onnx.If errors when the two branches yield with different shapes #696

renxida · 2024-05-21T01:52:05Z

I was working on #566

Found the problem in KeypointRCNN (gist with stripped IR at %5503 (the above link takes you to the correct line.

When lowering an If op with two branches returning different types, we encounter:

./repro3.mlir:2:10: error: 'torch.prim.If' op  along control flow edge from Region #0 to parent results: source type #0 '!torch.vtensor<[1],si64>' should match input type #0 '!torch.vtensor<[?],si64>'
    %0 = torch.operator "onnx.If"(%arg0) : (!torch.vtensor<[1],i1>) -> !torch.vtensor<[?],si64> {
         ^
./repro3.mlir:2:10: note: see current operation: 
%2 = "torch.prim.If"(%1) ({
  %3 = "torch.vtensor.literal"() <{value = dense<0> : tensor<1xsi64>}> : () -> !torch.vtensor<[1],si64>
  "torch.prim.If.yield"(%3) : (!torch.vtensor<[1],si64>) -> ()
}, {
  "torch.prim.If.yield"(%arg1) : (!torch.vtensor<[?],si64>) -> ()
}) : (!torch.bool) -> !torch.vtensor<[?],si64>
// -----// IR Dump After ConvertTorchOnnxToTorch Failed (convert-torch-onnx-to-torch) //----- //
"func.func"() <{function_type = (!torch.vtensor<[1],i1>, !torch.vtensor<[?],si64>) -> !torch.vtensor<[?],si64>, sym_name = "minimal_example"}> ({
^bb0(%arg0: !torch.vtensor<[1],i1>, %arg1: !torch.vtensor<[?],si64>):
  %0 = "torch.aten.item"(%arg0) : (!torch.vtensor<[1],i1>) -> !torch.int
  %1 = "torch.aten.Bool.int"(%0) : (!torch.int) -> !torch.bool
  %2 = "torch.prim.If"(%1) ({
    %3 = "torch.vtensor.literal"() <{value = dense<0> : tensor<1xsi64>}> : () -> !torch.vtensor<[1],si64>
    "torch.prim.If.yield"(%3) : (!torch.vtensor<[1],si64>) -> ()
  }, {
    "torch.prim.If.yield"(%arg1) : (!torch.vtensor<[?],si64>) -> ()
  }) : (!torch.bool) -> !torch.vtensor<[?],si64>
  "func.return"(%2) : (!torch.vtensor<[?],si64>) -> ()
}) {torch.onnx_meta.ir_version = 8 : si64, torch.onnx_meta.opset_version = 17 : si64, torch.onnx_meta.producer_name = "pytorch", torch.onnx_meta.producer_version = "1.13.1"} : () -> ()

Reproducer:

func.func @minimal_example(%arg0: !torch.vtensor<[1],i1>, %arg1: !torch.vtensor<[?], si64>) -> !torch.vtensor<[?],si64> attributes {torch.onnx_meta.ir_version = 8 : si64, torch.onnx_meta.opset_version = 17 : si64, torch.onnx_meta.producer_name = "pytorch", torch.onnx_meta.producer_version = "1.13.1"} {
    %0 = torch.operator "onnx.If"(%arg0) : (!torch.vtensor<[1],i1>) -> !torch.vtensor<[?],si64> {
        torch.operator_terminator %arg1 : !torch.vtensor<[?],si64>
    }, {
        %1 = torch.operator "onnx.Constant"() {torch.onnx.value = dense<0> : tensor<1xsi64>} : () -> !torch.vtensor<[1],si64>
        torch.operator_terminator %1 : !torch.vtensor<[1],si64>
    }
    return %0 : !torch.vtensor<[?],si64>
}

The text was updated successfully, but these errors were encountered:

renxida · 2024-05-21T01:55:31Z

I'm not sure if we want to support this, but if we do I think we need to insert a type conversion from vtensor<[1], si64> to vtensor<[?], si64>. Not sure how to materialize that.

renxida · 2024-05-30T18:41:51Z

KeypointRCNN_vaiq_int8 has if statements that returns not just different shapes but different ranks. We can't support that and will need something to be done with the model.

renxida · 2024-06-04T16:38:37Z

Stella mentions a way to deal with similar problems

https://discord.com/channels/973663919757492264/1173330951791706113/1246504997269798945

and @zjgarvey's comment in this morning's meeting got me to put 2 and 2 together. Will try to just remove the small branch of these onnx.If

renxida · 2024-10-23T02:52:31Z

I vaguely remember the concensus is that this is a model issue and should be fixed by editing the model. Perhaps ask sharktank folks for help?

