Chinese Llama 2 7B

All open source, fully commercially available Chinese version of the Llama2 model and Chinese and English SFT data sets , the input format strictly follows the llama-2-chat format, and is compatible with all optimizations for the original llama-2-chat model.

basic demo

Play online

Talk is cheap, Show you the Demo.

• Demo Address / HuggingFace Spaces
• Colab (FP16/need to open high RAM, free version cannot be used)
• Colab (INT4/requires high RAM, free version cannot be used)

the latest update

• On July 31, the Chinese-English bilingual speech-text LLaSM multimodal model based on Chinese-llama2-7b was open source 🔥🔥🔥
• On July 31, the Chinese-English bilingual visual-text Chinese-LLaVA multimodal model based on Chinese-llama2-7b was open source 🔥🔥🔥
• Open source Chinese-llama2-7b-ggml model on July 26 🔥🔥
• Update 7b model on July 23, add API, provide 4bit quantization model 🔥🔥
• SFT training/reasoning code will be launched on July 22🔥
• On July 21st, docker one-click deployment went online🔥
• The demo will be launched on July 21st🔥
• On July 21, Chinese and English bilingual SFT data open source 🔥🔥
• Open source Chinese-llama2-7b model on July 21 🔥🔥

Download

• Model download
  • HuggingFace: Chinese Llama2 Chat Model
  • Baidu Netdisk: 1.0 official version
  • Baidu Netdisk: 1.1 enhanced firepower version
• 4bit quantization
  • HuggingFace: Chinese Llama2 4bit Chat Model
  • Baidu Netdisk: Chinese Llama2 4bit Chat Model
• GGML Q4 model:
  • https://huggingface.co/LinkSoul/Chinese-Llama-2-7b-ggml
  • https://huggingface.co/rffx0/Chinese-Llama-2-7b-ggml-model-q4_0
  • https://huggingface.co/soulteary/Chinese-Llama-2-7b-ggml-q4

We used the Chinese and English SFT datasets with a data volume of 10 million.

• Dataset: https://huggingface.co/datasets/LinkSoul/instruction_merge_set

Visit Official Website

https://github.com/LinkSoul-AI/Chinese-Llama-2-7b