Add tolerances to number of files

sandlbn · sandlbn · commit fc7aca16f6d9 · 2026-02-18T14:18:35.000-08:00
diff --git a/problems/specs/KernelBench/level1/10_3D_tensor_matrix_multiplication.yaml b/problems/specs/KernelBench/level1/10_3D_tensor_matrix_multiplication.yaml
@@ -25,3 +25,5 @@ bench-gpu:
       K: 2048
       L: 768
     flop: "2*N*M*L*K"
+    rtol: 0.016
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level1/11_4D_tensor_matrix_multiplication.yaml b/problems/specs/KernelBench/level1/11_4D_tensor_matrix_multiplication.yaml
@@ -27,3 +27,5 @@ bench-gpu:
       L: 256
       K: 768
     flop: "2*B*I*J*K*L"
+    rtol: 0.016
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level1/12_Matmul_with_diagonal_matrices_.yaml b/problems/specs/KernelBench/level1/12_Matmul_with_diagonal_matrices_.yaml
@@ -21,3 +21,5 @@ bench-gpu:
       M: 4096
       N: 4096
     flop: "2*M*N"
+    rtol: 0.016
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level1/16_Matmul_with_transposed_A.yaml b/problems/specs/KernelBench/level1/16_Matmul_with_transposed_A.yaml
@@ -23,3 +23,5 @@ bench-gpu:
       N: 2048
       K: 4096
     flop: "2*M*N*K"
+    rtol: 0.016
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level1/17_Matmul_with_transposed_B.yaml b/problems/specs/KernelBench/level1/17_Matmul_with_transposed_B.yaml
@@ -23,3 +23,5 @@ bench-gpu:
       N: 2048
       K: 4096
     flop: "2*M*N*K"
+    rtol: 0.016
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level1/18_Matmul_with_transposed_both.yaml b/problems/specs/KernelBench/level1/18_Matmul_with_transposed_both.yaml
@@ -23,3 +23,5 @@ bench-gpu:
       N: 2048
       K: 4096
     flop: "2*N*M*K"
+    rtol: 0.016
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level1/19_ReLU.yaml b/problems/specs/KernelBench/level1/19_ReLU.yaml
@@ -27,3 +27,5 @@ bench-gpu:
       DIM: 16384
     flop: "BATCH*DIM"
     mem_bytes: "(2*BATCH*DIM) * 2" # f16
+    rtol: 0.016
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level1/1_Square_matrix_multiplication_.yaml b/problems/specs/KernelBench/level1/1_Square_matrix_multiplication_.yaml
@@ -30,13 +30,19 @@ bench-gpu:
     dims:
       N: 1024
     flop: "2*N*N*N"
+    rtol: 0.016
+    atol: 1.0e-05
   - params: [A, B]
     dtype: float16
     dims:
       N: 2048
     flop: "2*N*N*N"
+    rtol: 0.016
+    atol: 1.0e-05
   - params: [A, B]
     dtype: float16
     dims:
       N: 4096
     flop: "2*N*N*N"
+    rtol: 0.016
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level1/20_LeakyReLU.yaml b/problems/specs/KernelBench/level1/20_LeakyReLU.yaml
@@ -18,3 +18,5 @@ bench-gpu:
       BATCH: 1024
       DIM: 16384
     flop: "2*BATCH*DIM"
+    rtol: 0.016
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level1/2_Standard_matrix_multiplication_.yaml b/problems/specs/KernelBench/level1/2_Standard_matrix_multiplication_.yaml
@@ -33,3 +33,5 @@ bench-gpu:
       K: 4096
     flop: "2*M*N*K"
     mem_bytes: "(M*K + K*N + M*N) * 2" # f16
+    rtol: 0.016
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level1/3_Batched_matrix_multiplication.yaml b/problems/specs/KernelBench/level1/3_Batched_matrix_multiplication.yaml
@@ -24,3 +24,5 @@ bench-gpu:
       N: 1024
       K: 2048
     flop: "2*BATCH*M*N*K"
+    rtol: 0.016
