Repository navigation
Add NASNet-a model - #1688
Open
easyeasydev wants to merge 4 commits into
Open
Add NASNet-a model#1688easyeasydev wants to merge 4 commits into
easyeasydev wants to merge 4 commits into
Conversation
lockshaw
pushed a commit
to lockshaw/FlexFlow
that referenced
this pull request
Oct 13, 2026
…n. (flexflow#1688) GPU kernel and test are updated and should be in sync with task-spec. CPU kernel is still not implemented. Properly implements dim argument to match PyTorch semantics, which as far as I can tell wasn't working previously.
lockshaw
pushed a commit
to lockshaw/FlexFlow
that referenced
this pull request
Oct 14, 2026
…n. (flexflow#1688) GPU kernel and test are updated and should be in sync with task-spec. CPU kernel is still not implemented. Properly implements dim argument to match PyTorch semantics, which as far as I can tell wasn't working previously.
lockshaw
added a commit
to lockshaw/FlexFlow
that referenced
this pull request
Oct 14, 2026
…lel views Pre machine space arg solution Get most of data parallelism pass functioning minus some debugging Prototype new copy insertion Fix bugs in new copy insertion, overhaul dot visualization Checkpoint before adding data dependence modelling for parallel operators Migrate utils sufficiently to get utils-tests passing again More op dependency modelling with hemiunique relation Format Cleanup docs and start on data parallelism Format Format Get utils building again Modernize a bunch of the old operators Add more restrictive MappedOperatorTaskGroup types along with tests verifying the relationships they should allow Pass data parallelism test Refactoring and working on getting all ops ready for parallelism More parallel mappings for operators Add parallel dependency correctness test for batch matmul Pass shard signature correctness tests for BatchMatmul and Transpose for nontrivial parallelisms Modernize softmax lib/kernels interface and cuda kernel implementation. (flexflow#1688) GPU kernel and test are updated and should be in sync with task-spec. CPU kernel is still not implemented. Properly implements dim argument to match PyTorch semantics, which as far as I can tell wasn't working previously. Add data parallelism correctness tests for a bunch more yolov10 operators Build fixes Fix all non-linker build errors Even more test fixes Fixes for compiler tests
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
Summary
Add NASNet-A Large to models.
The implementation includes:
dtgennasnet_a_largesupport inexport-model-archPCG and task-spec changes
NASNet requires asymmetric padding and cropping, but the current computation graph does not expose dedicated padding or slicing operators. The model therefore composes these operations using existing
split,concat, and scalar operators.This exposed two existing split infrastructure issues:
ComputationGraphBuilder::splitvalidated the requested output count against variadic input slots instead of output slots.OUTPUTtensor even though split producesOUTPUT_00,OUTPUT_01, etc. This resulted in incorrect output pointers and block sizes for multi-output execution.This PR corrects the PCG validation and updates the split forward/backward task implementations to collect every variadic output independently. It also adds coverage for rejecting unsupported output counts.
This change is