Train llm (bloom, llama, baichuan2-7b, chatglm3-6b) with deepspeed pipeline mode. Faster than zero/zero++/fsdp.

[ACL 2024 Main] NewsBench: A Systematic Evaluation Framework for Assessing Editorial Capabilities of Large Language Models in Chinese Journalism

LLM API Server , OpenAI 同时支持 ChatGLM3 ，Llama, Llama-3, Firefunction, Openfunctions ，BAAI/bge-m3 ,bge-large-zh-v1.5

使用 FastAPI 搭配預訓練 CodeGeeX2 模型，能通過應用在 API Hook 讓 AI 智能生成代碼