Zero Zeng

I don't get what the problem is but https://github.com/onnx/onnx-tensorrt/blob/main/docs/operators.md might be helpful

> This version of TensorRT only supports input K as an initializer. Try applying constant folding on the model using Polygraphy: https://github.com/NVIDIA/TensorRT/tree/master/tools/Polygraphy/examples/cli/surgeon/02_folding_constants This is why it fails, k must be...

and we only support sorted output.

> Which device is the Reformatting layer running on GPU > when TensorRT inserts the reformatting layer between DLA and GPU? For example, if the GPU output is FP16 linear...

@YouSenRong Setting those layer back to FP32 just solve the subnormal value issue, but FP16 indeed has less accuracy than FP32 due to less mantissa bit. if you set all...

@ttyio ^ ^

@rajeevsrao ^ ^

Filed internal bug 4411864 to track

Could you please try latest 9.2/9.3 release? Thanks!

IIRC most of the transformer support are added since TRT 8.6. And we won't back port feature back to old release.