Baichuan2-7B

Baichuan2-7B is the 7-billion-parameter open large language model from Baichuan Intelligence, released in September 2023 alongside a 13B sibling in Base and Chat variants. Trained on 2.6 trillion tokens of high-quality bilingual data, it achieved best-in-class results among same-size models on authoritative Chinese and English benchmarks.\n\nThe Chat version adds instruction following and a 4-bit quantized release for constrained hardware, and the family became one of the most widely deployed Chinese open-model bases for downstream fine-tunes.\n\nAll versions are fully open for academic research; commercial use requires a free commercial license obtained via email request to Baichuan.