nod-ai / SHARK-ModelDev

Unified compiler/runtime for interfacing with PyTorch Dynamo.
Apache License 2.0
95 stars 48 forks source link

[Tracking] E2E Tests (torch->onnx->torch->linalg) #549

Open vivekkhandelwal1 opened 7 months ago

vivekkhandelwal1 commented 7 months ago

This issue tracks the E2E op tests for the OnnxToLinalg lowering.

Failing Tests (count: 547): as on 08/04/24

Higher Priority:

Failure - incorrect numerics

Failure - incorrect shape

Failure - onnx_lowering: onnx.RandomNormal

Failure - onnx_lowering: onnx.RandomNormalLike

Failure - onnx_lowering: onnx.RandomUniform

Failure - onnx_lowering: onnx.RandomUniformLike

Failure - onnx_lowering: onnx.ReduceProd

Failure - onnx_lowering: onnx.Pad

Failure - onnx_lowering: onnx.ScatterND

Failure - onnx_lowering: onnx.ScatterElements

Failure - onnx_lowering: onnx.Squeeze

Failure - onnx_lowering: onnx.MaxPool

Failure - onnx_lowering: onnx.ReduceL1

Failure - onnx_lowering: onnx.Resize

Failure - onnx_lowering: onnx.AveragePool

Failure - onnx_lowering: onnx.SoftmaxCrossEntropyLoss

Failure - onnx_lowering: onnx.Cast

Failure - onnx_lowering: onnx.ReduceL2

Failure - onnx_lowering: onnx.ReduceSum

Failure - cast error

Failure - incorrect dtype

Failure - onnx_lowering: onnx.Clip

Failure - onnx_lowering: onnx.OneHot

Failure - torch.aten.view lower

Failure - unknown

Failure - onnx_import

Failure - "RuntimeError: linalg.cross: inputs dimension 1 must have length 3. Got 1 and 1"

Crashing tests

Lower Priority:

Failure - onnx_export

Fixed:

vivekkhandelwal1 commented 3 months ago

This tracker is no more maintained. @zjgarvey I think you've been reporting the status of these tests. What should we do with this tracker?