compaction, prefix caching, sparse attention

Publication timeline

Three complementary mitigations for the same superlinear cost curve, operating at application, serving, and model-architecture layers respectively.

1 source videos

No core insights are tagged with this concept.