whoreson/picolm

Cross-platform pure C+GPU inference for Llama 2-family, Mistral Nemo, GPT-2, Gemma-3n, Qwen 3.5+, GPT-NeoX, StableLM, GLM-5.3 GGUFs. Enhancements over parent repo are 99% coded by Qwen 3.6-27b with my own lightweight custom agentic Pythonshit. No NodeJS cancer has been installed during the development. 0 money spent on cloudslop. Contains 0% ngxson

3 best points · 0 comments · 2 Hacker News posts · 15 GitHub stars · View on GitHub

Hacker News discussions

• Show HN: PicoLM v1.0-rc1 (3 points, 0 comments, by gabucino, 2026-09-03)
• Show HN: PicoLM v1.0-rc2 ("Yura Kana"). Run an LLM on Digital Unix (1 points, 0 comments, by gabucino, 2026-09-23)