Emerging Systems

SwiGLU has no exploitable contextual sparsity

FINDINGS  ·  2026-08-16   sparsityswiglunegative-resultinference

Contextual sparsity — predicting which FFN neurons will be active and skipping the rest — is a well-established acceleration technique. It does not transfer to Llama-3.

Concentration

Channels required to hold 90% of the activation magnitude, out of 8192:

layerchannels%
0377246.0
7449354.9
14447054.6
27355843.4

Half. There is no small subset to skip.

Stability

Overlap of the top-k active set between consecutive tokens:

layerk=256k=512k=1024
010%13%18%
1429%27%31%
2738%43%56%

The set churns 70-90% every token, so it cannot be predicted cheaply from the previous step either.

Why

sat_gate(g)·u is smooth. Every one of the 8192 channels comes out nonzero, just small. ReLU zeroes 90%+ of channels outright — which is where the published results come from. Llama-3 traded that sparsity away for quality.

Best case with a perfect oracle is 2x on down_proj, 25% of a layer, so about 12% of total weight traffic — and the oracle cannot exist at 20% overlap.

The route that would work is ReLUfication: fine-tune the activation back to ReLU and recover the sparsity. That is a training problem, not an inference one.

← Back to the library