[torch] Segmented Polynomial Product #97
Conversation
cuequivariance_torch/cuequivariance_torch/primitives/segmented_polynomial_product.py
Show resolved
Hide resolved
cuequivariance_torch/cuequivariance_torch/primitives/segmented_polynomial_product.py
Outdated
Show resolved
Hide resolved
cuequivariance_torch/tests/primitives/segmented_polynomial_test.py
Outdated
Show resolved
Hide resolved
|
i don't get why those pipelines are failing, any input @mariogeiger ? |
Oh crap sorry about that. On the CI I've installed a version of cuequivariance-ops that is compatible with the "advanced batching" but not with main. Let me switch back |
|
cuequivariance-torch (ubuntu-latest, 3.10) is failing because you don't have an escape mechanism to run the tests on cpu. The simplest fist is to skip the tests when there are no GPU available. |
|
I changed the cuequivariance-ops packages in the CI, now cuequivariance-torch (self-hosted, 3.12) is failing with a different error |
|
@mariogeiger looks like the remaining failure is not even in code i changed. do you know what's going on? |
cuequivariance_torch/.coverage
Outdated
There was a problem hiding this comment.
Please remove this file
| if operand_extent is None: | ||
| (operand_extent,) = list(o.get_dims(0)) | ||
| else: | ||
| torch._assert( | ||
| operand_extent == list(o.get_dims(0))[0], | ||
| "all operands must have the same extent", | ||
| ) |
There was a problem hiding this comment.
| if operand_extent is None: | |
| (operand_extent,) = list(o.get_dims(0)) | |
| else: | |
| torch._assert( | |
| operand_extent == list(o.get_dims(0))[0], | |
| "all operands must have the same extent", | |
| ) | |
| if operand_extent is None: | |
| (operand_extent,) = o.segment_shape | |
| else: | |
| torch._assert( | |
| (operand_extent,) == o.segment_shape, | |
| "all operands must have the same extent", | |
| ) |
Sorry my bad, this is nicer. (segment_shape is a property that exist only if all_same_segment_shape() is true.)
| ) | ||
|
|
||
|
|
||
| class SegmentedPolynomial(nn.Module): |
There was a problem hiding this comment.
Can you add a docstring?
2 participants
No description provided.