Hacker GitHub

← All repos

L-Forster/open-jet

View on GitHub ↗ ★ 141 Python

An offline coding agent for local models. One command sets up the best model your GPU can run, and on 24 GB+ RTX 30/40/50-series cards Qwen3.8-27B runs up to 1.9x faster than llama.cpp.

4 mentions on Hacker News by 2 people: 2 stories, 2 comments · first 2026-03-02 · latest 2026-10-28

Mentions per year 4 mentions · busiest year 2026 (4)
20082026
AllStoriesComments NewestTopOldest
  1. Qwen3.8-27b can be ran at 100 tok/sec average https://github.com/L-Forster/open-jet/tree/master/megakernel . Gemma4:31b, which is the better model with more complex architecture (and thus can't be compiled into a megakernel), is gonna be…
    comment by ActorNightly · 2026-10-28 · in New iMac with M6
  2. >Qwen 3.8 27B Q4 with 64k, 3090: ~40tok/sec Try more like 70-130 tok/sec. https://github.com/L-Forster/open-jet/tree/master/megakernel . And also Qwen models famously chase benchmarks. They are good for writing code, but fall behind in…
    comment by ActorNightly · 2026-10-28 · in New iMac with M6