Skip to content
KernelIndex
Search⌘K

Liger

Liger-Kernel · triton · BSD-2-Clause
holds 18 records

Use it

Installable · pip · pinned to the measured build · BSD-2-ClauseView source →

pip install "liger-kernel==0.6.0"

Pinned: this command installs the newest release the evidence below measured.

interfacetriton
Compatibility
measured onNVIDIA H100, NVIDIA A100
declared hardwaredeclared only
architectures—
dtypes

Benchmark evidence

25 measurements across 2 GPUs, fastest first.

Operation / workload
Hardware
Latency
Rank
Observed
LayerNormfp32 · [4096, 1024]
NVIDIA H100
18.8µs
#1 of 2
2025-07-17
LayerNormfp32 · [4096, 2048]
NVIDIA H100
29.2µs
#1 of 2
2025-07-17
LayerNormfp32 · [4096, 1024]
NVIDIA A100
30.3µs
#1 of 2
2024-09-03
LayerNormfp32 · [4096, 2048]
NVIDIA A100
49.9µs
#1 of 2
2024-09-03
LayerNormfp32 · [4096, 4096]
NVIDIA H100
51.0µs
#1 of 2
2025-07-17
LayerNormfp32 · [4096, 4096]
NVIDIA A100
88.2µs
#1 of 2
2024-09-03
LayerNormfp32 · [4096, 8192]
NVIDIA H100
94.8µs
#1 of 2
2025-07-17
LayerNormfp32 · [4096, 1024]
NVIDIA H100
120.1µs
#2 of 2
2025-07-17
LayerNormfp32 · [4096, 2048]
NVIDIA H100
120.7µs
#2 of 2
2025-07-17
LayerNormfp32 · [4096, 8192]
NVIDIA A100
165.2µs
#1 of 2
2024-09-03
Show all 25 measurements ›
LayerNormfp32 · [4096, 4096]
NVIDIA H100
166.3µs
#1 of 2
2025-07-17
LayerNormfp32 · [4096, 16384]
NVIDIA H100
184.8µs
#1 of 2
2025-07-17
LayerNormfp32 · [4096, 1024]
NVIDIA H100
265.8µs
#2 of 2
2025-07-17
LayerNormfp32 · [4096, 2048]
NVIDIA H100
274.3µs
#2 of 2
2025-07-17
LayerNormfp32 · [4096, 4096]
NVIDIA H100
274.5µs
#2 of 2
2025-07-17
LayerNormfp32 · [4096, 2048]
NVIDIA A100
278.3µs
#2 of 2
2024-09-03
LayerNormfp32 · [4096, 1024]
NVIDIA A100
280.2µs
#2 of 2
2024-09-03
LayerNormfp32 · [4096, 4096]
NVIDIA A100
284.7µs
#1 of 2
2024-09-03
LayerNormfp32 · [4096, 8192]
NVIDIA H100
312.8µs
#1 of 2
2025-07-17
LayerNormfp32 · [4096, 16384]
NVIDIA A100
322.3µs
#1 of 2
2024-09-03
LayerNormfp32 · [4096, 8192]
NVIDIA H100
405.6µs
#1 of 2
2025-07-17
LayerNormfp32 · [4096, 8192]
NVIDIA A100
440.6µs
#1 of 2
2024-09-03
LayerNormfp32 · [4096, 16384]
NVIDIA H100
577.6µs
#1 of 2
2025-07-17
LayerNormfp32 · [4096, 16384]
NVIDIA H100
760.9µs
#1 of 2
2025-07-17
LayerNormfp32 · [4096, 16384]
NVIDIA A100
1.15ms
#1 of 2
2024-09-03

Reported · How evidence levels are derived →

Source and license

sourceavailable
revision digestsha256:77931b8372d01c2e92fd763373b2fdab449b85949aad6fcfa15bee3810319f2b
license declaredBSD-2-Clause
license concludedBSD-2-Clause
imported2026-08-18

Best evidence level for this revision: reported

JSON