Hacker News new | past | comments | ask | show | jobs | submit login
Show HN: Local fine tuning for Mistral and SDXL, GPU mem/latency optimization (helix.ml)
36 points by lewq 10 months ago | hide | past | favorite | 3 comments
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.

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

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).

Architecture: https://docs.helix.ml/docs/architecture

Demo walkthrough showing runner dashboard: https://docs.helix.ml/docs/overview

Run it yourself: https://docs.helix.ml/docs/controlplane

Discord: https://discord.gg/VJftd844GE

Please roast me!




This is huge. Luke I think you have a winner here, this is great. Can't wait to try it over the holidays.

If I can be cheeky, be sure to repost over the coming days at different hours – you're likely to spawn more traffic that way =)


Thank you! <3


Some resources:

Demo: https://www.youtube.com/watch?v=Ym4nPSzfer0

About Helix

Helix is a generative AI platform that you can run on our cloud or deploy in your own data center or cloud account. It provides an easy-to-use interface to using open source AI that's accessible to everyone.

Under the hood, it uses the best open source models and includes a GPU scheduler that can fit model instances into GPU memory to optimally trade off user facing latency with GPU memory utilization.

If you think this is cool, please vote for us on https://www.producthunt.com/posts/helix-5 today.

Docs: https://docs.helix.ml/docs/overview

Architecture: https://docs.helix.ml/docs/architecture

Things to try with LLM fine-tuning using Helix:

- https://docs.helix.ml/docs/papers

- https://docs.helix.ml/docs/engaging-content

- https://docs.helix.ml/docs/insights-data

- https://docs.helix.ml/docs/website-content

Sample sessions for SDXL:

- https://app.tryhelix.ai/session/e1b50789-a209-46c8-aa60-4d09...

- https://app.tryhelix.ai/session/cc6004cd-111b-48ae-9a8c-d651...

- https://app.tryhelix.ai/session/d50db369-4ffa-4a49-88dd-1cff...




Consider applying for YC's W25 batch! Applications are open till Nov 12.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: