Skip to main content
costa-huang
Projects
cleanRL
Reports
MAX learner load: direct device transfer IMPALA
Log in
Sign up
Share
Comment
Star
Share
Comment
Star
MAX learner load: direct device transfer IMPALA
Costa
Created on April 22
|
Last edited on April 23
Comment
jax 0.3.25
charts/SPS_update
charts/SPS_update
500k
1M
1.5M
2M
2.5M
global_step
5000
10000
15000
charts/SPS
charts/SPS
500k
1M
1.5M
2M
2.5M
global_step
2000
4000
6000
8000
10000
12000
14000
stats/rollout_time
stats/rollout_time
500k
1M
1.5M
2M
2.5M
global_step
0.5
1
1.5
charts/avg_episodic_return
charts/avg_episodic_return
500k
1M
1.5M
2M
2.5M
global_step
0
5
10
15
20
25
30
stats/inference_time
stats/inference_time
500k
1M
1.5M
2M
2.5M
global_step
0.2
0.4
0.6
0.8
1
1.2
stats/d2h_time
stats/d2h_time
500k
1M
1.5M
2M
2.5M
global_step
0.05
0.1
0.15
0.2
0.25
l0
1
l1
1
l1,2
1
l1,2,3,4,5
1
l0, concat
1
l1, concat
1
l1,2, concat
1
l1,2,3,4,5 concat
1
l1,2,3,4,5 block
1
l1,2 block
1
l1 block
1
l1,2,3,4,5 concat2
1
trace
l1
1
l1 2
1
l1 2 3 4 5
1
Add a comment