MoE handbook: Kimi, DeepSeek, GLM all use 4-bit QAT; DeepSeek EP64 decode 25% faster
Nouamanetazi · x · 2026-11-06
Chapter 4 of the MoE Training Handbook covers serving: Kimi K3 quantized its routed experts to MXFP4 via SFT and RL, while DeepSeek-V4 and GLM-5 use 4-bit QAT after pretraining.
DeepSeek decodes V4.1-Flash at EP64–EP128, roughly 25% ahead of EP8 per GPU at 64 requests per GPU.
Related event: Hugging Face Releases Free MoE Training Handbook Built on 800K GPU Hours(20 posts)→
More from Infra
- Qwen3.8-27B hits 381 tok/s on a single RTX 3090 via inference software alone — Redo · 2026-11-06
- Dolphin Networks details 3.8× throughput gains running DeepSeek-V4.1 Flash on PCIe GPUs — AlpinDale · 2026-11-06
- Lawyer runs DeepSeek-V4.1-Flash on 8x 2018 V100s: 215 tok/s on code, 524K context fully in VRAM — TumbleweedNew6515 · 2026-11-06
- Hyperkernel generates SOTA megakernels across 11 domains, now on GitHub — MillionInt · 2026-11-06
- Musk: space will be the cheapest place to run superintelligence in 2-3 years — XFreeze · 2026-11-06
- Ex-Microsoft engineer: Azure runs 1M+ VM services, adds a datacenter every 4 days — viksit · 2026-11-06