Shawn/Yuxuan Tong

icon indicating a website link tongyx361.github.io icon indicating an email [email protected]

icon company Tsinghua University icon location Beijing, China icon location Senior undergraduate @ DCST, Tsinghua University. Research intern @hkust-nlp (previously: @THUDM). Interested in LLM & AI for Education/Research/Software Eng.

Results 15 issues of Shawn/Yuxuan Tong

What are the temperatures used for nucleus sampling in data synthesis?

The temperatures seem not mentioned neither in the paper or the OpenRiview.

Experimental results on ARC-C subset for challeging reasoning?

“Notably, except for ORPO, almost all approaches lead to consistent drops in one or more settings.” But ARC shows that commonsense reasoning is improved. However, the contradiction between reasoning tasks...

Doc error: line number is `None`

https://nbdev.fast.ai/api/doclinks.html#nbdevlookup.code:~:text=fastcore/net.py%23-,LNone,-%27 Why is the line number is `None` here but correct in rendered docs?

[Bug] `Meta-Llama-3.1-8B-Instruct` triggers "Detected errors during sampling! NaN in the probability." under high concurrency/RPS.

3 comment

### Checklist - [X] 1. I have searched related issues but cannot get the expected help. - [X] 2. The bug has not been fixed in the latest version. -...

Should `tokenizer_legacy` be default as `false`?

### ⚠️ Please check that this feature request hasn't been suggested before. - [X] I searched previous [Ideas in Discussions](https://github.com/axolotl-ai-cloud/axolotl/discussions/categories/ideas) didn't find any similar feature requests. - [X] I searched...

enhancement