level 3 · 51/63 correct
trajectories
63
finished
63
correct
51
compiled
52
avg speedup (correct)
4.96x
sweep-wide eval metricsaveraged over correct kernels
outcomescorrect 51 · error 10 · incorrect 1 · compile_fail 1
avg speedup4.96x (over 51 correct)
avg memory ratio (kernel / ref)0.87x (50 samples)
avg fusion ratio (ref launches / kernel launches)73.60x (50 samples)
avg energy ratio (ref / kernel)1.86x (38 samples)
avg SOL score (0–1)0.394 (51 samples)
avg DRAM util %53.5%
avg occupancy %239.7%
avg max abs error4.53e-05
original · 40/50 correct
P1 · gpt-5.5-priority1_MLP.py
correct1.06x3.6μs
P2 · gpt-5.5-priority2_ShallowWideMLP.py
correct0.98x11.2μs
P3 · gpt-5.5-priority3_DeepNarrowMLP.py
correct5.11x0.4μs
P4 · gpt-5.5-priority4_LeNet5.py
correct0.87x0.9μs
P5 · gpt-5.5-priority5_AlexNet.py
error—
P6 · gpt-5.5-priority6_GoogleNetInceptionModule.py
correct1.38x16.4μs
P7 · gpt-5.5-priority7_GoogleNetInceptionV1.py
correct1.21x2.1μs
P8 · gpt-5.5-priority8_ResNetBasicBlock.py
correct0.83x2.1μs
P9 · gpt-5.5-priority9_ResNet18.py
correct1.18x1.8μs
P10 · gpt-5.5-priority10_ResNet101.py
correct0.65x15.9μs
P11 · gpt-5.5-priority11_VGG16.py
correct1.32x2.6μs
P12 · gpt-5.5-priority12_VGG19.py
correct1.33x2.8μs
P13 · gpt-5.5-priority13_DenseNet121TransitionLayer.py
correct5.86x1.6μs
P14 · gpt-5.5-priority14_DenseNet121DenseBlock.py
correct1.04x7.8μs
P15 · gpt-5.5-priority15_DenseNet121.py
correct1.71x5.4μs
P16 · gpt-5.5-priority16_DenseNet201.py
correct1.85x12.1μs
P17 · gpt-5.5-priority17_SqueezeNetFireModule.py
correct1.60x8.4μs
P18 · gpt-5.5-priority18_SqueezeNet.py
error—
P19 · gpt-5.5-priority19_MobileNetV1.py
error—
P20 · gpt-5.5-priority20_MobileNetV2.py
incorrect—
P21 · gpt-5.5-priority21_EfficientNetMBConv.py
correct1.14x8.7μs
P22 · gpt-5.5-priority22_EfficientNetB0.py
correct1.18x2.9μs
P23 · gpt-5.5-priority23_EfficientNetB1.py
correct2.75x1.9μs
P24 · gpt-5.5-priority24_EfficientNetB2.py
correct1.36x1.7μs
P25 · gpt-5.5-priority25_ShuffleNetUnit.py
correct0.53x15.5μs
P26 · gpt-5.5-priority26_ShuffleNet.py
correct1.02x9.6μs
P27 · gpt-5.5-priority27_RegNet.py
correct0.77x3.1μs
P28 · gpt-5.5-priority28_VisionTransformer.py
correct0.87x2.5μs
P29 · gpt-5.5-priority29_SwinMLP.py
correct2.61x4.6μs
P30 · gpt-5.5-priority30_SwinTransformerV2.py
error—
P31 · gpt-5.5-priority31_VisionAttention.py
correct1.51x14.4μs
P32 · gpt-5.5-priority32_ConvolutionalVisionTransforme
error—
P33 · gpt-5.5-priority33_VanillaRNN.py
correct1.00x6.8μs
P34 · gpt-5.5-priority34_VanillaRNNHidden.py
correct9.70x1.7μs
P35 · gpt-5.5-priority35_LSTM.py
error—
P36 · gpt-5.5-priority36_LSTMHn.py
correct0.91x28.9μs
P37 · gpt-5.5-priority37_LSTMCn.py
correct1.03x25.2μs
P38 · gpt-5.5-priority38_LSTMBidirectional.py
error—
P39 · gpt-5.5-priority39_GRU.py
correct1.02x42.8μs
P40 · gpt-5.5-priority40_GRUHidden.py
correct1.10x23.2μs
P41 · gpt-5.5-priority41_GRUBidirectional.py
error—
P42 · gpt-5.5-priority42_GRUBidirectionalHidden.py
correct0.94x326.0μs
P43 · gpt-5.5-priority43_MinGPTCausalAttention.py
error—
P44 · gpt-5.5-priority44_MiniGPTBlock.py
correct1.60x21.5μs
P45 · gpt-5.5-priority45_UNetSoftmax.py
correct1.09x5.0μs
P46 · gpt-5.5-priority46_NetVladWithGhostClusters.py
correct0.63x2.5μs
P47 · gpt-5.5-priority47_NetVladNoGhostClusters.py
correct0.28x5.2μs
P48 · gpt-5.5-priority48_Mamba2ReturnY.py
correct1.03x8.1μs
P49 · gpt-5.5-priority49_Mamba2ReturnFinalState.py
correct2.93x2.2μs
P50 · gpt-5.5-priority50_ReLUSelfAttention.py
correct—
popcorn · 11/13 correct
P1 · gpt-5.5-priority1_DilatedConvTower_GenomicBlock.
correct0.88x28.9μs
P2 · gpt-5.5-priority2_FusedMultiStepDiffusionSolver.
correct14.42x0.2μs
P3 · gpt-5.5-priority3_SSM2D_FusedBidirectionalCrossS
correct14.85x0.0μs
P4 · gpt-5.5-priority4_MambaBlock.py
error—
P5 · gpt-5.5-priority5_FlowMatchingProbabilityFlowODE
correct9.48x0.4μs
P6 · gpt-5.5-priority6_HyenaOperator.py
correct1.35x3.7μs
P7 · gpt-5.5-priority7_GaussianProcessRegression.py
correct1.47x0.6μs
P8 · gpt-5.5-priority8_EvoformerBlock.py
correct1.17x0.7μs
P9 · gpt-5.5-priority9_BayesianLinearRegression.py
correct8.30x0.0μs
P10 · gpt-5.5-priority10_GaussianMixtureModelEM.py
correct124.28x0.0μs
P11 · gpt-5.5-priority11_pipeline_three_stage_mlp.py
correct2.64x0.0μs
P12 · gpt-5.5-priority12_NormalizingFlowPlanar.py
compile_fail—
P13 · gpt-5.5-priority13_ConditionalVAELoss.py
correct3.97x0.1μs