Rhett Ying
Rhett Ying
Please share more details of your config for `dgl.nn.GraphConv`. how did you init it, what are the arguments when calling `forward`?
what if disable `dropout` ?
The non-deterministic result is caused by `weight=True` when instantiate `GraphConv`. This arguments is `True` in default which will init by below code: https://github.com/dmlc/dgl/blob/0f0e7c7fad6ec1399a67881ce9a7e60d14cfe1a0/python/dgl/nn/pytorch/conv/graphconv.py#L283-L284 Then the `weight` will be re-init by...
Why do we choose the nightly build for test instead of stable one? If we need the speed-up introduced in `22.08`, why not pick up `nvcr.io/nvidia/pytorch:22.08-py3` which will be released...
so for now, you need `rapidsai/rapidsai-core-nightly:22.08-cuda11.5-runtime-ubuntu20.04-py3.9`. and maybe in near future, you will need a newer nightly build like `nightly:22.09` if you add more new features? or is there any...
Nightly build is not quite stable? Why not pick up the incoming stable one: `nvcr.io/nvidia/pytorch:22.08-py3`? is this the stable counterpart of the nightly build you're using?
`population`: 2500K*2500K = 6250B `sample`: 2B
Commit ID: 9600d6e9b8850584e0fe05a7364594bb85021fe7 Build ID: 11 Status: ✅ CI test succeeded Report path: [link](https://dgl-ci-result.s3.us-west-2.amazonaws.com/dgl/PR-4269/11/11/logs/report.html) Full logs path: [link](https://dgl-ci-result.s3.us-west-2.amazonaws.com/dgl/PR-4269/11/11/logs/cireport.log)
Commit ID: 0b9e9c2d6d20dd6f80bb617faa9d68baa39f8b16 Build ID: 9 Status: ✅ CI test succeeded Report path: [link](https://dgl-ci-result.s3.us-west-2.amazonaws.com/dgl/PR-4389/9/9/logs/report.html) Full logs path: [link](https://dgl-ci-result.s3.us-west-2.amazonaws.com/dgl/PR-4389/9/9/logs/cireport.log)