#gpt-2
3
2026-10-21
Show HN: I removed every float from GPT-2 for identical neural compression
unrated
GPT-2 with zero floats. Compress on a Mac, decompress on Windows/CUDA/Vulkan: byte-for-byte identical. 7 backends, 1 hash. An alpha of lossless LLM compression that doesn’t break when the hardware changes.
2026-09-24
Show HN: PicoLM v1.0-rc2 ("Yura Kana"). Run an LLM on Digital Unix
unrated
PicoLM is an LLM inference engine written in C99. It currently supports llama-2, GPT-2, Qwen 3.6/3.8(+MoE) and Gemma-3n models. Significant amount of work went into CPU SIMD acceleration/testing/correctness, and wide cross-platform availability with constant testing to never lose portability (from DOS through OS/X 10.4 to modernity). CUDA/HIP is supported, and accelerated IMMA kernels are available. More work needs to be done on prompt processing speed, but text generation is quite fast already. With v1.0-rc2, the list of tested and supported platforms became even more obscene (iPhone 1! Tru64!), and some amount of work has gone into having a Vulkan backend.