Comment by sally_glance

Not an expert on this but I think the RL runs need to work sequentially? I wonder what the opportunities for distributed execution would be... Maybe parallelizing the benchmark task or inference