Kaisheng Xu

Results 14 comments of Kaisheng Xu

> That's maybe some confusions from the second page. That's indeed a same page what ever your choice. But the differnt behavior will triggered when tap the "Next" button. >...

Could I suggest we just give user three input boxes to input username, password, and server url if choosing "Login". When they typing server url, there can be a drop-down...

I hit same issue and eventually I found that I need to first deploy a model on Azure portal. Then use parameter "engine" rather than "model" and pass the deployed...

Hey, this feature is very useful to me too. Seems there're not update since 2022, I will try make a PR for it. Just FYI. 😀

I have similar problem now, using CDivTable. It seems the forward pass is ok. This error appears at backward pass.

I tested 7B model with batch size 2 and it takes 15-17GB of GPU memory. The `example.py` use a much larger batch size and may need 40GB GPU memory.

Mainland china and HK is blocked as I tested. Same link works for JP & US.

OpenAI API is the de facto standard in this area. That's why I start this project to call custom LLM in OpenAI way. Any suggestion?