Sangchun Ha comments

Results 35 comments of


                                            Sangchun Ha

Is there an easy way to set model max length to 10 or 20 seconds?

I referred and confirmed issue #66. However, if max padding is done with 10 seconds or 20 seconds using a preprocessor, an error occurs in the model. @sanchit-gandhi

Is there an easy way to set model max length to 10 or 20 seconds?

I even loaded it like this, but when I put the dummy input, the following error occurs. `UnfilteredStackTrace: TypeError: float() argument must be a string or a number, not 'ShapeDtypeStruct`...

squeezeformer 모델 학습 및 추론

KsponSpeech로 모델을 쉽게 학습해볼 수 있는 [SpeechBrain 코드](https://github.com/speechbrain/speechbrain/tree/develop/recipes/KsponSpeech) 를 참고해보시면 좋을 것 같습니다.

[Bug]: Gemma2 supports 8192 context with sliding window, but vllm only does 4196 or fails if try 8192

Even if `flashinfer==0.1.3` is installed, the vllm code does not support it. https://github.com/vllm-project/vllm/blob/v0.5.4/vllm/config.py#L163-L171 ``` pip install https://github.com/flashinfer-ai/flashinfer/releases/download/v0.1.3/flashinfer-0.1.3+cu121torch2.3-cp310-cp310-linux_x86_64.whl ``` @noamgat Is there another way to use **8k context window?**

`OpenspeechCTCModel.forward`: `self.fc` is undefined

@kuraga It should be defined as below link, but there seems to be a missing issue in the jasper model. https://github.com/openspeech-team/openspeech/blob/main/openspeech/models/contextnet/model.py#L71 https://github.com/openspeech-team/openspeech/blob/main/openspeech/models/conformer/model.py#L66