notactuallytreyanastasio
To make any language model fit your machine by maximizing LLM quality on available hardware through exact-fit mixed-precision quantization.
At a Glance
AI Tools by notactuallytreyanastasio
(1)
shoehorn
VRAM Aware LLM Quantization CLI
Discussions
No discussions yet
Be the first to start a discussion about notactuallytreyanastasio
Latest News
v0.3.0 release: discover drafters lose to the real source; surface native QAT builds
v0.2.0 release: eval + baseline deltas, AMD probe, UI, and split-selection hardening
Initial release of shoehorn: quantize models to fit your available VRAM
Show HN: deciduous — Build a Graph from Git History
Products & Services
A Rust-based mixed-precision LLM quantizer that fits models exactly into available VRAM using a Lagrangian knapsack solver.
A tool for building connected decision graphs from a repository's commit history, designed for AI assistants to recover context.
A TV show discovery tool that provides recommendations based on the writers of your favorite shows.
An interactive browser-based social art experiment featuring flocking simulations and peer-to-peer visitor interactions.
Market Position
Provides significantly better model quality at specific memory footprints compared to standard uniform quantization presets by using a solver that prioritizes bits for the most sensitive tensors.
Leadership
Founders
notactuallytreyanastasio
Software developer and creator of high-efficiency open-source tools. Experienced in Rust and Elixir, with a focus on LLM optimization and AI-assisted development tools.
Executive Team
notactuallytreyanastasio
Creator/Developer
Solo creator specializing in open-source LLM tools and development infrastructure.
Founding Story
Started as a response to a user request for an LLM runtime that could quantize BF16 models to fit exactly into available VRAM, leading to the creation of a specialized Rust-based solver.
Business Model
Revenue Model
Open source (MIT License)
Target Markets
- Running large LLMs on consumer hardware
- Optimizing model quality for specific VRAM envelopes
- Recovering development decision context from git history