为什么使用http api服务时，使用post_api.py发送参考音频并推理时出现cuda报错。

[eightfeetfish]

报错内容：
RuntimeError: CUDA error: device-side assert triggered
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

[AnyaCoder]

更新到1.4.2试试？

[beilyLan]

我更新用1.4.2的也有这个问题。
运行环境：
docker
cloud-T4 16G

堆栈与报错如下：
Traceback (most recent call last):
File "/usr/local/lib/python3.12/site-packages/kui/asgi/exceptions.py", line 27, in wrapper
return await endpoint()
File "/usr/local/lib/python3.12/site-packages/kui/asgi/views.py", line 29, in wrapper
return await function()
File "/usr/local/lib/python3.12/site-packages/kui/asgi/parameters.py", line 119, in callback_with_auto_bound_params
result = await result
File "/opt/fish-speech/tools/api.py", line 463, in api_invoke_model
fake_audios = next(inference(req))
File "/usr/local/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 36, in generator_context
response = gen.send(None)
File "/opt/fish-speech/tools/api.py", line 266, in inference
encode_reference(
File "/opt/fish-speech/tools/api.py", line 158, in encode_reference
audios = torch.from_numpy(reference_audio_content).to(decoder_model.device)[

RuntimeError: CUDA error: device-side assert triggered
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

[AnyaCoder]

就我而言，我不用docker，只在linux和windows环境下直接跑，没出现这样的问题。可能得问下冷月。 @leng-yue