torch.compile
fails with jacfwd
when multiplying/dividing float and tensor
#125078
Labels
high priority
module: dynamo
module: functorch
Pertaining to torch.func or pytorch/functorch
module: pt2-dispatcher
PT2 dispatcher-related issues (e.g., aotdispatch, functionalization, faketensor, custom-op,
module: vmap
oncall: pt2
triaged
This issue has been looked at a team member, and triaged and prioritized into an appropriate module
Milestone
馃悰 Describe the bug
The following minimal example fails
with following last few lines in the error message (longer error log below).
Similar error if it's
return x / two
. Trying insteadalso fails with a different
NotImplementedError
(full error below).Error logs
For the first example where
two = 2.0
.For the second example where
two = torch.tensor([2.0], dtype=x.dtype, device=x.device)
.Minified repro
No response
Versions
cc @ezyang @gchanan @zou3519 @kadeng @msaroufim @bdhirsh @anijain2305 @chauhang @Chillee @samdow @kshitij12345 @janeyx99 @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78
The text was updated successfully, but these errors were encountered: