search

m-abr / FCPCodebase

FC Portugal Codebase
GNU General Public License v3.0
35 stars 7 forks source link

About Basic_Run usage #22

Closed dongmingre closed 5 months ago

m-abr commented 6 months ago

Please refer to Section 10 of the documentation (and more specifically Section 10.3 for information on how to integrate trained models).

m-abr commented 6 months ago

The GPU can only be used during policy updates, not during episode rollouts, which is the most time-consuming task. This limitation exists because the server (rcssserver3d/SimSpark) relies solely on the CPU. Therefore, using the GPU offers no practical advantage in this context.

However, you can enable it by changing every instance of device="cpu" to device="auto" or device="cuda" in the gym you are using. This change will instruct PyTorch to use the GPU for policy updates.