📉 Optimize GRPO memory usage by redefining per_device_batch_size
as generations per device
#7292
Job | Run time |
---|---|
9s | |
25m 4s | |
29m 6s | |
13m 18s | |
26m 55s | |
28m 9s | |
25m 47s | |
27m 21s | |
33m 48s | |
31m 54s | |
23m 50s | |
29m 32s | |
4h 54m 53s |