E2E tests failing on main branch (Commit: b834f94b)

[hariprasadravi]

Hello,

I am encountering failures in the end-to-end (e2e) test suite on the current HEAD of the repository. The build completes successfully but e2e tests fail with the current syndrome:

./tools/e2e_test.sh

TORCH_VERSION_FOR_COMPARISON = 2.10.0.dev20251016
TORCH_VERSION_FOR_COMPARISON = 2.10.0.dev20251016
/local-ssd/hravisha/torch-mlir-upstream/mlir_venv/lib/python3.11/site-packages/torch/backends/mkldnn/__init__.py:89: UserWarning: TF32 acceleration on top of oneDNN is available for Intel GPUs. The current Torch version does not have Intel GPU Support. (Triggered internally at /pytorch/aten/src/ATen/Context.cpp:243.)
  torch._C._set_onednn_allow_tf32(_allow_tf32)
At least one of testing processes has exited with code != 0.

./tools/e2e_test.sh --filter Conv2d --verbose

      RuntimeError: 
        attribute lookup is not defined on builtin:
          File "/local-ssd/hravisha/torch-mlir-upstream/build/tools/torch-mlir/python_packages/torch_mlir/torch_mlir_e2e_test/test_suite/conv.py", line 1769
            def _forward(self, inputVec, weight, scales, zeropoints, bias):
                inputVec = torch.ops.quantized_decomposed.dequantize_per_tensor.default(
                           ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
                    inputVec, 0.01, 7, -128, 127, torch.int8
                )

Environment

OS: Debian 12

Commit Hash: b834f94

Python Version: Python 3.11.2

PyTorch Version: 2.10.0.dev20251016

Build Type: Release with Assertions

Reproduction Steps

Clone/Checkout llvm/torch-mlir at HEAD.

Sync submodules (git submodule update --init --recursive).

Build project using:

    cmake -GNinja -Bbuild \
            -DCMAKE_BUILD_TYPE=RelWithDebInfo \
            -DPython3_FIND_VIRTUALENV=ONLY \
            -DPython_FIND_VIRTUALENV=ONLY \
            -DLLVM_ENABLE_PROJECTS=mlir \
            -DLLVM_EXTERNAL_PROJECTS="torch-mlir" \
            -DLLVM_EXTERNAL_TORCH_MLIR_SOURCE_DIR="$PWD" \
            -DMLIR_ENABLE_BINDINGS_PYTHON=ON \
            -DLLVM_TARGETS_TO_BUILD=host \
            externals/llvm-project/llvm \
            -DCMAKE_C_COMPILER=clang \
            -DCMAKE_CXX_COMPILER=clang++ \
            -DCMAKE_C_COMPILER_LAUNCHER=ccache \
            -DCMAKE_CXX_COMPILER_LAUNCHER=ccache \
            -DLLVM_USE_LINKER=lld \
            -DLLVM_ENABLE_ASSERTIONS=ON \
            -DTORCH_MLIR_ENABLE_JIT_IR_IMPORTER=ON \
            -DTORCH_MLIR_ENABLE_PYTORCH_EXTENSIONS=ON \
            -DCMAKE_EXPORT_COMPILE_COMMANDS=ON
    cd build        
    cmake --build .

Run the e2e tests:

https://github.com/llvm/torch-mlir/blob/main/docs/development.md#running-execution-end-to-end-tests

[sahas3]

@hariprasadravi this failure is expected since the default e2e tests are running the torchscript path which has been deprecated. For now you can run the fx_importer/fx_importer_tosa/or other such configs as ./tools/e2e_test.sh --filter Conv2d --verbose --config fx_importer.