etaoxing
etaoxing
Seems like the issue could be `embree` not using the latest version v3.13 which has M1 support? ```bash [2022-05-20 14:08:36] ~/Documents/repos/point-cloud-utils $ cd external/embree [2022-05-20 14:08:44] ~/Documents/repos/point-cloud-utils/external/embree $ git status...
> How does the try/except block resolves this issue? I added: ```python n_tries = 0 while n_tries < 3: try: idx = fps(pos, batch, ratio=self.ratio) break except Exception as e:...
Yeah, it's a strange issue. Btw `pytorch3d.ops.sample_farthest_points(random_start_point=False)` is their [default setting](https://github.com/facebookresearch/pytorch3d/blob/97894fb37bdb9ecda25322a8e66999f5a1961b36/pytorch3d/ops/sample_farthest_points.py#L20). I'll try running with `fps(random_start=False)` when I get the chance, and see if the error occurs.
Update: I did a few runs with `fps(random_start=False)` and did not encounter the exception. Interestingly, a colleague started a job on the same GPU which I was using, and I...
Given the `graphgym/` rewrite and PyTorch Lightning integration (#4503) then, it might be good to wait until that is merged before proceeding on larger-scale benchmarking. Btw, the pyg wandb team...
> I am not really sure why one cannot see the wandb team. It is set to "Public" :( Strange, maybe only user profiles are allowed to be publicly viewed?...
@chrischoy should this fix be merged?
I'm running on a machine with a 3090 and 64GB RAM, so I lowered to `replay_cfg.capacity=5000` and `replay_cfg.cache_size=5000`. `Pinch-v0/trajectory.none.pd_ee_delta_pose_pointcloud.h5` is 40GB. ```bash python maniskill2_learn/apis/run_rl.py configs/brl/bc/pointnet_soft_body.py \ --work-dir workdir/ --gpu-ids 0...
Yes, the demos are on root ssd. Seems to start training, but grad_norm becomes 0 pretty quickly. True for `env_cfg.control_mode=pd_ee_target_delta_pose` and `env_cfg.control_mode=pd_ee_delta_pose` ``` python maniskill2_learn/apis/run_rl.py configs/brl/bc/pointnet_soft_body.py --work-dir workdir / --gpu-ids...
Finetuning / training from scratch pipeline is in the works, will probably be a while before it gets pushed though.