Skip to content
View eaglstun's full-sized avatar
21 followers · 44 following

Achievements

Achievement: Pair ExtraordinaireAchievement: QuickdrawAchievement: YOLOAchievement: Public SponsorAchievement: Arctic Code Vault Contributor

Achievements

Achievement: Pair ExtraordinaireAchievement: QuickdrawAchievement: YOLOAchievement: Public SponsorAchievement: Arctic Code Vault Contributor

Block or report eaglstun

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
eaglstun/README.md

Eric Eaglstun

I make CUDA-shaped machine learning run on Apple Silicon.

It started as a grudge. Nearly every interesting ML project assumes an NVIDIA card, and I have a Mac, so I kept hitting the same wall from different directions — until I got tired of going around it and started going through. That turned into an int8 Metal backend for CTranslate2, an MPS lane for finetrainers, and Open-Sora 2.0 generating video on a laptop.

Sometimes the fix belongs upstream. My favorite so far is bitsandbytes#1993: the Lion optimizer was folding weight decay into the gradient before the sign(), which corrupts the direction of the update, not just its size. It had been sitting there because the test suite only ever ran Lion with weight_decay=0 — the one setting where the bug is invisible. Merged.

Fifteen years of web before this (TypeScript, React, Node), and an art degree before that, which is my excuse for caring how a thing feels and not just whether it runs. The music, the synths, and a guitar wired to a Teensy live over at ericeaglstun.com.

Pinned Loading

  1. CTranslate2 CTranslate2 Public

    Forked from OpenNMT/CTranslate2

    An int8 Metal backend for CTranslate2 — transformer inference on Apple Silicon GPUs.

    C++

  2. ai ai Public

    A plain-language glossary of AI/ML terms, plus blog posts, practice notes, and deep dives.

    Python

  3. finetrainers finetrainers Public

    Forked from huggingface/finetrainers

    An Apple Silicon (MPS) lane for HF finetrainers — diffusion fine-tuning on a Mac.

    Python

  4. fable-mvp-gg fable-mvp-gg Public

    A séance you can run. Turn up a language model’s temperature and watch it lose signal.

    TypeScript

  5. Open-Sora Open-Sora Public

    Forked from hpcaitech/Open-Sora

    Open-Sora 2.0 on a Mac — text-to-video, the full image-to-video family, and 768px stills on Apple Silicon.

    Python

  6. skill-beaglebros skill-beaglebros Public

    A Claude skill that carries the Beagle Bros sensibility — wisecracking, generous, demystify-the-machine — into whatever you are making.