AmosLewis · 2024-10-23T02:55:18Z

reproducer extract from coat_mini model
The onnx.if op is the last 2 op in the model.

module {
  func.func @torch_jit(%6768: !torch.vtensor<[1],si64>, %6765: !torch.vtensor<[1,1,216],f32>) -> !torch.vtensor<[],f32> attributes {torch.onnx_meta.ir_version = 7 : si64, torch.onnx_meta.opset_version = 21 : si64, torch.onnx_meta.producer_name = "pytorch", torch.onnx_meta.producer_version = "1.12.1"} {
    %6769 = torch.operator "onnx.Constant"() {torch.onnx.value = dense_resource<__2134> : tensor<1xsi64>} : () -> !torch.vtensor<[1],si64> 
    %6770 = torch.operator "onnx.Equal"(%6768, %6769) : (!torch.vtensor<[1],si64>, !torch.vtensor<[1],si64>) -> !torch.vtensor<[1],i1> 
    %6771 = torch.operator "onnx.If"(%6770) : (!torch.vtensor<[1],i1>) -> !torch.vtensor<[],f32> {
      %6773 = torch.operator "onnx.Identity"(%6765) : (!torch.vtensor<[1,1,216],f32>) -> !torch.vtensor<[1,1,216],f32> 
      torch.operator_terminator %6773 : !torch.vtensor<[1,1,216],f32>
    }, {
      %6773 = torch.operator "onnx.Constant"() {torch.onnx.value = dense_resource<__2135> : tensor<1xsi64>} : () -> !torch.vtensor<[1],si64> 
      %6774 = torch.operator "onnx.Squeeze"(%6765, %6773) : (!torch.vtensor<[1,1,216],f32>, !torch.vtensor<[1],si64>) -> !torch.vtensor<[1,216],f32> 
      torch.operator_terminator %6774 : !torch.vtensor<[1,216],f32>
    }
    return %6771: !torch.vtensor<[],f32>
  }
}

{-#
  dialect_resources: {
    builtin: {
      __2134: "0x080000000100000000000000"
    }
  }
#-}

