[rllib] PPO doesn't work with fractional num gpus #3396
Conversation
|
Test FAILed. |
|
Test FAILed. |
|
would love to have this merged just so I can build with the wheel instead of from source |
|
@jhpenger just a note that if you want to build wheels on branches/commits that we don't provide, that can be done by following the instructions in https://github.com/ray-project/ray/blob/master/python/README-building-wheels.md. |
| @@ -44,7 +45,9 @@ def _init(self, | |||
| if not num_gpus: | |||
| self.devices = ["/cpu:0"] | |||
| else: | |||
| self.devices = ["/gpu:{}".format(i) for i in range(num_gpus)] | |||
| self.devices = [ | |||
| "/gpu:{}".format(i) for i in range(int(math.ceil(num_gpus))) | |||
There was a problem hiding this comment.
isn't numpy already imported?
There was a problem hiding this comment.
Better to avoid numpy if you can use standard libraries instead.
|
Test FAILed. |
|
jenkins retest this please |
|
Test FAILed. |
What do these changes do?
Round up to next int when allocating ids w/fractional gpus.
Related issue number
Closes #3394