JaheimLee
JaheimLee
By the way, I tried to install sherpa using conda, and it did work, even running on GPU. But you said conda installation of sherpa is to be done according...
> By the way, I just tried > > ``` > conda install -c k2-fsa -c kaldifeat -c pytorch cpuonly k2 kaldifeat pytorch=1.12.0 python=3.8 > pip install --verbose k2-sherpa >...
I also get this warning. But my issue is, when fine-tuning bert model , `new_lr` is always smaller than the `sefl.min_lf`.
> > The 'shape' in ReshapeOp is invalid. The input tensor X'size must be equal to the capacity of 'shape'. But received X's shape = [1, 16, 512, 19], X's...
@zh794390558 我在使用streaming_asr_server时，从长音频识别的文本到达一定长度后，client receive msg会被清空，重新以当前时间点开始返回后续文本，之前的就不见了，这个在哪可以设置啊
> The issue with the path seems to be very similar to the problems encountered with using Git LFS. See #14. They have similar issues and similar solutions. > >...
把register_buffer的persistent改为False好像就可以了
> > > > 修改了哪里？ https://huggingface.co/fnlp/moss-moon-003-sft/blob/7119d446173035561f40977fb9cb999995bb7517/modeling_moss.py#L63
It seems that '\n\n' should be the eos_token. @sgugger
And according to [this](https://huggingface.co/BlinkDL/rwkv-4-world), '\n\n' will be a single token in the new world model.