Looking is not storing

Sparse attention can stop looking at a token without stopping storing it. Linear layers do the reverse in most of the stack. Decode getting cheaper does not mean the GPU got emptier.

You bought a 128k window. The vendor shipped “sparse attention.” Time-per-output-token dropped. Batch-4 still OOMs. The growth lab asked what a token costs to carry. This lab asks which bill the architecture actually paid.

Architecture
stored (resident KV) looked at this decode
Looked at
Stored tokens
Resident KV
Indexer work
Two bills — yellow is compute, teal is occupancy
Cuts vs leaves — same identity, different axes
BetCutsLeavesWho
GQAH (8 KV heads vs 96 Q)T still linearGLM-4.5 2508.06471
MLAH further (latent ~70 KB/tok vs GQA hundreds)T still linear; still look at all TDeepSeek V2/V3 2405.04434
NSA / DSAT looked at (top-k / branches)T stored, unless serving offloadsDeepSeek 2502.11089 · 2512.02556
IndexCacheIndexer FLOPs (~75% of layers reuse top-k)core k and stored TZ.ai / THU 2603.12201
KDA 3:1T in 3/4 of layers (finite-state)MLA layers still store TKimi Linear 2510.26692
MooncakePrefill recompute across the clusterThis request’s TKimi serving 2407.00079

mHC (2512.24880) is residual topology. It is not on this grid. Offload is a serving choice, not what DSA ships by default.

What to try
  • Leave T at 32k. Click DSA. Yellow shrinks to k. Teal does not move. That is the OOM-with-faster-decode failure.
  • Now KDA 3:1. Teal drops to the MLA quarter. Kimi paid the occupancy bill; DeepSeek paid the look bill.
  • Click + IndexCache after DSA. Look and store stay put; only indexer work falls. Next tax after DSA, not a smaller cache.
  • DSA + offload makes teal follow yellow. Feel it — then remember this is extra serving work, not the paper’s default.
  • Drag T to 128k on DSA, then on GQA. GQA look and store both explode; DSA only the store bar does.
Ask first: did we stop looking, stop storing, or only thin each stored token? Sparse ≠ smaller KV. Linear ≠ free quality on the MLA layers. Cluster reuse (Mooncake) is a fourth bill, already in the prefix-reuse lab.