llama-3-2-1b
0 of 9 operations have a usable best known on NVIDIA B200 · 9 without one
Best known on NVIDIA B200
GEMM n16384 k2048fp16 · [80, 2048]torch / matmul0a4e73not usable20.9µsReported · Apache-2.0 · source2026-04-20
No entry in this cohort passes the usability policy (no install recipe); the fastest known is shown. Measured by the source's own harness. Ranked by its metric under ranking-v1.
GEMM n2048 k2048fp16 · [56, 2048]torch / matmul756deanot usable10.7µsReported · Apache-2.0 · source2026-04-20
No entry in this cohort passes the usability policy (no install recipe); the fastest known is shown. Measured by the source's own harness. Ranked by its metric under ranking-v1.
GEMM n2048 k8192fp16 · [2, 8192]torch / matmulff95bcnot usable15.9µsReported · Apache-2.0 · source2026-04-20
No entry in this cohort passes the usability policy (no install recipe); the fastest known is shown. Measured by the source's own harness. Ranked by its metric under ranking-v1.
GEMM n3072 k2048fp16 · [15, 2048]torch / matmul73342cnot usable10.8µsReported · Apache-2.0 · source2026-04-20
No entry in this cohort passes the usability policy (no install recipe); the fastest known is shown. Measured by the source's own harness. Ranked by its metric under ranking-v1.
GQA paged decode h32 kv8 d64 ps1bf16 · [1, 32, 64] · num pages 9 · num kv indices 8flashinfer / wrapper3f9411not usable8.15µsReported · Apache-2.0 · source2026-04-20
No entry in this cohort passes the usability policy (no install recipe); the fastest known is shown. Measured by the source's own harness. Ranked by its metric under ranking-v1. 1 more entry in this cohort.
GQA paged decode h32 kv8 d64 ps64bf16 · [1, 32, 64] · num pages 66 · num kv indices 2flashinfer / wrapper68ff49not usable10.2µsReported · Apache-2.0 · source2026-04-20
No entry in this cohort passes the usability policy (no install recipe); the fastest known is shown. Measured by the source's own harness. Ranked by its metric under ranking-v1. 1 more entry in this cohort.
GQA paged prefill causal h32 kv8 d64 ps1bf16 · [4, 32, 64]flashinfer / wrapperece89anot usable12.3µsReported · Apache-2.0 · source2026-04-20
No entry in this cohort passes the usability policy (no install recipe); the fastest known is shown. Measured by the source's own harness. Ranked by its metric under ranking-v1. 1 more entry in this cohort.
GQA paged prefill causal h32 kv8 d64 ps64bf16 · [4, 32, 64]flashinfer / wrapper06ced8not usable19.5µsReported · Apache-2.0 · source2026-04-20
No entry in this cohort passes the usability policy (no install recipe); the fastest known is shown. Measured by the source's own harness. Ranked by its metric under ranking-v1. 1 more entry in this cohort.
GQA ragged prefill causal h32 kv8 d64bf16 · [1, 8, 64]flashinfer / wrapper44b034not usable8.93µsReported · Apache-2.0 · source2026-04-20
No entry in this cohort passes the usability policy (no install recipe); the fastest known is shown. Measured by the source's own harness. Ranked by its metric under ranking-v1. 1 more entry in this cohort.
Coverage gaps on NVIDIA B200
No eligible evidence yet for these operations. Challenges →