search

NM512 / dreamerv3-torch

Implementation of Dreamer v3 in pytorch.
MIT License
422 stars 96 forks source link

I ran DreamerV3 with bigger input resolution, but it will be "Killed" after around 325k steps #59

Closed Taskii-Lei closed 3 months ago

Taskii-Lei commented 4 months ago

I want to run dreamer-v3 on task "hammer" in meta-world. I editted the config.yaml to :

task: 'metaworld_hammer'
size: [256, 256]
batch_size: 28
batch_length: 32

Other parameters remain the same as before. But I found when it trains around 325k steps, it gonna stop with error "Killed", and no other words.

Stderr 2024-06-21, 18:57:00
Logger.warn(f"Box bound precision lowered by casting to {self.dtype}")
Stderr 2024-06-21, 19:02:09
Killed

I tried several times but all reduced to the same results. I'm really confused.

dirkmcpherson commented 4 months ago

Have you checked your memory usage?