level 2 · 82/144 correct
trajectories
144
finished
144
correct
82
compiled
105
avg speedup (correct)
21.93x
sweep-wide eval metricsaveraged over correct kernels
outcomescorrect 82 · compile_fail 38 · incorrect 23 · error 1
avg speedup21.93x (over 82 correct)
avg memory ratio (kernel / ref)0.73x (82 samples)
avg fusion ratio (ref launches / kernel launches)62.93x (82 samples)
avg energy ratio (ref / kernel)2.54x (55 samples)
avg SOL score (0–1)0.627 (82 samples)
avg DRAM util %49.3%
avg occupancy %159.2%
avg max abs error2.72e-03
original · 63/100 correct
P1 · gpt-5.5-priority1_Conv2D_ReLU_BiasAdd.py
compile_fail—
P2 · gpt-5.5-priority2_ConvTranspose2d_BiasAdd_Clamp_
correct1.93x6.8μs
P3 · gpt-5.5-priority3_ConvTranspose3d_Sum_LayerNorm_
incorrect—
P4 · gpt-5.5-priority4_Conv2d_Mish_Mish.py
compile_fail—
P5 · gpt-5.5-priority5_ConvTranspose2d_Subtract_Tanh.
correct1.09x8.5μs
P6 · gpt-5.5-priority6_Conv3d_Softmax_MaxPool_MaxPool
correct0.73x1.2μs
P7 · gpt-5.5-priority7_Conv3d_ReLU_LeakyReLU_GELU_Sig
correct1.42x4.5μs
P8 · gpt-5.5-priority8_Conv3d_Divide_Max_GlobalAvgPoo
correct0.30x8.5μs
P9 · gpt-5.5-priority9_Matmul_Subtract_Multiply_ReLU.
compile_fail—
P10 · gpt-5.5-priority10_ConvTranspose2d_MaxPool_Hardt
correct1.17x8.3μs
P11 · gpt-5.5-priority11_ConvTranspose2d_BatchNorm_Tan
correct1.01x2.7μs
P12 · gpt-5.5-priority12_Gemm_Multiply_LeakyReLU.py
correct1.00x2.7μs
P13 · gpt-5.5-priority13_ConvTranspose3d_Mean_Add_Soft
correct1.36x7.6μs
P14 · gpt-5.5-priority14_Gemm_Divide_Sum_Scaling.py
incorrect—
P15 · gpt-5.5-priority15_ConvTranspose3d_BatchNorm_Sub
incorrect—
P16 · gpt-5.5-priority16_ConvTranspose2d_Mish_Add_Hard
correct1.61x7.1μs
P17 · gpt-5.5-priority17_Conv2d_InstanceNorm_Divide.py
correct1.26x4.0μs
P18 · gpt-5.5-priority18_Matmul_Sum_Max_AvgPool_LogSum
correct96.52x0.0μs
P19 · gpt-5.5-priority19_ConvTranspose2d_GELU_GroupNor
correct1.08x11.4μs
P20 · gpt-5.5-priority20_ConvTranspose3d_Sum_ResidualA
correct2.26x1.7μs
P21 · gpt-5.5-priority21_Conv2d_Add_Scale_Sigmoid_Grou
compile_fail—
P22 · gpt-5.5-priority22_Matmul_Scale_ResidualAdd_Clam
incorrect—
P23 · gpt-5.5-priority23_Conv3d_GroupNorm_Mean.py
correct193.35x0.0μs
P24 · gpt-5.5-priority24_Conv3d_Min_Softmax.py
compile_fail—
P25 · gpt-5.5-priority25_Conv2d_Min_Tanh_Tanh.py
incorrect—
P26 · gpt-5.5-priority26_ConvTranspose3d_Add_HardSwish
incorrect—
P27 · gpt-5.5-priority27_Conv3d_HardSwish_GroupNorm_Me
compile_fail—
P28 · gpt-5.5-priority28_BMM_InstanceNorm_Sum_Residual
compile_fail—
P29 · gpt-5.5-priority29_Matmul_Mish_Mish.py
correct1.00x2.7μs
P30 · gpt-5.5-priority30_Gemm_GroupNorm_Hardtanh.py
incorrect—