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level1/4_Matrix_vector_multiplication_.yaml b/problems/specs/KernelBench/level1/4_Matrix_vector_multiplication_.yaml
@@ -23,3 +23,5 @@ bench-gpu:
       N: 1
       K: 16384
     flop: "2*M*N*K"
+    rtol: 0.016
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level1/5_Matrix_scalar_multiplication.yaml b/problems/specs/KernelBench/level1/5_Matrix_scalar_multiplication.yaml
@@ -23,3 +23,5 @@ bench-gpu:
       N: 4096
       UNIT: 1
     flop: "M*N"
+    rtol: 0.016
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level1/6_Matmul_with_large_K_dimension_.yaml b/problems/specs/KernelBench/level1/6_Matmul_with_large_K_dimension_.yaml
@@ -23,3 +23,5 @@ bench-gpu:
       N: 256
       K: 16384
     flop: "2*M*N*K"
+    rtol: 0.016
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level1/7_Matmul_with_small_K_dimension_.yaml b/problems/specs/KernelBench/level1/7_Matmul_with_small_K_dimension_.yaml
@@ -23,3 +23,5 @@ bench-gpu:
       N: 8192
       K: 64
     flop: "2*M*N*K"
+    rtol: 0.016
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level1/8_Matmul_with_irregular_shapes_.yaml b/problems/specs/KernelBench/level1/8_Matmul_with_irregular_shapes_.yaml
@@ -23,3 +23,5 @@ bench-gpu:
       N: 5921
       K: 2949
     flop: "2*M*N*K"
+    rtol: 0.016
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level1/93_masked_cumsum.yaml b/problems/specs/KernelBench/level1/93_masked_cumsum.yaml
@@ -24,3 +24,5 @@ bench-gpu:
       BATCH_SIZE: 8192
       INPUT_DIM: 8192
       SCAN_DIM: 1
+    rtol: 0.763880968
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level1/9_Tall_skinny_matrix_multiplication_.yaml b/problems/specs/KernelBench/level1/9_Tall_skinny_matrix_multiplication_.yaml
@@ -21,3 +21,5 @@ bench-gpu:
       M: 8192
       N: 32
     flop: "2*M*M*N"
+    rtol: 0.016
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level2/12_Gemm_Multiply_LeakyReLU.yaml b/problems/specs/KernelBench/level2/12_Gemm_Multiply_LeakyReLU.yaml
@@ -30,3 +30,5 @@ bench-gpu:
       MUL: 2.0
       NEG_SLOPE: 0.1
     flop: "2*BATCH*IN_FEAT*OUT_FEAT + 4*BATCH*OUT_FEAT"
+    rtol: 0.010389173
+    atol: 7.8e-03
diff --git a/problems/specs/KernelBench/level2/14_Gemm_Divide_Sum_Scaling.yaml b/problems/specs/KernelBench/level2/14_Gemm_Divide_Sum_Scaling.yaml
@@ -27,3 +27,5 @@ bench-gpu:
       HIDDEN_SIZE: 4096
       SCALE: 1.5
     flop: "2*BATCH*IN_SIZE*HIDDEN_SIZE + BATCH*HIDDEN_SIZE"
+    rtol: 5.748400211
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level2/1_Conv2D_ReLU_BiasAdd.yaml b/problems/specs/KernelBench/level2/1_Conv2D_ReLU_BiasAdd.yaml
@@ -34,3 +34,5 @@ bench-gpu:
       KERNEL_SIZE: 3
       BIAS_SHAPE: [128, 1, 1]
     flop: "2*BATCH*OUT_CHANNELS*HEIGHT*WIDTH*IN_CHANNELS*KERNEL_SIZE*KERNEL_SIZE + BATCH*OUT_CHANNELS*HEIGHT*WIDTH"
+    rtol: 0.011184794
+    atol: 0.01091092
diff --git a/problems/specs/KernelBench/level2/81_Gemm_Swish_Divide_Clamp_Tanh_Clamp.yaml b/problems/specs/KernelBench/level2/81_Gemm_Swish_Divide_Clamp_Tanh_Clamp.yaml
@@ -24,3 +24,5 @@ bench-gpu:
       IN_FEAT: 4096
       OUT_FEAT: 4096
     flop: "2*BATCH*IN_FEAT*OUT_FEAT"
+    rtol: 0.01567347
+    atol: 9.5e-04
diff --git a/problems/specs/KernelBench/level2/95_Matmul_Add_Swish_Tanh_GELU_Hardtanh.yaml b/problems/specs/KernelBench/level2/95_Matmul_Add_Swish_Tanh_GELU_Hardtanh.yaml
