I have tried to write some dockerfiles for compilation with cuda 11.3. and the Ubuntu versions 18.04 and 20.04. I also adjusted the location of the Dockerfile to enable various Dockerfiles for different environments. Furthermore, I removed the necessity to clone the tvm repo as it is cloned in the container anyway.
During the development, I updated the transformers version since I could not find a working version for the given environment.
I also was not able to install python3.7 via ppa in the docker container so I used the default version of the OS.
Since I was not able to build and run the code as discribed originally, I couldn't test whether the tests fail because of the changes I made, or because they didn't worked before. Please let me know, if the upgrade of tvm to 0.8.0. causes the original code to fail.
Nevertheless, I hope this work helpes to support newer cuda and os versions.
I have tried to write some dockerfiles for compilation with cuda 11.3. and the Ubuntu versions 18.04 and 20.04. I also adjusted the location of the Dockerfile to enable various Dockerfiles for different environments. Furthermore, I removed the necessity to clone the tvm repo as it is cloned in the container anyway.
During the development, I updated the transformers version since I could not find a working version for the given environment.
I also was not able to install python3.7 via ppa in the docker container so I used the default version of the OS.
Since I was not able to build and run the code as discribed originally, I couldn't test whether the tests fail because of the changes I made, or because they didn't worked before. Please let me know, if the upgrade of tvm to 0.8.0. causes the original code to fail.
Nevertheless, I hope this work helpes to support newer cuda and os versions.