`torch-mlir-opt --convert-torch-onnx-to-torch if.onnx.mlir --debug

if.onnx.mlir:5:13: error: 'torch.prim.If' op  along control flow edge from Region #0 to parent results: source type #0 '!torch.vtensor<[1,216],f32>' should match input type #0 '!torch.vtensor<[],f32>'
    %6771 = torch.operator "onnx.If"(%6770) : (!torch.vtensor<[1],i1>) -> !torch.vtensor<[],f32> {
            ^
if.onnx.mlir:5:13: note: see current operation: 
%4 = "torch.prim.If"(%3) ({
  %7 = "torch.vtensor.literal"() <{value = dense_resource<__onnx_constant_not_found_possibly_due_to_being_elided__> : tensor<1xsi64>}> : () -> !torch.vtensor<[1],si64>
  %8 = "torch.constant.int"() <{value = 1 : i64}> : () -> !torch.int
  %9 = "torch.prim.ListConstruct"(%8) : (!torch.int) -> !torch.list<int>
  %10 = "torch.prims.squeeze"(%arg1, %9) : (!torch.vtensor<[1,1,216],f32>, !torch.list<int>) -> !torch.vtensor<[1,216],f32>
  "torch.prim.If.yield"(%10) : (!torch.vtensor<[1,216],f32>) -> ()
}, {
  %5 = "torch.constant.none"() : () -> !torch.none
  %6 = "torch.aten.clone"(%arg1, %5) : (!torch.vtensor<[1,1,216],f32>, !torch.none) -> !torch.vtensor<[1,1,216],f32>
  "torch.prim.If.yield"(%6) : (!torch.vtensor<[1,1,216],f32>) -> ()
}) : (!torch.bool) -> !torch.vtensor<[],f32>
```

AmosLewis · 2024-10-23T02:58:06Z

I vaguely remember the concensus is that this is a model issue and should be fixed by editing the model. Perhaps ask sharktank folks for help?

we have 6 models https://github.com/pdhirajkumarprasad/SHARK-TestSuite/blob/feature/qa/issue/onnx-to-torch/onnx.if failed. Not sure if it is feasible to edit all the models. will ask tomorrow.

zjgarvey · 2024-10-23T05:01:46Z

Yeah, we could try adding the models to the basic opt list in the test suite `azure_models.py`

…

On Tue, Oct 22, 2024, 9:58 PM Chi_Liu ***@***.***> wrote: I vaguely remember the concensus is that this is a model issue and should be fixed by editing the model. Perhaps ask sharktank folks for help? we have 6 models https://github.com/pdhirajkumarprasad/SHARK-TestSuite/blob/feature/qa/issue/onnx-to-torch/onnx.if failed. Not sure if it is the right way to edit all the models. We ask tomorrow. — Reply to this email directly, view it on GitHub <#696 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ALODRYIK6DRMAE3SE26QWRDZ44GFJAVCNFSM6AAAAABIAVWOZKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIMZQG4ZDKMBQGY> . You are receiving this because you were mentioned.Message ID: ***@***.***>

AmosLewis · 2024-10-24T18:06:45Z

Yeah, we could try adding the models to the basic opt list in the test suite azure_models.py

With adding to basic_opt list, 2 models "coat_mini", "coat_tiny", pass the comiple and then get FAILED (Numerics) .
Others got this kind of bug:

python ./run.py --mode=cl-onnx-iree -v -t model--splinter-large-few-shot-k-16-finetuned-squad-seed-0--anas-awadalla --torchtolinalg
Traceback (most recent call last):
  File "/proj/gdba/shark/chi/src/SHARK-TestSuite/alt_e2eshark/./run.py", line 481, in <module>
    main(parser.parse_args())
  File "/proj/gdba/shark/chi/src/SHARK-TestSuite/alt_e2eshark/./run.py", line 114, in main
    test_list = get_tests(args.groups, args.test_filter, args.testsfile)
  File "/proj/gdba/shark/chi/src/SHARK-TestSuite/alt_e2eshark/./run.py", line 70, in get_tests
    from onnx_tests.models import model
  File "/proj/gdba/shark/chi/src/SHARK-TestSuite/alt_e2eshark/onnx_tests/models/model.py", line 12, in <module>
    from .nlp import *
  File "/proj/gdba/shark/chi/src/SHARK-TestSuite/alt_e2eshark/onnx_tests/models/nlp.py", line 107, in <module>
    register_test(dim_param_constructor(default_nlp_params), model_name)
  File "/proj/gdba/shark/chi/src/SHARK-TestSuite/alt_e2eshark/e2e_testing/registry.py", line 17, in register_test
    raise ValueError(
