Conversation
Bubullzz
added
feature request
|
/ok to test 9082458 |
CI Test Summary1 failed · 30 passed · 0 skipped
|
| CUOPT_LOG_ERROR("num-gpus=%d exceeds the number of visible CUDA devices: %d.", | ||
| requested_gpus, | ||
| device_count); | ||
| return -1; |
There was a problem hiding this comment.
I think it would be better to just issue the warning and continue the solve with the current available GPUs
There was a problem hiding this comment.
I don't understand why we wouldn't rather want to fail loudly ? if the user specifically asks to solve with --num-gpus 4 it is very likely they explicitely want a solve with 4 gpus. If we don't see 4 GPUs then the user will want to check their setup rather than a solve on 1 or 2 GPUs. If they don't care and want to use the max number of GPUs they can just set -1.
Did I miss something ?
There was a problem hiding this comment.
It's a design choice. Usually we prefer to complete a solve rather if we can recover, rather than failing.
Note this only happens through the command-line interface. What are you doing with the other APIs? Are you failing hard there as well?
2 participants
No description provided.