SGLang adapts to Kimi K3’s hybrid architecture with memory and kernel optimizations
SonglinYang4 · x · 2026-07-28
SGLang says it has adapted to Kimi K3’s hybrid KDA-MLA architecture with a set of infrastructure changes, including:
- Hybrid KDA management
- Unified memory
- ReplaySSM
- Chunked pipeline-parallel and decode-communication parallelism
- Extensive kernel optimizations and fusions
The post also cites 423 tok/s on GSM8K for day-0 Kimi K3 speed, says RL support is ready in Miles, and claims it has passed Kimi Vendor Verifier for production use. It links to blog posts, cookbooks, and benchmarks, and notes that the demo video was generated by Kimi K3 itself.
Related event: SGLang Day-0 Support for Kimi K3 Boosts Throughput to 423 tok/s(8 posts)→
More from Infra
- Engram's random reads don't suit SSDs; CPU-memory over NVLink could serve all 72 GPUs — bookwormengr · 2026-09-11
- 80% of the DIY LLM inference hype posters have already quit — it's brutally hard systems work — abhijithneil · 2026-09-11
- Hugging Face's Ultra Scale Playbook: a free book on training LLMs on GPU clusters — mdancho84 · 2026-09-11
- Is inference latency becoming the biggest bottleneck for production AI agents? — Euphoric_Sea632 · 2026-09-11
- LLM Serving Metrics Thread: Why TPOT and Uptime Make or Break User Experience — abhijithneil · 2026-09-11
- PlanetScale launches sharded Postgres: 768 servers acting as one, 1PB scale — dhruv2038 · 2026-09-11