Mac M2 Max 源码运行如何调用gpu

[Oceannew]

我修改了funasr_wss_server.py文件的配置，把cpu改为mps并没有效果。

我在google上没有找到解决方法。

修改的代码：

parser.add_argument("--ngpu", type=int, default=1, help="0 for cpu, 1 for gpu")
parser.add_argument("--device", type=str, default="mps", help="cuda, cpu")
parser.add_argument("--ncpu", type=int, default=10, help="cpu cores")

environment：

• OS (e.g., Linux):MAC
• FunASR Version (e.g., 1.0.0):1.0.18
• ModelScope Version (e.g., 1.11.0):1.14.0
• PyTorch Version (e.g., 2.0.0):2.3.0
• How you installed funasr (pip, source):source
• Python version:3.8.0
• GPU (e.g., V100M32):M2
• CUDA/cuDNN version (e.g., cuda11.7):none
• Docker version (e.g., funasr-runtime-sdk-cpu-0.4.1)
• Any other relevant information:

[LauraGPT]

We don't know M2, you could ref to torch. funasr is torch-based.

[XUJiahua]

可以看下这段代码：

FunASR/funasr/auto/auto_model.py

Line 184 in 1c52e36

device = "cpu"

指定了 mps，也会被回退到 cpu。

除了 spk_model 不兼容，其他模型都能在 mps 下跑。

[AlienKevin]

I can confirm that SenseVoiceSmall is runnable on mps, after commenting out the lines that coerse the device to CPU.
@LauraGPT Could you please remove the device coersion lines? I was extremely confused at first because I specified the device as MPS but saw no GPU usage on my Mac. If you are worried about device compatability for certain model-device pairs, maybe a warning message is more appropriate/helpful.

[LauraGPT]

I can confirm that SenseVoiceSmall is runnable on mps, after commenting out the lines that coerse the device to CPU. @LauraGPT Could you please remove the device coersion lines? I was extremely confused at first because I specified the device as MPS but saw no GPU usage on my Mac. If you are worried about device compatability for certain model-device pairs, maybe a warning message is more appropriate/helpful.

Fine. Do you know how to detect running on mps?

[AlienKevin]

@LauraGPT You can use torch.backends.mps.is_available() to detect if the MPS backend is supported. And the device name for MPS is simply the string mps.