add clue-ai/ChatYuan

[alreadydone]

https://github.com/clue-ai/ChatYuan

I opened an issue there but got no response. Maybe someone here will volunteer to add it.

[nichtdax]

Can you provide the link to some kind of Readme in English?

[alreadydone]

The official readme contains English:

ChatYuan large v2 is an open-source large language model for dialogue, supports both Chinese and English languages, and in ChatGPT style.

Based on the original functions of Chatyuan-large-v1, we optimized the model as follows:

-Added the ability to speak in both Chinese and English.

-Added the ability to refuse to answer. Learn to refuse to answer some dangerous and harmful questions.

-Added code generation functionality. Basic code generation has been optimized to a certain extent.

-Enhanced basic capabilities. The original contextual Q&A and creative writing skills have significantly improved.

-Added a table generation function. Make the generated table content and format more appropriate.

-Enhanced basic mathematical computing capabilities.

-The maximum number of length tokens has been expanded to 4096.

-Enhanced ability to simulate scenarios

But there are some parts in the Chinese readme that isn't translated:

It uses the same technical approach as the v1 version and has been optimized in aspects such as fine-tuning data, RLHF, and chain-of-thoughts.

ChatYuan-large-v2 is a lightweight model in the ChatYuan series that achieves high quality. Inference can be perform on consumer-level graphics cards, PCs, or even mobile phones (only 400M for INT4).

online Demo(Huggingface) | online Demo(ModelScope, in Chinese) | try online in Colab

The base model is PromptCLUE (770M parameters, download) pre-trained on a 100B-token Chinese corpus, which has also undergone Prompt-task-based training on hundreds of tasks. It was then trained on 100M functional dialogue / multi-turn dialogue data to obtain the ChatYuan model. ClueAI's online service is based on proprietary models of dozens of billions of parameters, and they have plans to open-source larger models, according to this article (in Chinese).

[nichtdax]

But there are some parts in the Chinese readme that isn't translated:

I would prefer a fully English readme with example. But do you think we should just add your comment in Related Links (if this project is added to the list)

[alreadydone]

Indeed it seems they are too focused on Chinese users (their company is centered around the CLUE benchmark as you can see from its name ClueAI, where C stands for Chinese), so that all examples provided are in Chinese. Maybe ChatYuan is not friendly enough to English users, and maybe it's not a good fit for this repo for this reason, but if you decide to add it, feel free to link to my comment above.

[nichtdax]

Resolved by 50ee7e2