P31 · gpt-5.5-priority31_Conv2d_Min_Add_Multiply.py
correct1.43x3.5μs
P32 · gpt-5.5-priority32_Conv2d_Scaling_Min.py
compile_fail—
P33 · gpt-5.5-priority33_Gemm_Scale_BatchNorm.py
correct1.01x2.8μs
P34 · gpt-5.5-priority34_ConvTranspose3d_LayerNorm_GEL
correct3.06x3.3μs
P35 · gpt-5.5-priority35_Conv2d_Subtract_HardSwish_Max
compile_fail—
P36 · gpt-5.5-priority36_ConvTranspose2d_Min_Sum_GELU_
correct251.71x0.0μs
P37 · gpt-5.5-priority37_Matmul_Swish_Sum_GroupNorm.py
correct1.42x6.0μs
P38 · gpt-5.5-priority38_ConvTranspose3d_AvgPool_Clamp
compile_fail—
P39 · gpt-5.5-priority39_Gemm_Scale_BatchNorm.py
compile_fail—
P40 · gpt-5.5-priority40_Matmul_Scaling_ResidualAdd.py
correct1.03x11.2μs
P41 · gpt-5.5-priority41_Gemm_BatchNorm_GELU_ReLU.py
correct1.05x11.3μs
P42 · gpt-5.5-priority42_ConvTranspose2d_GlobalAvgPool
compile_fail—
P43 · gpt-5.5-priority43_Conv3d_Max_LogSumExp_ReLU.py
compile_fail—
P44 · gpt-5.5-priority44_ConvTranspose2d_Multiply_Glob
compile_fail—
P45 · gpt-5.5-priority45_Gemm_Sigmoid_LogSumExp.py
correct0.97x9.2μs
P46 · gpt-5.5-priority46_Conv2d_Subtract_Tanh_Subtract
correct1.77x3.3μs
P47 · gpt-5.5-priority47_Conv3d_Mish_Tanh.py
correct1.08x2.4μs
P48 · gpt-5.5-priority48_Conv3d_Scaling_Tanh_Multiply_
compile_fail—
P49 · gpt-5.5-priority49_ConvTranspose3d_Softmax_Sigmo
correct0.16x19.1μs
P50 · gpt-5.5-priority50_ConvTranspose3d_Scaling_AvgPo
correct12.16x0.5μs
P51 · gpt-5.5-priority51_Gemm_Subtract_GlobalAvgPool_L
correct77.09x0.1μs
P52 · gpt-5.5-priority52_Conv2d_Activation_BatchNorm.p
correct1.44x2.5μs
P53 · gpt-5.5-priority53_Gemm_Scaling_Hardtanh_GELU.py
compile_fail—
P54 · gpt-5.5-priority54_Conv2d_Multiply_LeakyReLU_GEL
correct1.19x5.2μs
P55 · gpt-5.5-priority55_Matmul_MaxPool_Sum_Scale.py
incorrect—
P56 · gpt-5.5-priority56_Matmul_Sigmoid_Sum.py
correct0.99x5.3μs
P57 · gpt-5.5-priority57_Conv2d_ReLU_HardSwish.py
correct1.12x2.7μs
P58 · gpt-5.5-priority58_ConvTranspose3d_LogSumExp_Har
correct7.39x0.9μs
P59 · gpt-5.5-priority59_Matmul_Swish_Scaling.py
correct1.01x5.3μs
P60 · gpt-5.5-priority60_ConvTranspose3d_Swish_GroupNo
incorrect—
P61 · gpt-5.5-priority61_ConvTranspose3d_ReLU_GroupNor
compile_fail—
P62 · gpt-5.5-priority62_Matmul_GroupNorm_LeakyReLU_Su
correct1.10x2.8μs
P63 · gpt-5.5-priority63_Gemm_ReLU_Divide.py
correct0.99x2.8μs
P64 · gpt-5.5-priority64_Gemm_LogSumExp_LeakyReLU_Leak
compile_fail—
P65 · gpt-5.5-priority65_Conv2d_AvgPool_Sigmoid_Sum.py
correct0.90x13.5μs
P66 · gpt-5.5-priority66_Matmul_Dropout_Softmax.py
compile_fail—
P67 · gpt-5.5-priority67_Conv2d_GELU_GlobalAvgPool.py
correct0.86x7.6μs
P68 · gpt-5.5-priority68_Matmul_Min_Subtract.py
