// the find
X-D-Lab/LangChain-ChatGLM-Webui
基于LangChain和ChatGLM-6B等系列LLM的针对本地知识库的自动问答
A Gradio web UI for doing RAG over local documents using ChatGLM-6B and LangChain. Upload PDFs, Word docs, or markdown, pick your LLM and embedding model, ask questions. Aimed at Chinese-language users who want a local alternative to ChatGPT with their own data.
Supports multiple quantized ChatGLM variants (int4, int8) so it runs on consumer GPUs with limited VRAM. Ships a custom Chinese text splitter that handles CJK punctuation better than LangChain's default. Multiple embedding model options including Chinese-specific ones like text2vec-large-chinese and ERNIE variants. Has a Dockerfile and documented ModelScope/HuggingFace deployment paths for people without local GPU access.
Last commit was April 2024 and the models it wraps (ChatGLM-6B) have been superseded by ChatGLM3 and GLM-4 — you'd be building on outdated foundations. The repo ships binary NLTK data and a compiled Python .pyc file checked into source control, which is sloppy and inflates clone size for no reason. No tests anywhere in the tree. The whole thing is essentially a thin Gradio wrapper — if LangChain's RAG API changes (which it does constantly), this breaks silently.