DeepSeek's CED vs GLM's KV reuse: a developer unpacks how the cache-sharing modes actually differ
stochasticchasm · x · 2026-09-11
- stochasticchasm finds the emerging naming taxonomy for KV-cache reuse modes a bit funny but useful, and wonders what you'd call a mode that reuses indices but not KV, as in GLM 5.2.
- In a follow-up they detail DeepSeek's CED: the same idea, but applied only to the global branch of each CSA2 layer, uniquely per layer. In the decoder, every CSA2 layer has a global branch and a local SWA branch — the SWA branch computes from current-layer hidden states, while the global branch's KV cache is just a linear projection of encoder hidden states, and it's skipped during prefill.
- Useful architectural detail for anyone following long-context inference and KV-cache economics.
Related event: Inside YOCO: later layers share a single KV cache(4 posts)→
More from Research
- Survey Reviews Overparameterized ML Theory Behind the Fall of Bias-Variance Tradeoff — udmrzn · 2026-09-11
- OpenAI Paper Constructs Finite-Time Blowup for Navier–Stokes, Verified in Lean by GPT-6 Astra — burny_tech · 2026-09-11
- TDDN Fuses DINOv3 with CleanDIFT, More Than Triples CLIP's Dense Prediction Accuracy — udmrzn · 2026-09-11
- Scientist wireheads a fly with boosted dopamine and forces it to doomscroll 'FlyTok' — DynamicWebPaige · 2026-09-11
- NeurIPS 2026 asks all authors to pick preferred presentation sites by Sept 30 — NeurIPSConf · 2026-09-11
- Another Model Shifts to Muon Optimizer as It Emerges as the Training Default — stochasticchasm · 2026-09-11