LLMZoo is an open-source research project from FreedomIntelligence (CUHK Shenzhen) that bundles data, model artifacts and evaluation benchmarks for large language models in one place. Rather than releasing a single model, the project curates the ingredients around LLM development โ instruction datasets, domain models such as medical LLMs, and benchmark suites for comparing them. It has collected about 3,000 stars on GitHub.
LLMZoo grew out of the lab’s series of open LLM efforts, including the Phoenix multilingual models and the ChatMed line of Chinese medical models. The repository organizes their released datasets and evaluation resources so other researchers can reproduce results and build on them: instruction-tuning data with an emphasis on multilingual and Chinese coverage, trained model checkpoints, and benchmark definitions for measuring capabilities such as reasoning and domain knowledge. The project was most active through 2023, when open LLM research was moving fastest, and it remains a useful archive of data and evaluation material from that period โ particularly for Chinese-language and medical-domain work, where resources were scarce. The lab’s related releases, such as the AceGPT Arabic models, grew out of the same research program, which makes the repository a useful map of how one lab’s open models evolved.
LLMZoo is free and open source under the Apache-2.0 license. Its datasets, benchmarks and model artifacts can be downloaded and used for research at no cost.
LLMZoo is for LLM researchers and graduate students โ especially those working on Chinese-language or medical-domain models โ who need open training data, baselines and evaluation benchmarks.