search

google-research / seed_rl

SEED RL: Scalable and Efficient Deep-RL with Accelerated Central Inference. Implements IMPALA and R2D2 algorithms in TF2 with SEED's architecture.
Apache License 2.0
798 stars 146 forks source link

how to detach replay buffer from GPU memory during training and inference #36

Closed turmeric-blend closed 4 years ago

turmeric-blend commented 4 years ago

hi, is there a simple way to detach replay buffer from gpu for both training and inference? in particular for R2D2. Thanks.

lespeholt commented 4 years ago

Not sure what you mean. The replay should be on CPU due to this line: https://github.com/google-research/seed_rl/blob/master/common/utils.py#L57