[Fix][Relax][Torch] Materialize runtime scalar shape values - #20138
Open
akaashrp wants to merge 5 commits into
Open
[Fix][Relax][Torch] Materialize runtime scalar shape values#20138akaashrp wants to merge 5 commits into
akaashrp wants to merge 5 commits into
Conversation
Contributor
Author
|
@tvm-bot rerun |
Contributor
|
Failed to re-run CI in https://github.com/apache/tvm/actions/runs/32556288920 Detailswith response |
tlopex
requested changes
Aug 26, 2026
tlopex
left a comment
Member
There was a problem hiding this comment.
The overall direction makes sense, but several cases still need to be fixed. Symbolic eq is currently lowered to a constant True, which causes dynamic torch.cond to select the wrong branch. The new scalar materialization helper is also not used by fill/fill_, so importing them without decomposition still fails. In addition, full incorrectly defaults runtime integer fill values to float32, and full_like ignores an explicitly specified dtype. Please fix these cases and add corresponding regression tests.
Member
|
and also resolve the conflict |
akaashrp
force-pushed
the
upstream/torch-runtime-scalar-shapes
branch
from
August 27, 2026 18:46
bf23939 to
6e73650
比较
Member
|
Ahh I have a look, now explicit |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
注册 for free
to join this conversation on GitHub.
Already have an account?
登录 to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Materialize runtime integer scalars from
torch.exportas Relax symbolic values and convert them back to rank-zero tensors for fill operations. Preservefull/full_likedtype semantics, support directfill/fill_without decomposition, and lower symbolic shape comparisons as real predicates. Runtime scalar extraction now also accepts statically one-element integer tensors at higher ranks and signed/unsigned integer dtypes, preserving negative, unsigned, and 64-bit values. Python scalar fills are materialized in the destination dtype acrossfull,full_like,fill,fill_,masked_fill, andmasked_fill_, avoiding 32-bit narrowing or precision loss. Symbolic boolean fills, negation, floor division, and modulo are also supported.