Chengdong Liang
Chengdong Liang
I use two GPUs. I solved this problem by updating the pytorch version. Now, there is no problem with single GPU now.
@pengzhendong can you follow this?
> > @pengzhendong can you follow this? > > 带娃比较忙，有点慢，你能跟进更好。改改 wekws 好的，我来搞下 :)
> Any updates on this? Can we do on-device speaker verification on mobile devices? Coming soon. However, if you have already implemented it, welcome PR.
I don't think it can be supported in UIO mode. But in [dataset_deprecated.py](https://github.com/wenet-e2e/wespeaker/blob/master/wespeaker/dataset/dataset_deprecated.py), it may be implemented.
Good job. It is valid during the initial stage. When I ran 100 data, the results of the version before the fixing are as follows. (cpu run) `I0228 09:51:54.538599 149817...
@Sang-Hoon-Pakr Yes, I used the same test method.
You can use another Python package: [Pynini](https://www.openfst.org/twiki/bin/view/GRM/Pynini) > The latest version on PyPI is 1.7.3, which is very old and only supports up to Python 3.7. Is it possible to...
Maybe try temperature=0.
> @cdliang11 @JiJiJiang Hi Chendong and Hongji, can you help to check whether it's the problem of the onnx export? Thanks `voxblink2_samresnet34.onnx` model has a fixed input shape, so we...