deepgemm_fp8_fp4_gemm_nt_n2048_k2048
FlashInfer-Bench baselines · python · Apache-2.0
Kernel source · 12 lines ↓holds 3 records
Use it
Vendorable · source mirrored · Apache-2.0View source →
No package. Vendor the mirrored source: 12 lines, Apache-2.0, pinned at da91508.
main.py
curl "https://kernelindex.com/api/v1/implementations/flashinfer-deepgemm-fp8-fp4-gemm-nt-n2048-k2048?include=source"interfacepython
revisionda915083d4c7
symbolrun
pathmain.py
Compatibility
measured onNVIDIA B200
declared hardwareNVIDIA B200
architecturessm_100
dtypesfp32, fp8_e4m3, int8
Benchmark evidence
3 measurements across 1 GPU, fastest first.
Operation / workload
Hardware
Latency
Rank
Observed
Reported · How evidence levels are derived →
Source and license
sourcehttps://huggingface.co/datasets/flashinfer-ai/flashinfer-trace
commitda915083d4c7c5e61aa3005e3d17ae488e0fc71c
revision digestsha256:069490f864b1f05e2e2098f284dd47184d2e9094b138d2322bce4502e478bd87
license declaredApache-2.0
license concludedApache-2.0
authorsbaseline
imported2026-08-16
Kernel source
main.py12 lines
import torch
import deep_gemm
def run(a_fp8, a_scale, b_fp4, b_scale):
m = a_fp8.shape[0]
n = b_fp4.shape[0]
d = torch.empty(m, n, device=a_fp8.device, dtype=torch.bfloat16)
deep_gemm.fp8_fp4_gemm_nt((a_fp8, a_scale), (b_fp4, b_scale), d,
recipe_a=(1, 128), recipe_b=(1, 32))
return d
Source code from FlashInfer-Bench (flashinfer-ai/flashinfer-trace) · Apache-2.0
Best evidence level for this revision: reported
JSON