optimization when lack of CPU

[LukeLIN-web]

top - 15:41:43 up 2 days,  6:18,  8 users,  load average: 148.76, 159.89, 162.12
Tasks: 1130 total,   2 running, 1128 sleeping,   0 stopped,   0 zombie
%Cpu(s): 98.3 us,  1.6 sy,  0.0 ni,  0.0 id,  0.0 wa,  0.0 hi,  0.0 si,  0.0 st
MiB Mem : 257602.6 total,  95593.6 free,  27249.9 used, 134759.2 buff/cache
MiB Swap:   4096.0 total,   4096.0 free,      0.0 used. 228307.0 avail Mem 

    PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ COMMAND                                                                                                                          
3439911      20   0   39.7g   3.9g   1.0g R  1302   1.6  23:52.89 pt_main_thread                                                                                                                   
3412216     20   0   40.0g   4.0g   1.0g S  1072   1.6 131:49.12 pt_main_thread                                                                                                                   
3372667     20   0   40.6g   4.7g   1.0g S 162.7   1.9 290:17.24 pt_main_thread                                                                                                                   
3444674       20   0 9892.2m 134088  37632 S 121.9   0.1   0:03.73 ffmpeg             

I launched 4 Google robot evaluations using run_openvla.sh to evaluate 4 checkpoints simultaneously.
However, my 96-core machine is running out of CPU resources.
Do you have any suggestions for improving performance? Thank you!

[xuanlinli17]

Looks like your model is running on cpu instead of gpu. If it's using cpu, it will consume many cpu resources (and the env itself is slow). Pay attention to the tensorflow / jax info printed out in command line.