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.
A star delta has been computed for 63,087 of 63,708 tracked repositories in this period, and only those can be ranked in it.
A repository missing here has not been measured far enough back yet. That does not mean it did not move.
Star counts measured between 09:50 and 17:18 (BKK).
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.
A scalable, agentic-first, and HuggingFace-native RL framework for research (9k lines).
Switchless RoCE collectives, NCCL profiles and fabric tooling for DGX Spark inference recipes
That's all 3 repositories in this period.