Tom Tseng
Tom Tseng
Yup, it still does.
Interesting—conda works. For reference: The CircleCI `config.yml` script I used to test this: ``` version: 2.1 jobs: import-ray: macos: xcode: 13.4.1 steps: - checkout - run: name: Install conda command:...
> @tomtseng what did you find confusing when you started working on `imitation`? I agree that the code in `scripts/` was difficult to follow when I was debugging a test...
@Rocamonde if you click "View file" you can get a more sane view of the notebook  It seems like [there is a way...
No problem, thanks! (This isn't blocking me on anything, I was just looking at this for part of a class project. We decided to pivot elsewhere. So there's no rush...
(cc @hyln9 since they wrote the TensorRT backend)
FWIW I can reproduce on TensorRT 8.4.2 as well — I tried changing the first line of the [Dockerfile](https://github.com/tomtseng/KataGo/commit/fee37de93535d488a17693a98451d84e97ac0717#diff-dd2c0eb6ea5cfc6c4bd4eac30934e2d5746747af48fef6da689e85b752f39557) to be `FROM nvcr.io/nvidia/tensorrt:22.08-py3` (instead of `FROM nvcr.io/nvidia/tensorrt:22.09-py3`) and the error...
Interesting, thanks for that extra info. Do you think we should bother trying to report this in forums.developer.nvidia.com? I guess the problem with trying to write a report is that...
OK, I made a post on the NVIDIA developer forums to see if anyone has seen something like this before: https://forums.developer.nvidia.com/t/tensorrt-produces-unexpected-nan-values-during-inference/232130
Yeah I worked around it by using deduplicated indices in the config file. I didn't think about whether it's easy to write a code fix but figured I would report...