How severe does this issue affect your experience of using Ray?
- High: It blocks me to complete my task.
I am working with External Simulators and Policy Client/Servers. In my setting, the episode management is actually done on the Client side. Is there a way to save a checkpoint of the most recent model (i.e. after client.update_policy_weights() ) from the client side?
This is required because, with variable episode length and variable amount of experience (as in a turn-based multi-agent setting), the checkpoints on the server side might be outdated once the run finishes.