Skip to main content

sebulba (common setting)

Created on January 27|Last edited on January 27
4 settings:
# throughput
python sebulba_ppo_envpool.py --exp-name sebulba_ppo_envpool_thpt --actor-device-ids 0 --learner-device-ids 1 --update-epochs 8 --profile --test-actor-learner-throughput --track

# shared: actor on GPU0 and learner on GPU0
python sebulba_ppo_envpool.py --exp-name sebulba_ppo_envpool_1gpu --actor-device-ids 0 --learner-device-ids 0 --track

# separate: actor on GPU0 and learner on GPU1
python sebulba_ppo_envpool.py --exp-name sebulba_ppo_envpool_a0_l1 --actor-device-ids 0 --learner-device-ids 1 --track

# shared: actor on GPU0 and learner on GPU0,1
python sebulba_ppo_envpool.py --exp-name sebulba_ppo_envpool_a0_l01 --actor-device-ids 0 --learner-device-ids 0 1 --track

Select runs that logged charts/SPS
to visualize data in this line chart.
10M20M30M40M50Mglobal_step0100200300400500
50100150200250300Time (minutes)0100200300400500
sebulba_ppo_envpool
1
baseline
1
sebulba_ppo_envpool (1 GPU)
1
sebulba_ppo_envpool (pmap 2GPU)
1
sebulba_ppo_envpool (lambdalabs pmap 4GPU)
1
throughput (no a-l sync)
1
sebulba_ppo_envpool_1gpu
1
sebulba_ppo_envpool_a0_l1
1
sebulba_ppo_envpool_a0_l01
1




sebulba_ppo_envpool
1
baseline
1
sebulba_ppo_envpool (1 GPU)
1
sebulba_ppo_envpool (pmap 2GPU)
1
sebulba_ppo_envpool (lambdalabs pmap 4GPU)
1
sebulba_ppo_envpool (lambdalabs pmap 2GPU)
1
sebulba_ppo_envpool (lambdalabs pmap 8GPU)
1
sebulba_ppo_envpool (lambdalabs pmap 1GPU a, 4GPU l)
1