LM head projection with logit slicing
Final language model head projection from hidden states to vocabulary logits with optional slicing for generation efficiency. Projects [batch, seq_len, 2048] to [batch, logits_to_keep, 102400]. During generation, logits_to_keep specifies how many trailing positions to compute (typically 1 for autoregressive decoding).
Reported evidence · last observed 2026-07-16. Reported by source; not independently reproduced. No public source. License unknown.
Current records
Workloadsuite of 16 cases · mean latencyvocab_size = 102400 · hidden_size = 204816 cases
Not measured on H100 for this workload. Challenges →
1375.7µs52% SOL1.00×Reported · license unknown · no source2026-07-16
Measured exactly what you asked. Reported by source; not independently reproduced.
2377.4µs51% SOL1.00×Reported · license unknown · no source2026-08-07
Measured exactly what you asked. Reported by source; not independently reproduced.
3387.3µs48% SOL1.03×Reported · license unknown · no source2026-07-25
Measured exactly what you asked. Reported by source; not independently reproduced.
4391.3µs46% SOL1.04×Reported · license unknown · no source2026-08-13
Measured exactly what you asked. Reported by source; not independently reproduced.
5392.1µs46% SOL1.04×Reported · license unknown · no source2026-08-06
Measured exactly what you asked. Reported by source; not independently reproduced.
6392.4µs46% SOL1.04×Reported · license unknown · no source2026-08-06
Measured exactly what you asked. Reported by source; not independently reproduced.
7392.4µs46% SOL1.04×Reported · license unknown · no source2026-08-06
Measured exactly what you asked. Reported by source; not independently reproduced.
8392.6µs46% SOL1.04×Reported · license unknown · no source2026-08-07
Measured exactly what you asked. Reported by source; not independently reproduced.
9392.7µs46% SOL1.05×Reported · license unknown · no source2026-08-07
Measured exactly what you asked. Reported by source; not independently reproduced.
10393.1µs46% SOL1.05×Reported · license unknown · no source2026-07-12
Measured exactly what you asked. Reported by source; not independently reproduced.
11393.4µs46% SOL1.05×Reported · license unknown · no source2026-08-07
Measured exactly what you asked. Reported by source; not independently reproduced.
12393.9µs45% SOL1.05×Reported · license unknown · no source2026-07-15
Measured exactly what you asked. Reported by source; not independently reproduced.
13394.5µs45% SOL1.05×Reported · license unknown · no source2026-08-05
Measured exactly what you asked. Reported by source; not independently reproduced.
14394.8µs45% SOL1.05×Reported · license unknown · no source2026-07-31
Measured exactly what you asked. Reported by source; not independently reproduced.
15395.3µs45% SOL1.05×Reported · license unknown · no source2026-08-05
Measured exactly what you asked. Reported by source; not independently reproduced.
16395.6µs45% SOL1.05×Reported · license unknown · no source2026-08-05
Measured exactly what you asked. Reported by source; not independently reproduced.
17396.4µs44% SOL1.06×Reported · license unknown · no source2026-08-05
Measured exactly what you asked. Reported by source; not independently reproduced.
18396.5µs45% SOL1.06×Reported · license unknown · no source2026-07-14
Measured exactly what you asked. Reported by source; not independently reproduced.
19396.6µs45% SOL1.06×Reported · license unknown · no source2026-07-31
Measured exactly what you asked. Reported by source; not independently reproduced.
20398.3µs44% SOL1.06×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
21400.8µs43% SOL1.07×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
22401.5µs43% SOL1.07×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
23402.0µs44% SOL1.07×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
24403.6µs43% SOL1.07×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
25404.4µs42% SOL1.08×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
26406.4µs41% SOL1.08×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
27408.0µs41% SOL1.09×Reported · license unknown · no source2026-07-15
Measured exactly what you asked. Reported by source; not independently reproduced.
28412.0µs40% SOL1.10×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
29425.0µs36% SOL1.13×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.