I am trying to run tuned mbmpo Cartpole example on 32-Core machine from here:
https://github.com/ray-project/ray/blob/master/rllib/tuned_examples/mbmpo/cartpole-mbmpo.yaml
num_workers: 1 and num_workers: 2 are only working options. Any other value of num_workers gives below error:
(pid=2410138) File “/home/dl-user/miniconda3/envs/ki4lsa/lib/python3.6/site-packages/ray/rllib/agents/maml/maml_torch_policy.py”, line 240, in split_placeholders
(pid=2410138) placeholder, torch.sum(split, dim=1).tolist(), dim=0)
(pid=2410138) File “/home/dl-user/miniconda3/envs/ki4lsa/lib/python3.6/site-packages/torch/functional.py”, line 156, in split
(pid=2410138) return tensor.split(split_size_or_sections, dim)
(pid=2410138) File “/home/dl-user/miniconda3/envs/ki4lsa/lib/python3.6/site-packages/torch/_tensor.py”, line 515, in split
(pid=2410138) return super(Tensor, self).split_with_sizes(split_size, dim)
(pid=2410138) RuntimeError: split_with_sizes expects split_sizes to sum exactly to 32 (input tensor’s size at dimension 0), but got split_sizes=[30]
I tried 5, 10, 30, 31, 32 and few others, none of them worked. Please provide input if you have encountered this issue or have any suggestions.