Adjust threshold for nn.ConvTranspose3d unit test (#1372)

Adjust threshold for nn.ConvTranspose3d unit test. --------- Co-authored-by: Yutao Xu <[email protected]>
intel · Feb 18, 2025 · a14d1ea · a14d1ea
1 parent 9712a05
commit a14d1ea
Show file tree

Hide file tree

Showing 2 changed files with 39 additions and 34 deletions.
diff --git a/test/xpu/skip_list_common.py b/test/xpu/skip_list_common.py
@@ -1740,40 +1740,40 @@
         "test_fn_fwgrad_bwgrad_nn_functional_conv3d_xpu_complex128",
         "test_fn_fwgrad_bwgrad_nn_functional_conv3d_xpu_float64",
     ),
-    "test_matmul_cuda_xpu.py": (
-        # AssertionError: "Bias is not supported when out_dtype is set to Float32" does not match "Could not run 'aten::_scaled_mm' with arguments from the 'CPU' backend.
-        "test_float32_output_errors_with_bias_xpu",
-        # RuntimeError: "eye" not implemented for 'Float8_e4m3fn'
-        "test_float8_basics_xpu",
-        # AssertionError: "For row-wise scaling, scale_a must be size 1024 but got 1 and scale_b must be size 2048 but got 2" does not match "Could not run 'aten::_scaled_mm' with arguments from the 'CPU' backend.
-        "test_float8_error_messages_xpu",
-        # NotImplementedError: Could not run 'aten::_scaled_mm' with arguments from the 'CPU' backend.
-        "test_float8_bias_relu_edgecase_xpu",
-        "test_float8_bias_xpu",
-        "test_float8_rowwise_scaling_sanity_use_fast_accum_False_xpu",
-        "test_float8_rowwise_scaling_sanity_use_fast_accum_True_xpu",
-        "test_float8_scale_fast_accum_xpu",
-        "test_float8_scale_xpu",
-        "test_non_divisible_leading_dim_bias_False_xpu",
-        "test_non_divisible_leading_dim_bias_True_xpu",
-        "test_scaled_mm_change_stride_bfloat16_xpu",
-        "test_scaled_mm_change_stride_float16_xpu",
-        "test_scaled_mm_change_stride_float32_xpu",
-        "test_scaled_mm_vs_emulated_bfloat16_xpu",
-        "test_scaled_mm_vs_emulated_float16_xpu",
-        "test_scaled_mm_vs_emulated_float32_xpu",
-        "test_scaled_mm_vs_emulated_row_wise_bfloat16_xpu",
-        # AssertionError: Torch not compiled with CUDA enabled
-        "test_zero_dim_tensorwise_which_dim_zero",
-        # New added case in 2.7
-        "test_cublas_addmm_reduced_precision_fp16_accumulate_size_10000_xpu_bfloat16",
-        "test_cublas_addmm_reduced_precision_fp16_accumulate_size_10000_xpu_float16",
-        "test_cublas_addmm_reduced_precision_fp16_accumulate_size_1000_xpu_bfloat16",
-        "test_cublas_addmm_reduced_precision_fp16_accumulate_size_1000_xpu_float16",
-        "test_cublas_addmm_reduced_precision_fp16_accumulate_size_100_xpu_bfloat16",
-        "test_cublas_addmm_reduced_precision_fp16_accumulate_size_100_xpu_float16",
-        "test_cublas_and_lt_reduced_precision_fp16_accumulate_xpu",
-    ),
+    #    "test_matmul_cuda_xpu.py": (
+    #        # AssertionError: "Bias is not supported when out_dtype is set to Float32" does not match "Could not run 'aten::_scaled_mm' with arguments from the 'CPU' backend.
+    #        "test_float32_output_errors_with_bias_xpu",
+    #        # RuntimeError: "eye" not implemented for 'Float8_e4m3fn'
+    #        "test_float8_basics_xpu",
+    #        # AssertionError: "For row-wise scaling, scale_a must be size 1024 but got 1 and scale_b must be size 2048 but got 2" does not match "Could not run 'aten::_scaled_mm' with arguments from the 'CPU' backend.
+    #        "test_float8_error_messages_xpu",
+    #        # NotImplementedError: Could not run 'aten::_scaled_mm' with arguments from the 'CPU' backend.
+    #        "test_float8_bias_relu_edgecase_xpu",
+    #        "test_float8_bias_xpu",
+    #        "test_float8_rowwise_scaling_sanity_use_fast_accum_False_xpu",
+    #        "test_float8_rowwise_scaling_sanity_use_fast_accum_True_xpu",
+    #        "test_float8_scale_fast_accum_xpu",
+    #        "test_float8_scale_xpu",
+    #        "test_non_divisible_leading_dim_bias_False_xpu",
+    #        "test_non_divisible_leading_dim_bias_True_xpu",
+    #        "test_scaled_mm_change_stride_bfloat16_xpu",
+    #        "test_scaled_mm_change_stride_float16_xpu",
+    #        "test_scaled_mm_change_stride_float32_xpu",
+    #        "test_scaled_mm_vs_emulated_bfloat16_xpu",
+    #        "test_scaled_mm_vs_emulated_float16_xpu",
+    #        "test_scaled_mm_vs_emulated_float32_xpu",
+    #        "test_scaled_mm_vs_emulated_row_wise_bfloat16_xpu",
+    #        # AssertionError: Torch not compiled with CUDA enabled
+    #        "test_zero_dim_tensorwise_which_dim_zero",
+    #        # New added case in 2.7
+    #        "test_cublas_addmm_reduced_precision_fp16_accumulate_size_10000_xpu_bfloat16",
+    #        "test_cublas_addmm_reduced_precision_fp16_accumulate_size_10000_xpu_float16",
+    #        "test_cublas_addmm_reduced_precision_fp16_accumulate_size_1000_xpu_bfloat16",
+    #        "test_cublas_addmm_reduced_precision_fp16_accumulate_size_1000_xpu_float16",
+    #        "test_cublas_addmm_reduced_precision_fp16_accumulate_size_100_xpu_bfloat16",
+    #        "test_cublas_addmm_reduced_precision_fp16_accumulate_size_100_xpu_float16",
+    #        "test_cublas_and_lt_reduced_precision_fp16_accumulate_xpu",
+    #    ),
     "test_maskedtensor_xpu.py": None,
     "quantization/core/test_quantized_op_xpu.py": (
         # AssertionError: Torch not compiled with CUDA enabled

diff --git a/test/xpu/xpu_test_utils.py b/test/xpu/xpu_test_utils.py
@@ -424,6 +424,11 @@
             torch.float32: tol(atol=3e-5, rtol=5e-5),
         }
     },
+    "nn.ConvTranspose3d": {
+        ("TestModule", "test_non_contiguous_tensors"): {
+            torch.float32: tol(atol=2e-5, rtol=5e-5),
+        }
+    },
 }