Show HN: Kimi K3 inference on about 100k PS3 nodes (github.com)
I forked RAM coffers and implemented Kimi K3 on PS3, using one-layer-expert-per-node inference. Idea mine, used AI assistance to implement it. Inspired by AirLLM PR 116 (https://github.com/lyogavin/airllm/pull/316).
1 comment
[ 0.35 ms ] story [ 2.1 ms ] thread