Merge pull request #298 from MuyleangIng/mac-mps-gradio-support

Add Mac MPS support for Gradio app

.gitignore

@@ -1,6 +1,7 @@
 launch.json
+.venv/
 __pycache__
 voxcpm.egg-info
 .DS_Store
 ./pretrained_models/
-app_local.py
+app_local.py

README.md

@@ -242,6 +242,14 @@ voxcpm --help
 python app.py --port 8808  # then open in browser: http://localhost:8808
 ```
 
+Use `--device` to choose the runtime device:
+
+```bash
+python app.py --device auto
+```
+
+Supported values are `auto`, `cpu`, `mps`, `cuda`, and `cuda:N`. On Apple Silicon Macs, `auto` uses MPS when available.
+
 ### 🚢 Production Deployment (Nano-vLLM)
 
 For high-throughput serving, use [**Nano-vLLM-VoxCPM**](https://github.com/a710128/nanovllm-voxcpm) — a dedicated inference engine built on Nano-vLLM with concurrent request support and an async API.

README_zh.md

@@ -241,6 +241,14 @@ voxcpm --help
 python app.py --port 8808  # 然后在浏览器打开 http://localhost:8808
 ```
 
+使用 `--device` 选择运行设备：
+
+```bash
+python app.py --device auto
+```
+
+支持的取值包括 `auto`、`cpu`、`mps`、`cuda` 和 `cuda:N`。在 Apple Silicon Mac 上，`auto` 会在可用时使用 MPS。
+
 ### 🚢 生产部署（Nano-vLLM）
 
 如需高吞吐量部署，使用 [**Nano-vLLM-VoxCPM**](https://github.com/a710128/nanovllm-voxcpm) — 基于 Nano-vLLM 构建的专用推理引擎，支持并发请求和异步 API。

app.py

@@ -3,7 +3,6 @@ import re
 import sys
 import logging
 import numpy as np
-import torch
 import gradio as gr
 from typing import Optional, Tuple
 from funasr import AutoModel
@@ -12,6 +11,7 @@ from pathlib import Path
 os.environ["TOKENIZERS_PARALLELISM"] = "false"
 
 import voxcpm
+from voxcpm.model.utils import resolve_runtime_device
 
 logging.basicConfig(
     level=logging.INFO,
@@ -220,17 +220,14 @@ _APP_THEME = gr.themes.Soft(
 # ---------- Model ----------
 
 class VoxCPMDemo:
-    def __init__(self, model_id: str = "openbmb/VoxCPM2") -> None:
-        self.device = "cuda" if torch.cuda.is_available() else "cpu"
-        logger.info(f"Running on device: {self.device}")
+    def __init__(self, model_id: str = "openbmb/VoxCPM2", device: str = "auto") -> None:
+        self.device = resolve_runtime_device(device, "cuda")
+        logger.info(f"Running VoxCPM on device: {self.device}")
+        self.optimize = self.device.startswith("cuda")
 
         self.asr_model_id = "iic/SenseVoiceSmall"
-        self.asr_model: Optional[AutoModel] = AutoModel(
-            model=self.asr_model_id,
-            disable_update=True,
-            log_level="DEBUG",
-            device="cuda:0" if self.device == "cuda" else "cpu",
-        )
+        self.asr_device = "cuda:0" if self.device.startswith("cuda") else "cpu"
+        self.asr_model: Optional[AutoModel] = None
 
         self.voxcpm_model: Optional[voxcpm.VoxCPM] = None
         self._model_id = model_id
@@ -239,14 +236,37 @@ class VoxCPMDemo:
         if self.voxcpm_model is not None:
             return self.voxcpm_model
         logger.info(f"Loading model: {self._model_id}")
-        self.voxcpm_model = voxcpm.VoxCPM.from_pretrained(self._model_id, optimize=True)
+        self.voxcpm_model = voxcpm.VoxCPM.from_pretrained(
+            self._model_id,
+            optimize=self.optimize,
+            device=self.device,
+        )
         logger.info("Model loaded successfully.")
         return self.voxcpm_model
 
+    def get_or_load_asr_model(self) -> AutoModel:
+        if self.asr_model is not None:
+            return self.asr_model
+        logger.info(
+            f"Loading ASR model: {self.asr_model_id} on device: {self.asr_device}"
+        )
+        self.asr_model = AutoModel(
+            model=self.asr_model_id,
+            disable_update=True,
+            log_level="DEBUG",
+            device=self.asr_device,
+        )
+        logger.info("ASR model loaded successfully.")
+        return self.asr_model
+
     def prompt_wav_recognition(self, prompt_wav: Optional[str]) -> str:
         if prompt_wav is None:
             return ""
-        res = self.asr_model.generate(input=prompt_wav, language="auto", use_itn=True)
+        res = self.get_or_load_asr_model().generate(
+            input=prompt_wav,
+            language="auto",
+            use_itn=True,
+        )
         return res[0]["text"].split("|>")[-1]
 
     def _build_generate_kwargs(
@@ -487,8 +507,9 @@ def run_demo(
     server_port: int = 8808,
     show_error: bool = True,
     model_id: str = "openbmb/VoxCPM2",
+    device: str = "auto",
 ):
-    demo = VoxCPMDemo(model_id=model_id)
+    demo = VoxCPMDemo(model_id=model_id, device=device)
     interface = create_demo_interface(demo)
     interface.queue(max_size=10, default_concurrency_limit=1).launch(
         server_name=server_name,
@@ -508,5 +529,11 @@ if __name__ == "__main__":
         help="Local path or HuggingFace repo ID (default: openbmb/VoxCPM2)",
     )
     parser.add_argument("--port", type=int, default=8808, help="Server port")
+    parser.add_argument(
+        "--device",
+        type=str,
+        default="auto",
+        help="Runtime device: auto, cpu, mps, cuda, or cuda:N (default: auto)",
+    )
     args = parser.parse_args()
-    run_demo(model_id=args.model_id, server_port=args.port)
+    run_demo(model_id=args.model_id, server_port=args.port, device=args.device)

tests/test_cli.py

@@ -54,6 +54,11 @@ def run_main(monkeypatch, argv):
     cli.main()
 
