Zhihao Jia

Thanks for pointing out the issue. The bug should have been fixed in 6b884. The logic for new_weight is that TASO will always allocate a device memory for the new...

I have replaced the assertion in noop.cc to a warning message in commit 49e804e8. Presumably, when TASO encounters an uninitialized weight tensor, it is likely due to a bug in...

@OuHangKresnik The support for PRelu is added in comment 3723d31.

@OuHangKresnik I pushed a temporary fix for this. The output shape of Pad depends on the second input tensor, so we will need the shape information from the ONNX runtime...

It seems there are potential misconfiguration when you did `cmake`. And people have experienced similar issues with onnx https://github.com/onnx/onnx-tensorrt/issues/150. Can you try to set `USE_CUDA` to your CUDA path in...

The failure should have been fixed by commit c76ee70

@puddingfjz Thanks for your interests in TASO. TASO currently uses around 700 default substitutions, and you can easily customize the substitutions you want to explore by running https://github.com/jiazhihao/TASO/blob/master/src/generator/generator.cc. Line 1731-1783...

@qunluo I was able to reproduce the bug. It is because the ONNX file you used didn't specify a batch size for the input tensor, and TASO assumed a batch...

@sonfiree are you using the pre-built docker image? Make sure you have installed nvidia-docker, and launch the TASO docker image via `run_docker.sh tasoml/cuda100`.

You don't need to build the docker image. Once you have cloned the TASO github repo, you can directly launch our prebuilt docker image via `/path/to/taso/docker/run_docker.sh tasoml/cuda100`.