correct1.01x1.4μs
P69 · gpt-5.5-priority69_Conv2d_HardSwish_ReLU.py
correct1.21x1.6μs
P70 · gpt-5.5-priority70_Gemm_Sigmoid_Scaling_Residual
correct1.02x2.7μs
P71 · gpt-5.5-priority71_Conv2d_Divide_LeakyReLU.py
correct1.11x1.7μs
P72 · gpt-5.5-priority72_ConvTranspose3d_BatchNorm_Avg
incorrect—
P73 · gpt-5.5-priority73_Conv2d_BatchNorm_Scaling.py
correct1.16x2.2μs
P74 · gpt-5.5-priority74_ConvTranspose3d_LeakyReLU_Mul
incorrect—
P75 · gpt-5.5-priority75_Gemm_GroupNorm_Min_BiasAdd.py
correct1.11x2.8μs
P76 · gpt-5.5-priority76_Gemm_Add_ReLU.py
correct1.10x0.4μs
P77 · gpt-5.5-priority77_ConvTranspose3d_Scale_BatchNo
correct0.02x232.0μs
P78 · gpt-5.5-priority78_ConvTranspose3d_Max_Max_Sum.p
correct0.23x209.0μs
P79 · gpt-5.5-priority79_Conv3d_Multiply_InstanceNorm_
compile_fail—
P80 · gpt-5.5-priority80_Gemm_Max_Subtract_GELU.py
correct488.29x0.0μs
P81 · gpt-5.5-priority81_Gemm_Swish_Divide_Clamp_Tanh_
correct1.05x2.7μs
P82 · gpt-5.5-priority82_Conv2d_Tanh_Scaling_BiasAdd_M
correct1.65x6.0μs
P83 · gpt-5.5-priority83_Conv3d_GroupNorm_Min_Clamp_Dr
correct25.47x0.1μs
P84 · gpt-5.5-priority84_Gemm_BatchNorm_Scaling_Softma
correct1.01x2.8μs
P85 · gpt-5.5-priority85_Conv2d_GroupNorm_Scale_MaxPoo
correct1.62x1.6μs
P86 · gpt-5.5-priority86_Matmul_Divide_GELU.py
correct0.99x2.8μs
P87 · gpt-5.5-priority87_Conv2d_Subtract_Subtract_Mish
correct1.58x5.9μs
P88 · gpt-5.5-priority88_Gemm_GroupNorm_Swish_Multiply
correct1.18x2.7μs
P89 · gpt-5.5-priority89_ConvTranspose3d_MaxPool_Softm
compile_fail—
P90 · gpt-5.5-priority90_Conv3d_LeakyReLU_Sum_Clamp_GE
correct0.62x30.6μs
P91 · gpt-5.5-priority91_ConvTranspose2d_Softmax_BiasA
compile_fail—
P92 · gpt-5.5-priority92_Conv2d_GroupNorm_Tanh_HardSwi
compile_fail—
P93 · gpt-5.5-priority93_ConvTranspose2d_Add_Min_GELU_
incorrect—
P94 · gpt-5.5-priority94_Gemm_BiasAdd_Hardtanh_Mish_Gr
incorrect—
P95 · gpt-5.5-priority95_Matmul_Add_Swish_Tanh_GELU_Ha
correct1.04x2.7μs
P96 · gpt-5.5-priority96_ConvTranspose3d_Multiply_Max_
correct14.96x0.4μs
P97 · gpt-5.5-priority97_Matmul_BatchNorm_BiasAdd_Divi
correct1.04x2.8μs
P98 · gpt-5.5-priority98_Matmul_AvgPool_GELU_Scale_Max
correct1.96x1.4μs
P99 · gpt-5.5-priority99_Matmul_GELU_Softmax.py
compile_fail—
P100 · gpt-5.5-priority100_ConvTranspose3d_Clamp_Min_Di
correct1.21x11.5μs
popcorn · 19/44 correct
P1 · gpt-5.5-priority1_DeepSeekMLALoRAExpansion.py
compile_fail—
P3 · gpt-5.5-priority3_FusedCrossAttentionRoPEDropout
incorrect—
P4 · gpt-5.5-priority4_GFVLA_FusedSceneGraphUpdate.py
compile_fail—
P5 · gpt-5.5-priority5_LayerNorm_GELU_DilatedConv1d_R
incorrect—
P6 · gpt-5.5-priority6_TriangularAttention.py
