Skip to content
KernelIndex
Search⌘K
Sign in

Liger

Liger-Kernel · triton · BSD-2-Clause
holds 14 records

Use it

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

pip install "liger-kernel==0.6.1"

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

interfacetriton
Compatibility
measured onNVIDIA H100
declared hardwaredeclared only
architectures—
dtypes

Benchmark evidence

24 measurements across 1 GPU, fastest first.

Operation / workload
Hardware
Latency
Rank
Observed
Llama 4 rotary position embeddingbf16 · [1, 1024, 8, 256]
NVIDIA H100
74.2µs
#1 of 2
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 2048, 8, 256]
NVIDIA H100
81.3µs
#1 of 4
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 2048, 8, 64]
NVIDIA H100
81.7µs
#2 of 2
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 2048, 8, 256]
NVIDIA H100
81.8µs
#2 of 4
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 2048, 8, 16]
NVIDIA H100
82.5µs
#2 of 2
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 4096, 8, 256]
NVIDIA H100
117.1µs
#1 of 2
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 1024, 8, 256]
NVIDIA H100
118.4µs
#1 of 2
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 2048, 8, 256]
NVIDIA H100
123.4µs
#1 of 4
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 4096, 8, 256]
NVIDIA H100
123.8µs
#1 of 2
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 2048, 8, 256]
NVIDIA H100
152.6µs
#2 of 4
2025-08-07
Run →
Show all 24 measurements ›
Llama 4 rotary position embeddingbf16 · [1, 2048, 8, 64]
NVIDIA H100
153.6µs
#2 of 2
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 2048, 8, 16]
NVIDIA H100
154.0µs
#2 of 2
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 8192, 8, 256]
NVIDIA H100
216.5µs
#1 of 2
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 8192, 8, 256]
NVIDIA H100
217.1µs
#1 of 2
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 2048, 8, 256]
NVIDIA H100
286.5µs
#1 of 4
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 2048, 8, 256]
NVIDIA H100
287.4µs
#2 of 4
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 1024, 8, 256]
NVIDIA H100
287.4µs
#2 of 2
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 4096, 8, 256]
NVIDIA H100
299.0µs
#1 of 2
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 2048, 8, 64]
NVIDIA H100
359.9µs
#2 of 2
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 2048, 8, 16]
NVIDIA H100
365.2µs
#2 of 2
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 16384, 8, 256]
NVIDIA H100
417.5µs
#1 of 2
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 16384, 8, 256]
NVIDIA H100
417.6µs
#1 of 2
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 8192, 8, 256]
NVIDIA H100
431.6µs
#1 of 2
2025-08-07
Run →
Llama 4 rotary position embeddingbf16 · [1, 16384, 8, 256]
NVIDIA H100
833.2µs
#1 of 2
2025-08-07
Run →

Reported · How evidence levels are derived →

Source and license

sourceavailable
revision digestsha256:7b16e7caa07229eb63274c52969e9742bd849258f08f13b57ebaa55928740b6d
license declaredBSD-2-Clause
license concludedBSD-2-Clause
imported2026-08-20

Best evidence level for this revision: reported

JSON