DiscoResearch

We are a open-source research collective focussed on the research and training of large language models. Some of our focus areas include non-english models, advanced evaluation and RL techniques.

We have extensive experience in continued pre-training and finetuning, primarily for non-english models (at the time of writing, our models have been downloaded more than 70.000 times in the last 30 days) and want to apply this knowledge to push the boundaries of open LLMs.

We aim to bring together researchers from different communities to work together on a common cause. We are in close collaboration with in LAION, AlignmentLab AI and other communities.

If you want to collaborate or have any questions/feedback, please reach out on our Discord.