KnowLM is an open-source framework for building knowledgeable large language models, developed by Zhejiang University’s NLP lab (zjunlp) and hosted on GitHub with about 1,400 stars. Released under the MIT license, it covers the full lifecycle from data processing and pre-training to fine-tuning and knowledge-augmented use, with the latest code updates in 2025.
Rather than a single model, KnowLM is a standard framework for pre-training and fine-tuning LLMs with knowledge baked in. It ships with a model zoo of ready-to-use models, including ZhiXi, KnowLM-IE and OneKE for information and knowledge extraction, plus OceanGPT, together with open-source datasets. An instruction-processing module built on EasyInstruct handles instruction data, a knowledge augmentation module adds retrieval-augmented generation, and a hallucination detection module based on EasyDetect helps flag unreliable outputs. The bilingual project (English and Chinese) is aimed at researchers who want knowledgeable models that can extract, reason over and use structured knowledge. The codebase builds on common open training tooling, including DeepSpeed and LoRA-style parameter-efficient fine-tuning, so teams can adapt the released models to their own domains and data without repeating full pre-training runs from scratch.
KnowLM is free and open source under the MIT license. The framework, model weights and datasets are available from GitHub and Hugging Face at no cost; you only provide the compute needed to train or run the models.
Researchers and ML engineers working on knowledge-enhanced language models, information extraction and bilingual Chinese-English LLM research who want a complete, ready pipeline instead of assembling training code from scratch.