HomeAI Tools


About Chinese-LlaMA2

Not long ago, Meta recently open-sourced the Llama 2 model, which is fully commercially available. It seems that Meta is bound to fight OpenAI (ClosedAI) to the end. Although Llama 2 has upgraded the original LlaMA model, it still does not support Chinese very well and needs to be customized in Chinese. So we decided to carry out the Chinese Sinicization of Llama 2:

  • 🚀 Chinese-LlaMA2-chat-sft : directly supervised fine-tuning of Llama-2,
    • using open source instructions for fine-tuning Data, such as UltraChat, various versions of Chinese alpaca corpus (such as Chinese-alpaca, BELLE), etc; Version
  • Chinese-LlaMA2 : large-scale Chinese pre-training for Llama 2;

    • the first step: first train on 42G Chinese corpus; follow-up will increase the training scale
    • Chinese-LlaMA2-chat : Fine-tuning Chinese-LlaMA2 instructions and multiple rounds of dialogue to adapt to various application scenarios and multiple rounds of dialogue interaction.

    Note that following the corresponding license, we will release the complete and merged LoRA weights, and at the same time release the LoRA weights for the convenience of the open source community.

At the same time, we will create various vertical domain models around Chinese-LlaMA2:

  • Chinese-LlaMA2-chatmed : Chinese-LlaMA2 medical field large model, supports multiple Round online consultation;
  • Chinese-LlaMA2-tcm : Chinese-LlaMA2 large model of Chinese medicine, focusing on the subdivision of Chinese medicine, empowering the inheritance of Chinese medicine

Visit Official Website


Community Posts
no data
Nothing to display