Kimi K3 - GoKawiil

Deltafin fork runs full 2.8-trillion-parameter Kimi K3 model from SSDs on a MacBook Pro

A project called ARGODRIVE Deltafin, forked from gavamedia/deltafin, streams the complete, unpruned 2.8-trillion-parameter Kimi K3 model off four SSDs to run on an Apple Silicon M1 Max laptop. The latest benchmark shows throughput of about 0.29 tokens per second, roughly 2% faster than the prior measurement, with earlier logs showing the speed climbing from 0.0141 tokens/s in late July to current levels through iterative optimization.

github.com · 2026-09-08

vLLM 0.28.0 ships major performance upgrades for Kimi-K3 and DeepSeek V4

The vLLM project released version 0.28.0, combining 584 commits from 270 contributors. The update centers on deep performance work for Kimi-K3, including new decode context parallelism, fused kernels, memory-saving expert sharding, and ROCm support, alongside DeepSeek V4 improvements such as end-to-end sparse MLA and AMD Quark NVFP4 support.

github.com · 2026-08-29