ValueError: Duplicate test name: 'model--splinter-large-few-shot-k-16-finetuned-squad-seed-0--anas-awadalla'. Please make sure that the function wrapped by `register_test` has a unique name.

zjgarvey · 2024-10-24T18:42:07Z

Ah, I see. For nlp models, we need to register those differently.

zjgarvey · 2024-10-24T19:08:19Z

I tried doing basic optimizations for those nlp models locally, but it still has the problematic if statement, so it's likely unable to fold it when dynamic dims are present.

Add a PR to do those two static dim models as basic_opt tests for now.

AmosLewis · 2024-10-24T20:05:58Z

Add a PR to do those two static dim models as basic_opt tests for now.

nod-ai/SHARK-TestSuite#378

Fix the onnx.if different shapes issue nod-ai/SHARK-ModelDev#696. 4 dynamic models still left unfixed.

AmosLewis · 2024-10-25T00:07:06Z

4 related model--splinter-large-few-shot-k-16-finetuned-squad-seed-0--anas-awadalla, similar failed pattern:
!torch.vtensor<[?,?],f32> / !torch.vtensor<[?,?,?],f32> -> !torch.vtensor<[],f32>

%1900 = torch.operator "onnx.If"(%1899) : (!torch.vtensor<[1],i1>) -> !torch.vtensor<[],f32> {
  %1946 = torch.operator "onnx.Identity"(%1896) : (!torch.vtensor<[?,?,?],f32>) -> !torch.vtensor<[?,?,?],f32> 
  torch.operator_terminator %1946 : !torch.vtensor<[?,?,?],f32>
}, {
  %1946 = torch.operator "onnx.Constant"() {torch.onnx.value = dense<1> : tensor<1xsi64>} : () -> !torch.vtensor<[1],si64> 
  %1947 = torch.operator "onnx.Squeeze"(%1896, %1946) : (!torch.vtensor<[?,?,?],f32>, !torch.vtensor<[1],si64>) -> !torch.vtensor<[?,?],f32> 
  torch.operator_terminator %1947 : !torch.vtensor<[?,?],f32>
}

->

/proj/gdba/shark/chi/src/SHARK-TestSuite/alt_e2eshark/test-run/model--splinter-large-few-shot-k-16-finetuned-squad-seed-0--anas-awadalla/model.torch_onnx.mlir:1904:13: error: 'torch.prim.If' op  along control flow edge from Region #0 to parent results: source type #0 '!torch.vtensor<[?,?],f32>' should match input type #0 '!torch.vtensor<[],f32>'
    %1900 = torch.operator "onnx.If"(%1899) : (!torch.vtensor<[1],i1>) -> !torch.vtensor<[],f32> {
            ^
/proj/gdba/shark/chi/src/SHARK-TestSuite/alt_e2eshark/test-run/model--splinter-large-few-shot-k-16-finetuned-squad-seed-0--anas-awadalla/model.torch_onnx.mlir:1904:13: note: see current operation: 
%7620 = "torch.prim.If"(%7619) ({
  %7737 = "torch.vtensor.literal"() <{value = dense<1> : tensor<1xsi64>}> : () -> !torch.vtensor<[1],si64>
  %7738 = "torch.constant.int"() <{value = 0 : i64}> : () -> !torch.int
  %7739 = "torch.constant.int"() <{value = 0 : i64}> : () -> !torch.int
  %7740 = "torch.aten.select.int"(%7737, %7738, %7739) : (!torch.vtensor<[1],si64>, !torch.int, !torch.int) -> !torch.vtensor<[1],si64>
  %7741 = "torch.aten.item"(%7740) : (!torch.vtensor<[1],si64>) -> !torch.int
  %7742 = "torch.prim.ListConstruct"(%7741) : (!torch.int) -> !torch.list<int>
  %7743 = "torch.prims.squeeze"(%7604, %7742) : (!torch.vtensor<[?,?,?],f32>, !torch.list<int>) -> !torch.vtensor<[?,?],f32>
  "torch.prim.If.yield"(%7743) : (!torch.vtensor<[?,?],f32>) -> ()
}, {
  %7735 = "torch.constant.none"() : () -> !torch.none
  %7736 = "torch.aten.clone"(%7604, %7735) : (!torch.vtensor<[?,?,?],f32>, !torch.none) -> !torch.vtensor<[?,?,?],f32>
  "torch.prim.If.yield"(%7736) : (!torch.vtensor<[?,?,?],f32>) -> ()
}) : (!torch.bool) -> !torch.vtensor<[],f32>

AmosLewis · 2024-10-28T16:09:02Z

@jinchen62 you can take this 2. retinanet_resnet50_fpn_vaiq_int8 and KeypointRCNN_vaiq_int8 failed pattern are same, one branch dynamic/one branch static, return dynamic
[?]/[0] -> [?]

%6272 = torch.operator "onnx.If"(%6271) : (!torch.vtensor<[],i1>) -> !torch.vtensor<[?],si64> {
  %6346 = torch.operator "onnx.ReduceMax"(%6264) {torch.onnx.keepdims = 0 : si64} : (!torch.vtensor<[?,?],f32>) -> !torch.vtensor<[],f32> 
