[C++ API] L1Loss module #25902
[C++ API] L1Loss module #25902
Conversation
ShahriarRezghi
requested review from
ebetica,
goldsborough and
yf225
as code owners
pytorchbot
added
caffe2
module: build
|
@ShahriarSS Not keeping |
| s.backward(); | ||
|
|
||
| ASSERT_EQ(output.sizes(), torch::IntArrayRef()); | ||
| ASSERT_EQ(input.sizes(), input.grad().sizes()); |
There was a problem hiding this comment.
Is there a way we can also check the correctness of output value after loss->forward?
There was a problem hiding this comment.
I based this on python's lL1Loss test and I think that it didn't check. What is the expected value here?
There was a problem hiding this comment.
Got it, I think we can rely on the parity test to make sure the output value is in sync with the Python API.
|
I'll also work on enabling parity test for |
|
I will rename this PR to |
|
@pytorchbot rebase this please |
yf225
force-pushed
the
C++-loss
branch
9 times, most recently
from
0db0f10 to
25c1855
Compare
|
@pytorchbot rebase this please |
yf225
force-pushed
the
C++-loss
branch
9 times, most recently
from
fc699ab to
fa19561
Compare
yf225
left a comment
yf225
left a comment
There was a problem hiding this comment.
Thanks a lot @ShahriarSS !
facebook-github-bot
left a comment
facebook-github-bot
left a comment
There was a problem hiding this comment.
@yf225 is landing this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
6 participants
@yf225 This is L1Loss module. I don't think that
_Lossand_WeightedLossas base Python classes do anything. First one sets reduction type and also takes inreduceparameter which is deprecated. The second one only registersweightparameter. I don't think that we should keep this structure. What do you think?