correct1.82x0.3μs
P7 · gpt-5.5-priority7_Conv1d_Depthwise_Separable.py
incorrect—
P8 · gpt-5.5-priority8_DeepSeekMoEGroundedTop2Routing
compile_fail—
P9 · gpt-5.5-priority9_MetropolisHastingsStep.py
correct12.30x0.0μs
P10 · gpt-5.5-priority10_GFVLA_FusedPairwiseNodeAggreg
compile_fail—
P12 · gpt-5.5-priority12_BatchNorm_GELU_DilatedConv1d_
compile_fail—
P13 · gpt-5.5-priority13_QuantVLA_SelectiveLayoutFusio
correct2.42x0.0μs
P14 · gpt-5.5-priority14_TriangularMultiplicativeUpdat
incorrect—
P15 · gpt-5.5-priority15_GQAKVHeadExpansionAttention.p
compile_fail—
P16 · gpt-5.5-priority16_HamiltonianMonteCarloStep.py
correct37.76x0.0μs
P17 · gpt-5.5-priority17_FP8ScaledAttention.py
incorrect—
P19 · gpt-5.5-priority19_DepthwiseSeparableConv1D_GELU
incorrect—
P20 · gpt-5.5-priority20_GatedDeltaNetLinearAttention.
correct111.31x0.1μs
P21 · gpt-5.5-priority21_TriangularMultiplicativeUpdat
correct1.96x0.1μs
P22 · gpt-5.5-priority22_FusedBallQuery3D.py
correct19.48x0.0μs
P23 · gpt-5.5-priority23_GibbsSamplingStep.py
incorrect—
P24 · gpt-5.5-priority24_KimiDeltaAttentionChannelwise
correct141.45x0.0μs
P25 · gpt-5.5-priority25_OuterProductMean.py
correct2.70x0.1μs
P26 · gpt-5.5-priority26_RoPEKVCacheUpdate.py
incorrect—
P28 · gpt-5.5-priority28_FusedGroupNormSiLU.py
compile_fail—
P29 · gpt-5.5-priority29_MSARowAttention.py
compile_fail—
P30 · gpt-5.5-priority30_SelectiveScan_Mamba.py
correct159.76x0.8μs
P31 · gpt-5.5-priority31_FusedMLAAttention.py
compile_fail—
P32 · gpt-5.5-priority32_MSAColumnAttention.py
correct6.64x0.1μs
P33 · gpt-5.5-priority33_VectorizedPoseTransformFused.
correct43.25x0.0μs
P35 · gpt-5.5-priority35_FusedConv1dGroupNorm.py
incorrect—
P36 · gpt-5.5-priority36_InvariantPointAttention.py
correct16.53x0.0μs
P37 · gpt-5.5-priority37_CSRFusedAttentionValue.py
compile_fail—
P39 · gpt-5.5-priority39_Fused1DTemporalConvolution.py
correct0.69x0.1μs
P40 · gpt-5.5-priority40_SmithWatermanDPScore.py
compile_fail—
P41 · gpt-5.5-priority41_VariationalELBO.py
correct2.30x0.1μs
P42 · gpt-5.5-priority42_BatchedForwardKinematics.py
compile_fail—
P43 · gpt-5.5-priority43_ContactMapPrediction.py
correct0.80x0.1μs
P44 · gpt-5.5-priority44_ReparameterizationTrick.py
correct5.55x0.1μs
P45 · gpt-5.5-priority45_AxialAttention.py
incorrect—
P46 · gpt-5.5-priority46_SteinVariationalGradient.py
correct0.05x20.1μs
P47 · gpt-5.5-priority47_virtual_all_gather_concat_lin
compile_fail—
P48 · gpt-5.5-priority48_ParticleFilter.py
compile_fail—
P49 · gpt-5.5-priority49_sum_gelu_across_virtual_ranks
correct1.53x0.0μs
P50 · gpt-5.5-priority50_HiddenMarkovForward.py
error—