  %6347 = torch.operator "onnx.Cast"(%6266) {torch.onnx.to = 1 : si64} : (!torch.vtensor<[?],si64>) -> !torch.vtensor<[?],f32> 
  %6348 = torch.operator "onnx.Constant"() {torch.onnx.value = dense_resource<__3571> : tensor<f32>} : () -> !torch.vtensor<[],f32> 
  %6349 = torch.operator "onnx.Add"(%6346, %6348) : (!torch.vtensor<[],f32>, !torch.vtensor<[],f32>) -> !torch.vtensor<[],f32> 
  %6350 = torch.operator "onnx.Mul"(%6347, %6349) : (!torch.vtensor<[?],f32>, !torch.vtensor<[],f32>) -> !torch.vtensor<[?],f32> 
  %6351 = torch.operator "onnx.Constant"() {torch.onnx.value = dense_resource<__3572> : tensor<1xsi64>} : () -> !torch.vtensor<[1],si64> 
  %6352 = torch.operator "onnx.Unsqueeze"(%6350, %6351) : (!torch.vtensor<[?],f32>, !torch.vtensor<[1],si64>) -> !torch.vtensor<[?,1],f32> 
  %6353 = torch.operator "onnx.Add"(%6264, %6352) : (!torch.vtensor<[?,?],f32>, !torch.vtensor<[?,1],f32>) -> !torch.vtensor<[?,?],f32> 
  %6354 = torch.operator "onnx.Constant"() {torch.onnx.value = dense_resource<__3573> : tensor<1xsi64>} : () -> !torch.vtensor<[1],si64> 
  %6355 = torch.operator "onnx.Unsqueeze"(%6353, %6354) : (!torch.vtensor<[?,?],f32>, !torch.vtensor<[1],si64>) -> !torch.vtensor<[1,?,?],f32> 
  %6356 = torch.operator "onnx.Constant"() {torch.onnx.value = dense_resource<__3574> : tensor<1xsi64>} : () -> !torch.vtensor<[1],si64> 
  %6357 = torch.operator "onnx.Unsqueeze"(%6265, %6356) : (!torch.vtensor<[?],f32>, !torch.vtensor<[1],si64>) -> !torch.vtensor<[1,?],f32> 
  %6358 = torch.operator "onnx.Constant"() {torch.onnx.value = dense_resource<__3575> : tensor<1xsi64>} : () -> !torch.vtensor<[1],si64> 
  %6359 = torch.operator "onnx.Unsqueeze"(%6357, %6358) : (!torch.vtensor<[1,?],f32>, !torch.vtensor<[1],si64>) -> !torch.vtensor<[1,1,?],f32> 
  %6360 = torch.operator "onnx.Constant"() {torch.onnx.value = dense_resource<__3576> : tensor<1xsi64>} : () -> !torch.vtensor<[1],si64> 
  %6361 = torch.operator "onnx.Constant"() {torch.onnx.value = dense_resource<__3577> : tensor<1xf32>} : () -> !torch.vtensor<[1],f32> 
  %6362 = torch.operator "onnx.NonMaxSuppression"(%6355, %6359, %6360, %6361) : (!torch.vtensor<[1,?,?],f32>, !torch.vtensor<[1,1,?],f32>, !torch.vtensor<[1],si64>, !torch.vtensor<[1],f32>) -> !torch.vtensor<[?,3],si64> 
  %6363 = torch.operator "onnx.Constant"() {torch.onnx.value = dense_resource<__3578> : tensor<1xsi64>} : () -> !torch.vtensor<[1],si64> 
  %6364 = torch.operator "onnx.Gather"(%6362, %6363) {torch.onnx.axis = 1 : si64} : (!torch.vtensor<[?,3],si64>, !torch.vtensor<[1],si64>) -> !torch.vtensor<[?,1],si64> 
  %6365 = torch.operator "onnx.Constant"() {torch.onnx.value = dense_resource<__3579> : tensor<1xsi64>} : () -> !torch.vtensor<[1],si64> 
  %6366 = torch.operator "onnx.Squeeze"(%6364, %6365) : (!torch.vtensor<[?,1],si64>, !torch.vtensor<[1],si64>) -> !torch.vtensor<[?],si64> 
  torch.operator_terminator %6366 : !torch.vtensor<[?],si64>
}, {
  %6346 = torch.operator "onnx.Constant"() {torch.onnx.value = dense_resource<__3580> : tensor<0xsi64>} : () -> !torch.vtensor<[0],si64> 
  torch.operator_terminator %6346 : !torch.vtensor<[0],si64>
}

AmosLewis · 2024-10-30T01:09:34Z

4 related model--splinter-large-few-shot-k-16-finetuned-squad-seed-0--anas-awadalla, similar failed pattern: !torch.vtensor<[?,?],f32> / !torch.vtensor<[?,?,?],f32> -> !torch.vtensor<[],f32>

With onnx-modifier.exe, This can be fixed by remove if node and add the squeeze node(in if then branch) pass its output to the next add node as input directly. Here is the test output.

python ./run.py --mode=cl-onnx-iree -v -t model--splinter-large-few-shot-k-16-finetuned-squad-seed-0--anas-awadalla --torchtolinalg  


Stages to be run: ['setup', 'import_model', 'preprocessing', 'compilation', 'construct_inputs', 'native_inference', 'compiled_inference', 'postprocessing']
