[iree_tests/onnx/node] Mass convert dims loaded from .npy files into onnx.Constant built into the model.mlir

[renxida]

We have a lot of tests in SHARK-TestSuite/iree_tests/onnx/node/generated/ like https://github.com/nod-ai/SHARK-TestSuite/blob/main/iree_tests/onnx/node/generated/test_reduce_l1_do_not_keepdims_example/model.mlir that uses a dim specified as an argument to the func.func & loaded from a npy file.

Since torch_mlir requires a lot of dims to be constant in order to compile, these tests fail, causing a lot of false positives.

I wonder if there is some sort of mass solution that could convert the args to onnx.Constant like:

└──╼ $git diff ./model.mlir
diff --git a/iree_tests/onnx/node/generated/test_reduce_l1_do_not_keepdims_example/model.mlir b/iree_tests/onnx/node/generated/test_reduce_l1_do_not_keepdims_example/model.mlir
index 182a1fc6..8ee10c96 100644
--- a/iree_tests/onnx/node/generated/test_reduce_l1_do_not_keepdims_example/model.mlir
+++ b/iree_tests/onnx/node/generated/test_reduce_l1_do_not_keepdims_example/model.mlir
@@ -1,5 +1,6 @@
 module {
-  func.func @test_reduce_l1_do_not_keepdims_example(%arg0: !torch.vtensor<[3,2,2],f32>, %arg1: !torch.vtensor<[1],si64>) -> !torch.vtensor<[3,2],f32> attributes {torch.onnx_meta.ir_version = 8 : si64, torch.onnx_meta.opset_version = 18 : si64, torch.onnx_meta.producer_name = "backend-test", torch.onnx_meta.producer_version = ""} {
+  func.func @test_reduce_l1_do_not_keepdims_example(%arg0: !torch.vtensor<[3,2,2],f32>) -> !torch.vtensor<[3,2],f32> attributes {torch.onnx_meta.ir_version = 8 : si64, torch.onnx_meta.opset_version = 18 : si64, torch.onnx_meta.producer_name = "backend-test", torch.onnx_meta.producer_version = ""} {
+    %arg1 = "torch.operator"() <{name = "onnx.Constant"}> {torch.onnx.value = dense<2> : tensor<1xsi64>} : () -> !torch.vtensor<[1],si64>
     %none = torch.constant.none
     %0 = torch.operator "onnx.ReduceL1"(%arg0, %arg1) {torch.onnx.keepdims = 0 : si64} : (!torch.vtensor<[3,2,2],f32>, !torch.vtensor<[1],si64>) -> !torch.vtensor<[3,2],f32> 
     return %0 : !torch.vtensor<[3,2],f32>

spicifically, instead of taking arg1 from the func arguments, we have:

%arg1 = "torch.operator"() <{name = "onnx.Constant"}> {

which causes the test cases to pass

[ScottTodd]

Don't change the tests to make them pass :P

Since torch_mlir requires a lot of dims to be constant in order to compile, these tests fail, causing a lot of false positives.

Those sound like true positives, not false positives? (edit: or true negatives vs false negatives? 🤔)

[renxida]

Yup that makes sense. Opening torch-mlir issue to fix this llvm/torch-mlir#3573

[ScottTodd]

Nothing to do here, closing this issue.
(and iree_tests/onnx/node moved to https://github.com/iree-org/iree-test-suites)