[omneity]

We develop tooling for low-resource languages for millions of underserved people, and Llama and other resources for Meta have been vital for us.

We picked Llama among the models we use for its ubiquity in the ecosystem and because many practitioners are getting quite familiar with it and its performance profile, strengths and limitations.

The major challenge we have is that given the speculative nature of our work (some of our research is on the frontier), we have to conduct multiple experiments in parallel (finetunes).

That has been proven prohibitive cost and compute wise which is bottlenecking our progress. An official finetuning platform from Meta or some other form of support from Meta that would help alleviate the impact of compute requirements would help us fly and deliver concrete results much faster.

Are you considering open a program for projects downstream of Llama?