Jee Jee Li

Results 248 comments of Jee Jee Li

Yes, no problem. So, the dockerfile provides `modelscope`. And for other deployment way, error messages guide users to install `modelscope`. We can capture the ImportError similar to how it's done...

I don't have any bias, I'm just describing the current situation. @DarkLight1337 WDYT

which version are you using? It looks like your version is outdated.

I can generate reasonable results by using the latest main branch with `gemma-2-9b`. I think you can upgrade vllm to 0.7.3, then try it again

I cannot reproduce the result you reported. I suspect it might be due to the influence of prefix caching. The impact of LoRA(just using `--enable-lora`) is not very significant. ![image](https://github.com/user-attachments/assets/67af4283-8c9f-4f83-bce8-c03fed8a4576)

@badrjd I didn't use the above image for testing. I built it locally based on the main branch. Maybe you could try that.

> the same problem @cmccxll Are you also using the above image for testing?

@badrjd Has this issue been resolved for you? If not, you can try adding `--max-seq-len-to-capture 48000`.It's most likely due to this reason