Update leaderboard so can support newer models?

#1
by CombinHorizon - opened

Any plans to update the leaderboard, so that it can support newer models such as mixtral (8x7B & 8x22B), gemma, & phi-3 ?

thx!

Logikon AI org

Hi @CombinHorizon , thanks for the question. Technically speaking, the leaderboard accepts all models supported by vLLM. We have already evaluated mixtral 8x7b and are busy running evals for other larger models (see https://github.com/logikon-ai/cot-eval/issues).

As to submitting larger models via the "submit" tab, the issue is really compute. It's just for lack of GPU resources that we're currently prioritising 7b models. But the plan is to get more compute. :-)

Logikon AI org

Brief follow-up: We've just upgraded to the latest lm-eval-harness and vllm releases, and successfully ran phi-3. Feel free to submit models based on latest architecture.

ggbetz changed discussion status to closed

Sign up or log in to comment