jp72: upgrade PyTorch to 2.13#2661
Closed
alexnorell wants to merge 4 commits into
Closed
Conversation
… RTSP HW-decode + latest fixes)
torch 2.13 requires setuptools>=77.0.3, but the 7.2 server image pinned setuptools<=75.5.0, making the dependency set unsatisfiable (uv: 'No solution found'). The upper bound had no documented reason and no requirements file demands it (only setuptools>=70.0.0). Raise the floor to satisfy torch 2.13. 6.2 keeps torch 2.6 and its existing pin.
PyTorch 2.13 pins triton==3.7.1 exactly, but the wheels image fetched triton 3.6.0, so the server dependency install failed with uv 'No solution found: torch 2.13 depends on triton==3.7.1, but there is no version of triton==3.7.1'. Fetch the cp312 aarch64 triton 3.7.1 wheel (sha256 verified). Confirmed via a local depot bake: the uv resolution now succeeds (253 packages resolved, triton==3.7.1 installed).
Contributor
Author
|
Closing as superseded by #2651. That PR carries the PyTorch 2.13 / TorchVision 0.28 upgrade, the same pinned FlashAttention revision, the setuptools compatibility fix, and the Triton 3.7.1 wheel/checksum update from this branch. It also contains the follow-on JP7.2 runtime, SM87 + SM110 validation, ORT 1.26 work, and the successful device-tested image build. No unique patch remains here. |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Upgrades the JetPack 7.2 wheels and server image to PyTorch 2.13.0 with its supported TorchVision 0.28.0 pair.
Updates the external FlashAttention source pin to an exact upstream commit that retains CUDA 13 Thor (
sm_110) support. The existing Orin (sm_87) patch applies cleanly to that revision.Pins setuptools to PyTorch 2.13 source-build requirements (
>=77,<82).Why
PyTorch 2.13 adds Thor/Blackwell-relevant Inductor work and lower-overhead CUDA launch machinery, which are directly relevant to the tensor-native inference and tracking path.
Validation
docker buildx bake -f docker/docker-bake.jetson.7.2.0.hcl --print jetson-wheels-jp72 jetson-server-jp72sm_87patch applies cleanly to the pinned source SHA.