Matthias Gehre

Please specify the behavior of min > max in Clip

12

# Ask a Question ### Question [onnx.Clip](https://onnx.ai/onnx/operators/onnx__Clip.html) leaves it open what happens when values for min and max are specified such that min > max. Should this - be an...

We used to move the SmallVector into an ArrayRef and then the SmallVector left the scope. Found by asan.

Example IR ``` func.func @torch.aten.pad.reflect(%input: !torch.tensor, %pads: !torch.vtensor) -> !torch.tensor { %int0 = torch.constant.int 0 %float0.000000e00 = torch.constant.float 0.000000e+00 %1 = torch.aten.select.int %pads, %int0, %int0 : !torch.vtensor, !torch.int, !torch.int ->...

convert-torch-onnx-to-torch generates invalid IR for onnx.Resize where scaling is in the first two dimensions

1

The code in that pass seems to silently assume the the first two dimensions are not scaled, but ONNX has no such restriction. With input ``` func.func @test_resize_middle(%arg0: !torch.vtensor) ->...

onnx.reference: Cast to float8e4m3fnuz treats +/-inf wrong

2

# Bug Report ### Describe the bug Casting +/-inf from float to float8e4m3fnuz / float8e5m2fnuz with `onnx.reference` yields +/- FLT_MAX where it should be NaN according to the table in...