Show HN: PicoLM v1.0-rc2 ("Yura Kana"). Run an LLM on Digital Unix

Category: ai-ml

Tags: llm-inference, c11, quantization, portability, gpu, simd

Score: 7.0/10 (Innovation: 8, Technical: 9, Documentation: 6, Utility: 5)

PicoLM is a C11 LLM inference engine derived from the mmap-streaming PicoLM concept, now with extensive quantization, SIMD, GPU backends, and an OpenAI-compatible HTTP server. It aims for extreme portability (from MS-DOS to modern GPUs) and competitive performance, supporting models up to 70B while streaming weights from disk.

Target audience: performance engineers, embedded developers, AI infrastructure engineers, retro computing enthusiasts

Repository: https://github.com/whoreson/picolm/ · C · MIT · 15 stars

View on Hacker News