Kimi K3 opens up with 2.8T parameters, 1M context, and day-0 vLLM support
青稞AI · wechat · 2026-07-28
A Qwen-hosted talk on Kimi K3 says the model has been open-sourced and got same-day vLLM support.
The session explains how a 2.8T-parameter native multimodal MoE model with 16 active experts out of 896 per token and a 1M-token context stresses inference stacks. It covers the hybrid KDA + periodic full-attention architecture, how vLLM handles cache management and expert parallelism, and a set of optimizations including FlashKDA prefill, fused KDA decode kernels, LatentMoE tail fusion, KV cache offloading, sequence parallelism, and speculative decoding.
Official tests claim DSpark raises single-user generation speed from 118 tok/s to 370 tok/s on SpeedBench, while KDAMetadataBuilder latency drops from 870 µs to 34 µs.
Related event: Kimi K3 shifts attention from scale to architecture(25 posts)→
More from Infra
- PyTorch details 2.12x faster on-device streaming ASR on Arm SME2 via ExecuTorch — PyTorch · 2026-10-27
- Revenue per GPU can rise as token prices fall, argues Gavin Baker — GavinSBaker · 2026-10-27
- Anthropic S-1: only $19B total compute spend since inception, overbuild warning — bookwormengr · 2026-10-27
- CoreWeave, You.com and NVIDIA lift BrowseComp 22.9% in 3-week RL post-training sprint — RichardSocher · 2026-10-27
- Grid queues make power architecture an AI-factory decision; Exowatt and Redwood offer fixes — ingliguori · 2026-10-27
- Two CVSS 9.9 Wasmtime flaws allow sandbox escape, one ARM64-only since 2020 — basedjensen · 2026-10-27