Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

KernelBench

Benchmarks

Task NameDataset NameSOTA ResultTrend
CUDA Kernel GenerationKernelBench Level 3
Executions Count24
31
CUDA Kernel GenerationKernelBench Level 2
Execution Count48
31
CUDA Kernel GenerationKernelBench Level 1
Exec Count46
31
Kernel OptimizationKernelBench 1.0 (test)
Latency (us)0.0063
27
Verifier Robustness EvaluationKernelBench task 001 (held-out)
Hinted ASR62
26
GPU kernel code generation and optimizationKernelBench Level 3 (test)
Correctness Score100
18
GPU kernel code generation and optimizationKernelBench Level 2 (test)
Correctness19
18
GPU kernel code generation and optimizationKernelBench Level 1 (test)
Correctness19
18
GPU kernel optimizationKernelBench Level 3
Fast132.2
17
GPU kernel optimizationKernelBench Level 1
Fast1 Score37.2
17
GPU Kernel OptimizationKernelBench Level 2
Fast1 Score70.6
17
Triton kernel generationKernelBench LEVEL3 1.0
Fast1 Score29.8
17
Triton kernel generationKernelBench LEVEL2 1.0
Fast1 Score80.9
17
Triton kernel generationKernelBench LEVEL1 1.0
Fast139.3
17
Kernel Code GenerationKernelBench Level-3
Accuracy100
14
Kernel Code GenerationKernelBench Level-2
Accuracy100
14
Matrix MultiplicationKernelBench FP4 Matmul
Throughput (TF/s)2,898
14
Kernel GenerationKernelBench
Speedup3.45
14
GPU programmingKernelBench level 1 (test)
Compilation Success69
12
GPU kernel generation and optimizationKernelBench Level 2
Accuracy100
11
GPU kernel generation and optimizationKernelBench Level 1
Correctness99
11
Kernel Code GenerationKernelBench v0.2.0.dev0 (full 250-task suite)
Correctness Rate98.4
11
Kernel GenerationKernelBench Overall
CR (Round 1)25
10
Kernel GenerationKernelBench Level 2
CR (Round 1)16
10
Kernel GenerationKernelBench Level 1
Correctness Rate (Round 1)34
10
Showing 25 of 51 rows