@@ -30,3 +30,5 @@ bench-gpu:
       OUT_FEAT: 4096
       ADD_VALUE_SHAPE: [4096]
     flop: "2*BATCH*IN_FEAT*OUT_FEAT"
+    rtol: 0.01567347
+    atol: 2.5e-03
diff --git a/problems/specs/KernelBench/level2/97_Matmul_BatchNorm_BiasAdd_Divide_Swish.yaml b/problems/specs/KernelBench/level2/97_Matmul_BatchNorm_BiasAdd_Divide_Swish.yaml
@@ -36,3 +36,5 @@ bench-gpu:
       BIAS_SHAPE: [1]
       DIV_VAL: 1.0
     flop: "8*BATCH*IN_FEAT"
+    rtol: 0.013459186
+    atol: 3.2e-03
diff --git a/problems/specs/KernelBench/level2/98_Matmul_AvgPool_GELU_Scale_Max.yaml b/problems/specs/KernelBench/level2/98_Matmul_AvgPool_GELU_Scale_Max.yaml
@@ -30,3 +30,5 @@ bench-gpu:
       POOL_KERNEL_SIZE: 16
       SCALE: 2.0
     flop: "BATCH*IN_FEAT*(POOL_KERNEL_SIZE-1) + 2*BATCH*OUT_FEAT*IN_FEAT + BATCH*OUT_FEAT"
+    rtol: 0.016
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level2/99_Matmul_GELU_Softmax.yaml b/problems/specs/KernelBench/level2/99_Matmul_GELU_Softmax.yaml
@@ -24,3 +24,5 @@ bench-gpu:
       IN_FEAT: 4096
       OUT_FEAT: 4096
     flop: "2*BATCH*IN_FEAT*OUT_FEAT"
+    rtol: 0.016
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level2/9_Matmul_Subtract_Multiply_ReLU.yaml b/problems/specs/KernelBench/level2/9_Matmul_Subtract_Multiply_ReLU.yaml
@@ -30,3 +30,5 @@ bench-gpu:
       SUB_VAL: 2.0
       MUL_VAL: 1.5
     flop: "2*BATCH*IN_FEAT*OUT_FEAT + 2*BATCH*OUT_FEAT"
+    rtol: 0.016
+    atol: 1.0e-05
diff --git a/problems/specs/KernelBench/level3/1_MLP.yaml b/problems/specs/KernelBench/level3/1_MLP.yaml
@@ -25,3 +25,5 @@ bench-gpu:
       IN_SIZE: 4096
       LAYER_SIZES: [4096, 4096]
       OUT_SIZE: 2048
+    rtol: 0.01567347
+    atol: 1.2e-03
diff --git a/problems/specs/KernelBench/level3/2_ShallowWideMLP.yaml b/problems/specs/KernelBench/level3/2_ShallowWideMLP.yaml
@@ -25,3 +25,5 @@ bench-gpu:
       IN_SIZE: 2048
       LAYER_SIZES: [4096, 4096]
       OUT_SIZE: 2048
+    rtol: 0.01567347
+    atol: 1.1e-03
diff --git a/problems/specs/KernelBench/level3/3_DeepNarrowMLP.yaml b/problems/specs/KernelBench/level3/3_DeepNarrowMLP.yaml
@@ -35,3 +35,5 @@ bench-gpu:
         1024, 1024, 1024, 1024
       ]
       OUT_SIZE: 4096
+    rtol: 0.011755102
+    atol: 1.1e-04
diff --git a/problems/specs/KernelBench/level3/43_MinGPTCausalAttention.yaml b/problems/specs/KernelBench/level3/43_MinGPTCausalAttention.yaml
@@ -33,3 +33,5 @@ bench-gpu:
       N_HEAD: 8
       ATTN_PDROP: 0.0
       RESID_PDROP: 0.0
+    rtol: 0.01567347
+    atol: 2.8e-03
diff --git a/problems/specs/KernelBench/level3/4_LeNet5.yaml b/problems/specs/KernelBench/level3/4_LeNet5.yaml
@@ -25,3 +25,5 @@ bench-gpu:
       IN_2: 32
       IN_3: 32
       NUM_CLASSES: 20
+    rtol: 0.01567347
+    atol: 2.5e-04
diff --git a/problems/specs/KernelBench/level3/5_AlexNet.yaml b/problems/specs/KernelBench/level3/5_AlexNet.yaml
@@ -25,3 +25,5 @@ bench-gpu:
       IN_2: 224
       IN_3: 224
       NUM_CLASSES: 500
+    rtol: 3.3e-03
+    atol: 1.0e-04
diff --git a/problems/specs/KernelBench/level3/9_ResNet18.yaml b/problems/specs/KernelBench/level3/9_ResNet18.yaml
@@ -27,3 +27,5 @@ bench-gpu:
       IN_2: 3
       IN_3: 224
       IN_4: 244
+    rtol: 8.2e-03
+    atol: 1.8e-04
