaces (Free CPU; paid GPU tiers supported).
 - Server is **stateless**. All voice library and history persistence lives in browser IndexedDB. Do not add server-side DBs.
 - YAGNI: don't add features beyond what the spec calls for.

README.md

@@ -17,8 +17,8 @@ short_description: Voice cloning studio for the Chatterbox TTS family.
 
 A multi-platform browser-based voice cloning studio for the Chatterbox TTS family
 (English, Turbo, Multilingual). Runs locally on macOS (MPS), Linux (CUDA/CPU),
-and Windows (CUDA/CPU). Deploys to Hugging Face Spaces (Free CPU by default,
-ZeroGPU-decorator-ready).
+and Windows (CUDA/CPU). Deploys to Hugging Face Spaces (Docker SDK, Free CPU
+by default; paid GPU tiers supported).
 
 ## Quick start (local)
 
@@ -47,11 +47,9 @@ and opens the studio at http://127.0.0.1:7860.
 This repo's `Dockerfile` is what HF Spaces uses to build the image.
 On Free CPU it runs as-is — generation will be slow (30–90s per clip).
 
-To get GPU on Spaces:
-
-- Subscribe to HF Pro and pick "ZeroGPU" hardware in your Space settings.
-  No code change needed; the `@spaces.GPU` decorator activates.
-- Or pick a paid GPU (T4 small / A10G).
+To get GPU on Spaces, switch the Space hardware to a paid tier (T4 small,
+A10G, L4). ZeroGPU is **not** available on Docker Spaces — it's currently
+restricted to the Gradio SDK only.
 
 ## Environment variables
 

server/main.py

@@ -25,7 +25,6 @@ from server.dialog import (
 )
 from server.progress import get_bus
 from server.registry import Registry
-from server.zerogpu import decorate
 
 
 STATIC_DIR = Path(__file__).parent / "static"
@@ -169,12 +168,11 @@ def build_app() -> FastAPI:
             tmp.close()
             ref_path = tmp.name
 
-        gen_fn = decorate(adapter.generate)
         bus = get_bus()
         try:
             async with bus.session("single", total_turns=1) as sess:
                 wav_bytes, _sr, seed_used = await asyncio.to_thread(
-                    gen_fn, text, ref_path, language, json.loads(params or "{}"),
+                    adapter.generate, text, ref_path, language, json.loads(params or "{}"),
                 )
                 sess.set_seed(seed_used)
         except Exception as exc:

server/zerogpu.py

@@ -1,25 +0,0 @@
-"""ZeroGPU decorator shim.
-
-When `spaces` is importable (HF ZeroGPU runtime), `decorate` wraps
-functions with `spaces.GPU(duration=...)`. Otherwise it is the
-identity decorator. Local installs and Free CPU Spaces hit the
-no-op branch.
-"""
-from __future__ import annotations
-
-from typing import Callable, TypeVar
-
-
-F = TypeVar("F", bound=Callable)
-
-
-try:  # pragma: no cover — covered by a test that injects a fake module
-    import spaces  # type: ignore[import-not-found]
-
-    def decorate(fn: F) -> F:
-        return spaces.GPU(duration=120)(fn)  # type: ignore[no-any-return]
-
-except ImportError:  # local / Free CPU
-
-    def decorate(fn: F) -> F:
-        return fn

tests/test_zerogpu.py

@@ -1,33 +0,0 @@
-import sys
-from unittest.mock import MagicMock
-
-from server.zerogpu import decorate
-
-
-def test_decorate_is_passthrough_when_spaces_missing(monkeypatch):
-    monkeypatch.setitem(sys.modules, "spaces", None)
-
-    @decorate
-    def fn(x):
-        return x * 2
-
-    assert fn(3) == 6
-
-
-def test_decorate_uses_spaces_gpu_when_available(monkeypatch):
-    fake_spaces = MagicMock()
-    fake_decorator = MagicMock(side_effect=lambda f: f)
-    fake_spaces.GPU = MagicMock(return_value=fake_decorator)
-    monkeypatch.setitem(sys.modules, "spaces", fake_spaces)
-
-    import importlib
-    import server.zerogpu as zg
-
-    importlib.reload(zg)
-
-    @zg.decorate
-    def fn(x):
-        return x + 1
-
-    assert fn(2) == 3
-    fake_spaces.GPU.assert_called_once()