Skip to content

Release v0.2.0

Latest
Compare
Choose a tag to compare
@Yu-Yang-Li Yu-Yang-Li released this 12 Oct 08:36
· 72 commits to main since this release
4a876df

1.通过数据集清洗再训练,缓解了先前版本经过Agent/工具学习训练后对原有知识的灾难性遗忘,并显著提升了数学推理、代码能力,可通过code interpreter解决一系列问题。

2.公布了现版本在CG-Eval评测上的结果,总排名达到第二,仅次于GPT-4,数学推理和天文能力接近或超过GPT 3.5 Turbo。