Test list: ['model--splinter-large-few-shot-k-16-finetuned-squad-seed-0--anas-awadalla']
running test model--splinter-large-few-shot-k-16-finetuned-squad-seed-0--anas-awadalla...
Unzipping - /proj/gdba/shark/chi/src/SHARK-TestSuite/alt_e2eshark/tmp/model--splinter-large-few-shot-k-16-finetuned-squad-seed-0--anas-awadalla/model.onnx.zip... 
Unzipping succeded. Look for extracted contents in /proj/gdba/shark/chi/src/SHARK-TestSuite/alt_e2eshark/test-run/model--splinter-large-few-shot-k-16-finetuned-squad-seed-0--anas-awadalla
2024-10-29 17:46:40.456511164 [W:onnxruntime:, graph.cc:4093 CleanUnusedInitializersAndNodeArgs] Removing initializer '/Constant_4_output_0'. It is not used by any node and should be removed from the model.
2024-10-29 17:46:40.456554925 [W:onnxruntime:, graph.cc:4093 CleanUnusedInitializersAndNodeArgs] Removing initializer '/Constant_2_output_0'. It is not used by any node and should be removed from the model.
2024-10-29 17:46:40.456562795 [W:onnxruntime:, graph.cc:4093 CleanUnusedInitializersAndNodeArgs] Removing initializer '/Constant_3_output_0'. It is not used by any node and should be removed from the model.
2024-10-29 17:46:40.456585936 [W:onnxruntime:, graph.cc:4093 CleanUnusedInitializersAndNodeArgs] Removing initializer '/Constant_1_output_0'. It is not used by any node and should be removed from the model.
2024-10-29 18:02:22.548917921 [W:onnxruntime:, graph.cc:4093 CleanUnusedInitializersAndNodeArgs] Removing initializer '/Constant_4_output_0'. It is not used by any node and should be removed from the model.
2024-10-29 18:02:22.548954032 [W:onnxruntime:, graph.cc:4093 CleanUnusedInitializersAndNodeArgs] Removing initializer '/Constant_2_output_0'. It is not used by any node and should be removed from the model.
2024-10-29 18:02:22.548960962 [W:onnxruntime:, graph.cc:4093 CleanUnusedInitializersAndNodeArgs] Removing initializer '/Constant_3_output_0'. It is not used by any node and should be removed from the model.
2024-10-29 18:02:22.548981482 [W:onnxruntime:, graph.cc:4093 CleanUnusedInitializersAndNodeArgs] Removing initializer '/Constant_1_output_0'. It is not used by any node and should be removed from the model.
2024-10-29 18:02:24.037809661 [W:onnxruntime:, graph.cc:4093 CleanUnusedInitializersAndNodeArgs] Removing initializer '/Constant_4_output_0'. It is not used by any node and should be removed from the model.
2024-10-29 18:02:24.037850531 [W:onnxruntime:, graph.cc:4093 CleanUnusedInitializersAndNodeArgs] Removing initializer '/Constant_2_output_0'. It is not used by any node and should be removed from the model.
2024-10-29 18:02:24.037858882 [W:onnxruntime:, graph.cc:4093 CleanUnusedInitializersAndNodeArgs] Removing initializer '/Constant_3_output_0'. It is not used by any node and should be removed from the model.
2024-10-29 18:02:24.037881712 [W:onnxruntime:, graph.cc:4093 CleanUnusedInitializersAndNodeArgs] Removing initializer '/Constant_1_output_0'. It is not used by any node and should be removed from the model.
2024-10-29 18:02:25.649182399 [W:onnxruntime:, graph.cc:4093 CleanUnusedInitializersAndNodeArgs] Removing initializer '/Constant_4_output_0'. It is not used by any node and should be removed from the model.
2024-10-29 18:02:25.649218780 [W:onnxruntime:, graph.cc:4093 CleanUnusedInitializersAndNodeArgs] Removing initializer '/Constant_2_output_0'. It is not used by any node and should be removed from the model.
2024-10-29 18:02:25.649228320 [W:onnxruntime:, graph.cc:4093 CleanUnusedInitializersAndNodeArgs] Removing initializer '/Constant_3_output_0'. It is not used by any node and should be removed from the model.
2024-10-29 18:02:25.649250351 [W:onnxruntime:, graph.cc:4093 CleanUnusedInitializersAndNodeArgs] Removing initializer '/Constant_1_output_0'. It is not used by any node and should be removed from the model.
