收录实现中文版ChatGPT的各种开源技术路线,数据及其他资料
Three steps to ChatGPT:
- LLM-pretrain
- Instruction tuning and code continual pretrain
- RLHF (SFT, RM, PPO-RL)
具体技术可参考 dalinvip/Awesome-ChatGPT
模型更新太快,收录难度较高,提供以下几个榜单作为参考。
- FlashAttention (Github, Torch 2.0 implementation)
- Positional Interpolation
- NTK-Aware Scaled RoPE (Reddit, Implementation)
- Toolformer (Paper, Implementation)
- Toolbench
- Paper List
- Stanford Alpaca: LLAMA-7B SFT
- Vicuna: LLAMA-7b&13B SFT,数据来自ShareGPT
- Baize: LLAMA聊天微调,数据采集自ChatGPT self-chat
- LoRA: popular低成本LLM参数高效微调方案,已集成到PEFT
如果你创建或发现了任何关于实现中文ChatGPT的优秀资源,请创建Issue或PR来贡献这个仓库!
If you created or found any awesome resource about Chinese ChatGPT, feel free to create issues or PRs to contribute to this repository!