100% bootstrapped new startup. It lets you fine tune Mistral-7B and SDXL. In particular, for the LLM fine tuning we implemented a dataprep pipeline that turns websites/pdfs/doc files into question-answer pairs for training the small LLM using an big LLM.<p>It includes a GPU scheduler that can do finegrained GPU memory scheduling (Kubernetes can only do whole-GPU, we do it per-GB of GPU memory to pack both inference and fine tuning jobs into the same fleet) to fit model instances into GPU memory to optimally trade off user facing latency with GPU memory utilization<p>It's a pretty simple stack of control plane and a fat container that runs anywhere you can get hold of a GPU (e.g. runpod).<p>Architecture: <a href="https://docs.helix.ml/docs/architecture" rel="nofollow noreferrer">https://docs.helix.ml/docs/architecture</a><p>Demo walkthrough showing runner dashboard:
<a href="https://docs.helix.ml/docs/overview" rel="nofollow noreferrer">https://docs.helix.ml/docs/overview</a><p>Run it yourself: <a href="https://docs.helix.ml/docs/controlplane" rel="nofollow noreferrer">https://docs.helix.ml/docs/controlplane</a><p>Discord: <a href="https://discord.gg/VJftd844GE" rel="nofollow noreferrer">https://discord.gg/VJftd844GE</a><p>Please roast me!