level 3 · 50/63 correct
trajectories
63
finished
63
correct
50
compiled
51
avg speedup (correct)
4.88x
sweep-wide eval metricsaveraged over correct kernels
outcomescorrect 50 · error 12 · incorrect 1
avg speedup4.88x (over 50 correct)
avg memory ratio (kernel / ref)0.83x (50 samples)
avg fusion ratio (ref launches / kernel launches)23.21x (50 samples)
avg energy ratio (ref / kernel)1.73x (40 samples)
avg SOL score (0–1)0.454 (50 samples)
avg DRAM util %53.3%
avg occupancy %375.2%
avg max abs error5.16e-05
original · 38/50 correct
P1 · gpt-5.5-priority1_MLP.py
error—
P2 · gpt-5.5-priority2_ShallowWideMLP.py
correct1.60x7.1μs
P3 · gpt-5.5-priority3_DeepNarrowMLP.py
correct1.09x1.7μs
P4 · gpt-5.5-priority4_LeNet5.py
correct2.03x0.4μs
P5 · gpt-5.5-priority5_AlexNet.py
correct1.40x17.6μs
P6 · gpt-5.5-priority6_GoogleNetInceptionModule.py
error—
P7 · gpt-5.5-priority7_GoogleNetInceptionV1.py
correct0.52x6.5μs
P8 · gpt-5.5-priority8_ResNetBasicBlock.py
correct1.08x1.7μs
P9 · gpt-5.5-priority9_ResNet18.py
correct3.14x2.3μs
P10 · gpt-5.5-priority10_ResNet101.py
correct1.83x9.5μs
P11 · gpt-5.5-priority11_VGG16.py
correct1.51x2.4μs
P12 · gpt-5.5-priority12_VGG19.py
correct1.37x2.9μs
P13 · gpt-5.5-priority13_DenseNet121TransitionLayer.py
correct0.97x9.7μs
P14 · gpt-5.5-priority14_DenseNet121DenseBlock.py
correct0.33x40.1μs
P15 · gpt-5.5-priority15_DenseNet121.py
correct1.51x5.5μs
P16 · gpt-5.5-priority16_DenseNet201.py
correct1.70x12.7μs
P17 · gpt-5.5-priority17_SqueezeNetFireModule.py
correct1.09x12.3μs
P18 · gpt-5.5-priority18_SqueezeNet.py
error—
P19 · gpt-5.5-priority19_MobileNetV1.py
correct1.37x1.9μs
P20 · gpt-5.5-priority20_MobileNetV2.py
error—
P21 · gpt-5.5-priority21_EfficientNetMBConv.py
correct1.12x8.2μs
P22 · gpt-5.5-priority22_EfficientNetB0.py
error—
P23 · gpt-5.5-priority23_EfficientNetB1.py
correct1.34x2.1μs
P24 · gpt-5.5-priority24_EfficientNetB2.py
correct0.98x1.9μs
P25 · gpt-5.5-priority25_ShuffleNetUnit.py
correct0.97x9.4μs
P26 · gpt-5.5-priority26_ShuffleNet.py
correct1.14x10.1μs
P27 · gpt-5.5-priority27_RegNet.py
correct1.22x1.9μs
P28 · gpt-5.5-priority28_VisionTransformer.py
correct10.48x0.2μs
P29 · gpt-5.5-priority29_SwinMLP.py
correct1.97x4.1μs
P30 · gpt-5.5-priority30_SwinTransformerV2.py
correct0.81x15.9μs
P31 · gpt-5.5-priority31_VisionAttention.py
correct1.46x14.9μs
P32 · gpt-5.5-priority32_ConvolutionalVisionTransforme
correct6.43x0.4μs
P33 · gpt-5.5-priority33_VanillaRNN.py
correct1.01x1.1μs
P34 · gpt-5.5-priority34_VanillaRNNHidden.py
error—
P35 · gpt-5.5-priority35_LSTM.py
correct1.26x20.5μs
P36 · gpt-5.5-priority36_LSTMHn.py
correct1.00x30.7μs
P37 · gpt-5.5-priority37_LSTMCn.py
correct1.05x24.1μs
P38 · gpt-5.5-priority38_LSTMBidirectional.py
correct1.01x49.2μs
P39 · gpt-5.5-priority39_GRU.py
error—
P40 · gpt-5.5-priority40_GRUHidden.py
correct3.73x12.4μs
P41 · gpt-5.5-priority41_GRUBidirectional.py
correct0.09x691.0μs
P42 · gpt-5.5-priority42_GRUBidirectionalHidden.py
error—
P43 · gpt-5.5-priority43_MinGPTCausalAttention.py
correct1.44x11.8μs
P44 · gpt-5.5-priority44_MiniGPTBlock.py
correct1.49x9.1μs
P45 · gpt-5.5-priority45_UNetSoftmax.py
error—
P46 · gpt-5.5-priority46_NetVladWithGhostClusters.py
correct1.13x1.2μs
P47 · gpt-5.5-priority47_NetVladNoGhostClusters.py
error—
P48 · gpt-5.5-priority48_Mamba2ReturnY.py
correct0.94x8.9μs
P49 · gpt-5.5-priority49_Mamba2ReturnFinalState.py
error—
P50 · gpt-5.5-priority50_ReLUSelfAttention.py
error—
popcorn · 12/13 correct
P1 · gpt-5.5-priority1_DilatedConvTower_GenomicBlock.
correct1.01x25.8μs
P2 · gpt-5.5-priority2_FusedMultiStepDiffusionSolver.
correct29.43x0.1μs
P3 · gpt-5.5-priority3_SSM2D_FusedBidirectionalCrossS
correct0.05x5.0μs
P4 · gpt-5.5-priority4_MambaBlock.py
correct41.78x3.8μs
P5 · gpt-5.5-priority5_FlowMatchingProbabilityFlowODE
correct9.45x0.4μs
P6 · gpt-5.5-priority6_HyenaOperator.py
correct1.88x2.0μs
P7 · gpt-5.5-priority7_GaussianProcessRegression.py
correct1.31x0.6μs
P8 · gpt-5.5-priority8_EvoformerBlock.py
incorrect—
P9 · gpt-5.5-priority9_BayesianLinearRegression.py
correct3.26x0.1μs
P10 · gpt-5.5-priority10_GaussianMixtureModelEM.py
correct83.48x0.0μs
P11 · gpt-5.5-priority11_pipeline_three_stage_mlp.py
correct2.28x0.0μs
P12 · gpt-5.5-priority12_NormalizingFlowPlanar.py
correct1.00x2.0μs
P13 · gpt-5.5-priority13_ConditionalVAELoss.py
correct4.63x0.1μs