Solutions Architect and AI developer
ntransformer. High-efficiency LLM inference engine in C++/CUDA. Run Llama 70B on RTX 3090.
465ps2-llm. Running a large language model on a PlayStation 2
42gpu-nvme-direct. GPU-initiated NVMe I/O via PCIe BAR MMIO — CUDA kernels directly issue NVMe commands, eliminating CPU from the storage data path
23browsergate-shield. A collection of small chrome extensions to defend and protest against LinkedIn spying behaviour on chrome based browsers.
4MP3FM. Open-source replacement for Sony MP3FileManager.exe — transfer music to Sony Network Walkman (NW-E4xx) players on Linux, macOS, and modern Windows
2brandon-tiny. Ultra-small instruction-following language models (10M-110M params) that run on a PlayStation 2. BLiMP 73.3%, HellaSwag 32.4% at just 10.7M parameters.
2osito-k. Bare-metal preemptive kernel for ESP8266 — 3D wireframe Elite demo, zForth interpreter, OsitoFS, all in 24KB of IRAM
2