Repository navigation
Conversation
Discover installed Warp adapters for Triton ONNX Runtime and NativeRT/PT2 models. Keep ONNX CUDA replay tensors on the GPU and support native Torch registration, stream ordering, and cleanup when models unload. Add explicit SDK selection, generic CMake argument forwarding, and install rules without embedding the absolute Warp library path in dependencies. Validate SDKs before replacing an existing cached runtime build. Validation: seven build-helper tests, two native CUDA replay tests, and two Triton integration tests covering twelve session placements and load errors. Verified SDK overrides, relocated installs, and formatting. Signed-off-by: Marc Alwan <malwan@nvidia.com>
Signed-off-by: Marc Alwan <malwan@nvidia.com>
Default the new engine option in the Warp integration test. Signed-off-by: Marc Alwan <malwan@nvidia.com>
This branch has not been deployed
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This adds Warp support to compiled LEAPP sessions through native adapters for Triton’s ONNX and PT2 backends. It supports CPU and CUDA tensor placement, CUDA stream ordering, noncontiguous Torch inputs, and graph cleanup when models unload. Build improvements add explicit SDK selection, installable libraries, and dependency validation before replacing cached builds. Validation covers CUDA replay, all 12 placement combinations, session lifecycle, and adapter load failures.