Scott Todd
Scott Todd
Would you happen to have the `model.mlir` on hand for this? I just hit a similar error where my program has `torch.onnx_meta.ir_version = 7` but the lowering pattern requires IR...
Here is the program (test case) I'm using: ```mlir module { func.func @test_loop13_seq(%arg0: !torch.vtensor, %arg1: !torch.vtensor, %arg2: !torch.list) -> !torch.list attributes {torch.onnx_meta.ir_version = 7 : si64, torch.onnx_meta.opset_version = 13 :...
> @ScottTodd I think this ticket can be closed, because of the improvements that were made in: #16452 Sort of - I only changed the CMake configuration. The compiler code...
> @ScottTodd can probably provide the files/instructions For `test_convtranspose_1d` * Download files from https://github.com/ScottTodd/SHARK-TestSuite/tree/test-githubaction/iree_tests/onnx/node/generated/test_convtranspose_1d * In that directory, `iree-compile model.mlir --iree-hal-target-backends=llvm-cpu -o model_cpu.vmfb` * In that directory, `iree-run-module --module=model_cpu.vmfb --device=local-task...
CI found some test failures: https://github.com/openxla/iree/actions/runs/8086732817/job/22097146456?pr=16607#step:5:151 ``` 20/1395 Test #62: torch-iree/InputConversion/test/scatter.mlir.test ..........................................................***Failed 0.35 sec -- Testing: 1 tests, 1 workers -- FAIL: IREE :: plugins/input/Torch/torch-iree/InputConversion/test/scatter.mlir (1 of 1) ******************** TEST...
FYI, I regenerated the external ONNX test suite using this torch-mlir commit: https://github.com/nod-ai/SHARK-TestSuite/pull/84. (prefetching this assuming nothing else is broken - please let me know if you find anything)
Are these failures legit? https://github.com/openxla/iree/actions/runs/8144463146/job/22259150196?pr=16607#step:6:49 I'll retry failed just in case...
> Are these failures legit? https://github.com/openxla/iree/actions/runs/8144463146/job/22259150196?pr=16607#step:6:49 > > I'll retry failed just in case... Oh, maybe preexisting + fixed by https://github.com/openxla/iree/pull/16661 .... phew. Thanks @antiagainst !
Yeah that failure was preexisting and should be fixed now. Could sync this PR.
> so I should just rebase yeah? Yes.