        FAILED (Numerics)                    

Test Summary:
        PASSES: 0
        TOTAL: 1
results stored in /proj/gdba/shark/chi/src/SHARK-TestSuite/alt_e2eshark/test-run

renxida mentioned this issue May 21, 2024

[tracking] E2EShark Model Tests Onnx Mode #566

Open

6 tasks

renxida mentioned this issue Jun 4, 2024

Elide bad onnx.If nodes with different shapes on two branches llvm/torch-mlir#3419

Closed

AmosLewis mentioned this issue Aug 9, 2024

[TRACKER] Onnx Op' Torch->Linalg Lowering #797

Open

jinchen62 mentioned this issue Aug 23, 2024

[tracking] HF CNN fp32 model tests #801

Open

30 tasks

kumardeepakamd mentioned this issue Sep 10, 2024

failed to legalize operation onnx.If #815

Closed

kumardeepakamd assigned renxida Sep 10, 2024

pdhirajkumarprasad mentioned this issue Sep 11, 2024

[Tracker] All the issue related with e2e shark test suite #812

Open

AmosLewis mentioned this issue Oct 24, 2024

Fix the coat model with onnx.if different shapes bug nod-ai/SHARK-TestSuite#378

Merged

AmosLewis added a commit to nod-ai/SHARK-TestSuite that referenced this issue Oct 24, 2024

Fix the coat model with onnx.if different shapes bug (#378)

9aa03f5

Fix the onnx.if different shapes issue nod-ai/SHARK-ModelDev#696. 4 dynamic models still left unfixed.

This was referenced Oct 30, 2024

Cast static/dynamic shape for onnx.If branches to match result type llvm/torch-mlir#3828

Merged

Fix onnx.If lowering with scalar condition tensor llvm/torch-mlir#3846

Merged

jinchen62 closed this as completed in llvm/torch-mlir#3846 Nov 1, 2024

jinchen62 closed this as completed in llvm/torch-mlir@032a636 Nov 1, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

onnx.If errors when the two branches yield with different shapes #696

onnx.If errors when the two branches yield with different shapes #696

renxida commented May 21, 2024 •

edited

Loading

renxida commented May 21, 2024

renxida commented May 30, 2024

renxida commented Jun 4, 2024 •

edited

Loading

renxida commented Oct 23, 2024

AmosLewis commented Oct 23, 2024 •

edited

Loading

AmosLewis commented Oct 23, 2024 •

edited

Loading

zjgarvey commented Oct 23, 2024 via email

AmosLewis commented Oct 24, 2024 •

edited

Loading

zjgarvey commented Oct 24, 2024

zjgarvey commented Oct 24, 2024

AmosLewis commented Oct 24, 2024 •

edited

Loading

AmosLewis commented Oct 25, 2024 •

edited

Loading

AmosLewis commented Oct 28, 2024 •

edited

Loading

AmosLewis commented Oct 30, 2024 •

edited

Loading

onnx.If errors when the two branches yield with different shapes #696

onnx.If errors when the two branches yield with different shapes #696

Comments

renxida commented May 21, 2024 • edited Loading

renxida commented May 21, 2024

renxida commented May 30, 2024

renxida commented Jun 4, 2024 • edited Loading

renxida commented Oct 23, 2024

AmosLewis commented Oct 23, 2024 • edited Loading

AmosLewis commented Oct 23, 2024 • edited Loading

zjgarvey commented Oct 23, 2024 via email

AmosLewis commented Oct 24, 2024 • edited Loading

zjgarvey commented Oct 24, 2024

zjgarvey commented Oct 24, 2024

AmosLewis commented Oct 24, 2024 • edited Loading

AmosLewis commented Oct 25, 2024 • edited Loading

AmosLewis commented Oct 28, 2024 • edited Loading

AmosLewis commented Oct 30, 2024 • edited Loading

renxida commented May 21, 2024 •

edited

Loading

renxida commented Jun 4, 2024 •

edited

Loading

AmosLewis commented Oct 23, 2024 •

edited

Loading

AmosLewis commented Oct 23, 2024 •

edited

Loading

AmosLewis commented Oct 24, 2024 •

edited

Loading

AmosLewis commented Oct 24, 2024 •

edited

Loading

AmosLewis commented Oct 25, 2024 •

edited

Loading

AmosLewis commented Oct 28, 2024 •

edited

Loading

AmosLewis commented Oct 30, 2024 •

edited

Loading