On July 18, Meta AI open-sourced its own new generation of large language model models, the Llama2 series. However, after trial, many friends found that whether it is the base version or the chat version, it is almost impossible to constrain the model to conduct Chinese dialogue. Therefore, the majority of colleagues urgently hope to have a Llama2 with Chinese ability for everyone to use and study.
At this point, our ChinChunMei team decided to launch a Chinese version of Llama2 open source project to provide you with a full range of Chinese version of Llama2 under different data scales, different tasks, and different training programs. This project aims to promote the community's research on large language models and their chat version models, including but not limited to the robustness and reliability of the model; the automatic evaluation scheme of the chat/dialogue version model; the ability of In Context Learning of different models and Analyze the key features behind their differences in capabilities; the model's stable training regimen, and more.
At present, the Chinese version model based on Llama2-Chat-7B has been open sourced to the warehouse:
https://huggingface.co/RicardoLee/Llama2-chat-Chinese-50W
Everyone is welcome to use it and give your valuable feedback.
At the same time, colleagues are also welcome to join the ChinChunMei-Llama2 team to explore the charm of large language models.
ChinChunMei (Jinjunmei, a kind of tea) is the name of an open source project~
Click to read the original text directly to the project warehouse~