Liger
Liger-Kernel · triton · BSD-2-Clause
holds 8 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 A100
declared hardwaredeclared only
architectures—
dtypes
Benchmark evidence
24 measurements across 1 GPU, fastest first.
Operation / workload
Hardware
Latency
Rank
Observed
GRPO loss, token-level importance samplingbf16 · [2, 1025, 128256]
NVIDIA A100
1.81ms
#1 of 4
2025-08-05
GRPO loss, token-level importance samplingbf16 · [2, 1025, 128256]
NVIDIA A100
1.82ms
#2 of 4
2025-08-05
GRPO loss, token-level importance samplingbf16 · [4, 1025, 128256]
NVIDIA A100
1.85ms
#1 of 4
2025-08-05
GRPO loss, token-level importance samplingbf16 · [4, 1025, 128256]
NVIDIA A100
1.85ms
#1= of 4
2025-08-05
GRPO loss, token-level importance samplingbf16 · [8, 1025, 128256]
NVIDIA A100
1.89ms
#1 of 4
2025-08-05
GRPO loss, token-level importance samplingbf16 · [8, 1025, 128256]
NVIDIA A100
1.89ms
#1= of 4
2025-08-05
GRPO loss, token-level importance samplingbf16 · [16, 1025, 128256]
NVIDIA A100
1.97ms
#1 of 4
2025-08-05
GRPO loss, token-level importance samplingbf16 · [16, 1025, 128256]
NVIDIA A100
1.98ms
#2 of 4
2025-08-05
GRPO loss, token-level importance samplingbf16 · [2, 1025, 128256]
NVIDIA A100
40.7ms
#3 of 4
2025-08-05
GRPO loss, token-level importance samplingbf16 · [2, 1025, 128256]
NVIDIA A100
40.8ms
#4 of 4
2025-08-04
Show all 24 measurements ›Showing all 24 measurements ⌄
GRPO loss, token-level importance samplingbf16 · [2, 1025, 128256]
NVIDIA A100
42.3ms
#1 of 4
2025-08-05
GRPO loss, token-level importance samplingbf16 · [2, 1025, 128256]
NVIDIA A100
42.7ms
#2 of 4
2025-08-05
GRPO loss, token-level importance samplingbf16 · [4, 1025, 128256]
NVIDIA A100
81.0ms
#3 of 4
2025-08-04
GRPO loss, token-level importance samplingbf16 · [4, 1025, 128256]
NVIDIA A100
81.7ms
#4 of 4
2025-08-05
GRPO loss, token-level importance samplingbf16 · [4, 1025, 128256]
NVIDIA A100
82.2ms
#1 of 4
2025-08-05
GRPO loss, token-level importance samplingbf16 · [4, 1025, 128256]
NVIDIA A100
82.5ms
#2 of 4
2025-08-05
GRPO loss, token-level importance samplingbf16 · [8, 1025, 128256]
NVIDIA A100
162.8ms
#3 of 4
2025-08-05
GRPO loss, token-level importance samplingbf16 · [8, 1025, 128256]
NVIDIA A100
163.4ms
#1 of 4
2025-08-05
GRPO loss, token-level importance samplingbf16 · [8, 1025, 128256]
NVIDIA A100
163.6ms
#4 of 4
2025-08-04
GRPO loss, token-level importance samplingbf16 · [8, 1025, 128256]
NVIDIA A100
167.0ms
#2 of 4
2025-08-05
GRPO loss, token-level importance samplingbf16 · [16, 1025, 128256]
NVIDIA A100
323.7ms
#3 of 4
2025-08-05
GRPO loss, token-level importance samplingbf16 · [16, 1025, 128256]
NVIDIA A100
323.7ms
#4 of 4
2025-08-04
GRPO loss, token-level importance samplingbf16 · [16, 1025, 128256]
NVIDIA A100
325.4ms
#1 of 4
2025-08-05
GRPO loss, token-level importance samplingbf16 · [16, 1025, 128256]
NVIDIA A100
327.1ms
#2 of 4
2025-08-05
Reported · How evidence levels are derived →
Source and license
sourceavailable
revision digestsha256:28174c2e416258f1e1b426e1c8becc299d62316ec2dec05909fb7a5e076e001a
license declaredBSD-2-Clause
license concludedBSD-2-Clause
imported2026-08-20
Best evidence level for this revision: reported
JSON