Omer Ullman Argov comments

Repositories
Issues
Comments

Results 6 comments of


                                            Omer Ullman Argov

Conversion from onnx hurts accuracy when model is using fp16

This is a continuation of https://github.com/NVIDIA/TensorRT/issues/2899 , in which I couldn't figure out how to stop an overflowing weight from being converted. I am getting some warnings but can't manage...

Conversion from onnx hurts accuracy when model is using fp16

@zerollzeng Anyway I can use 8.6 without updating to CUDA 12?

Conversion from onnx hurts accuracy when model is using fp16

Ok, I ended up updating to cuda12 and trt 8.6.1 and I'm still getting all-zero output for some reason

fix/ replace sanity test for nemotron h with a correctness test

/bot run

fix/ replace sanity test for nemotron h with a correctness test

/bot run

fix/ replace sanity test for nemotron h with a correctness test

/bot run