 
+def patch_soundfile_write(monkeypatch):
+    soundfile_stub = types.SimpleNamespace(write=lambda *args, **kwargs: None)
+    monkeypatch.setitem(sys.modules, "soundfile", soundfile_stub)
+
+
 def test_parser_defaults_to_voxcpm2():
     parser = cli._build_parser()
     args = parser.parse_args(["design", "--text", "hello", "--output", "out.wav"])
@@ -70,7 +75,7 @@ def test_load_model_respects_no_optimize_for_local_model(monkeypatch):
             calls["kwargs"] = kwargs
             self.tts_model = DummyTTSModel()
 
-    monkeypatch.setattr(cli, "VoxCPM", FakeVoxCPM)
+    monkeypatch.setattr(core_stub, "VoxCPM", FakeVoxCPM)
     args = cli._build_parser().parse_args(
         [
             "design",
@@ -99,7 +104,7 @@ def test_load_model_defaults_optimize_for_hf(monkeypatch):
             calls["kwargs"] = kwargs
             return DummyModel()
 
-    monkeypatch.setattr(cli, "VoxCPM", FakeVoxCPM)
+    monkeypatch.setattr(core_stub, "VoxCPM", FakeVoxCPM)
     args = cli._build_parser().parse_args(
         [
             "design",
@@ -125,7 +130,7 @@ def test_load_model_respects_no_optimize_for_hf(monkeypatch):
             calls["kwargs"] = kwargs
             return DummyModel()
 
-    monkeypatch.setattr(cli, "VoxCPM", FakeVoxCPM)
+    monkeypatch.setattr(core_stub, "VoxCPM", FakeVoxCPM)
     args = cli._build_parser().parse_args(
         [
             "design",
@@ -152,7 +157,7 @@ def test_load_model_passes_explicit_device_to_hf(monkeypatch):
             calls["kwargs"] = kwargs
             return DummyModel()
 
-    monkeypatch.setattr(cli, "VoxCPM", FakeVoxCPM)
+    monkeypatch.setattr(core_stub, "VoxCPM", FakeVoxCPM)
     args = cli._build_parser().parse_args(
         [
             "design",
@@ -173,7 +178,7 @@ def test_load_model_passes_explicit_device_to_hf(monkeypatch):
 def test_design_subcommand_applies_control(monkeypatch, tmp_path):
     dummy_model = DummyModel()
     monkeypatch.setattr(cli, "load_model", lambda args: dummy_model)
-    monkeypatch.setattr(cli.sf, "write", lambda *args, **kwargs: None)
+    patch_soundfile_write(monkeypatch)
 
     run_main(
         monkeypatch,
@@ -201,7 +206,7 @@ def test_clone_subcommand_reads_prompt_file(monkeypatch, tmp_path):
     prompt_file.write_text("prompt transcript\n", encoding="utf-8")
 
     monkeypatch.setattr(cli, "load_model", lambda args: dummy_model)
-    monkeypatch.setattr(cli.sf, "write", lambda *args, **kwargs: None)
+    patch_soundfile_write(monkeypatch)
 
     run_main(
         monkeypatch,
@@ -273,7 +278,7 @@ def test_clone_rejects_reference_audio_for_v1_hf_model_id(monkeypatch, tmp_path)
 def test_legacy_root_args_still_work_and_warn(monkeypatch, tmp_path, capsys):
     dummy_model = DummyModel()
     monkeypatch.setattr(cli, "load_model", lambda args: dummy_model)
-    monkeypatch.setattr(cli.sf, "write", lambda *args, **kwargs: None)
+    patch_soundfile_write(monkeypatch)
 
     run_main(
         monkeypatch,
@@ -296,7 +301,7 @@ def test_batch_subcommand_applies_control(monkeypatch, tmp_path):
     input_file.write_text("hello\nworld\n", encoding="utf-8")
 
     monkeypatch.setattr(cli, "load_model", lambda args: dummy_model)
-    monkeypatch.setattr(cli.sf, "write", lambda *args, **kwargs: None)
+    patch_soundfile_write(monkeypatch)
 
     run_main(
         monkeypatch,
@@ -325,7 +330,7 @@ def test_legacy_clone_with_prompt_file_still_works(monkeypatch, tmp_path, capsys
     prompt_file.write_text("legacy transcript", encoding="utf-8")
 
     monkeypatch.setattr(cli, "load_model", lambda args: dummy_model)
-    monkeypatch.setattr(cli.sf, "write", lambda *args, **kwargs: None)
+    patch_soundfile_write(monkeypatch)
 
     run_main(
         monkeypatch,