performance

5 posts tagged “performance”.

1. 2x Decode From Block-Diffusion Drafting (2026-09-03)
2. One Environment Variable Cost Us 60x on Cache Hits (2026-08-24)
3. A 9.5x Pessimization That Looked Like a Hardware Limit (2026-08-18)
4. We Spent a Week Tuning Prefill. The Bottleneck Was Decode. (2026-08-11)
5. Our GPU Cluster Is Idle 97.4% of the Time (2026-08-09)