ahtx / kindi

A custom LLM inference engine for low-cost NVIDIA GPUs: Quadro K420 and K1200, two K1200s joined into one 8 GB card, GTX 1080 and RTX 3080. Runs Bonsai 2B, 4B, 9B and 27B models with an OpenAI-compatible server.