Ladislav Rampášek

Hi, I didn't think of it but it is a good idea! However I don't have the checkpoints at the moment.

Hi @yanzhangnlp, I briefly tired a basic fine-tuning and didn't get to a competitive performance in the initial experiment. For Graphormer, the authors used FLAG [1] and considerable hyper-parameter search...

Hi @yanzhangnlp, I'll look into it in the next few days and try to rerun `ogbg-code2` with the current codebase (after the update to PyG2.0.4). Are you running [the GPS...

Hi @yanzhangnlp, Here is W&B with 10 repeated runs (random seeds from 0 to 9) for the default `ogbg_code2` config: https://wandb.ai/gtransformers/rerun_code2?workspace=user-rampasek (search for `best/test_f1` panel) So far it seems to...

Hi @msadegh97, please see PR #11 for running inference using a pre-trained GPS model. With that all questions in this issue are closed.

Hi @amiltonwong, Seems it is a compatibility issue between different PyG, PyTorch and Cuda versions. For the current GraphGPS you need PyG=2.0.4 and PyTorch=1.10 (from my experience `performer-pytorch` package has...

Hi @Ranceeeee, unfortunately I have not run GraphGPS in a multiple GPU setting. The latest [PyG release 2.1.0](https://github.com/pyg-team/pytorch_geometric/releases/tag/2.1.0) has brought PyTorch Lightning support for GraphGym, upgrading to which in the...

Hi, You can also access it in GraphGPS that has been updated to PyG 2.0.4 compatibility: https://github.com/rampasek/GraphGPS I will refactor GraphGPS to PyG 2.1 in the future. However the dataset...