Is your feature request related to a problem? Please describe.
We maintain a downstream Triton image and are trying to understand when ONNX Runtime 1.26.0 is expected to be officially supported by Triton. The public compatibility matrix currently lists Triton 26.05 with ONNX Runtime 1.24.4, and server/main still defaults to ORT 1.24.4 in build.py.
Before carrying a downstream override, we'd like to understand whether ORT 1.26.0 is already planned for an upcoming Triton release train and what the preferred contribution path would be.
Describe the solution you'd like
Could the Triton team clarify the planned timeline for ONNX Runtime 1.26.0 support?
In particular:
- Is ORT 1.26.0 targeted for a specific Triton release, such as 26.07 or a later release?
- Is the expected change limited to updating
DEFAULT_TRITON_VERSION_MAP in server/build.py, or does it require coordinated changes in triton-inference-server/onnxruntime_backend as well?
- If external help is useful, would the preferred contribution be a
server PR, an onnxruntime_backend PR, coordinated PRs across both repos, or waiting for an existing maintainer branch to land?
Describe alternatives you've considered
We can override the ONNX Runtime version downstream by passing --ort-version 1.26.0 to build.py, but that leaves us outside the documented support matrix unless the full Triton/ONNX Runtime/backend combination is validated upstream.
We also considered opening a small PR to bump server/build.py, but from the public repo history it looks like this may already be part of release sequencing rather than a missing one-line update.
Additional context
Some public breadcrumbs I found while investigating:
Happy to help validate CPU-only builds and provide a downstream test plan if that would be useful.
Is your feature request related to a problem? Please describe.
We maintain a downstream Triton image and are trying to understand when ONNX Runtime 1.26.0 is expected to be officially supported by Triton. The public compatibility matrix currently lists Triton 26.05 with ONNX Runtime 1.24.4, and
server/mainstill defaults to ORT 1.24.4 inbuild.py.Before carrying a downstream override, we'd like to understand whether ORT 1.26.0 is already planned for an upcoming Triton release train and what the preferred contribution path would be.
Describe the solution you'd like
Could the Triton team clarify the planned timeline for ONNX Runtime 1.26.0 support?
In particular:
DEFAULT_TRITON_VERSION_MAPinserver/build.py, or does it require coordinated changes intriton-inference-server/onnxruntime_backendas well?serverPR, anonnxruntime_backendPR, coordinated PRs across both repos, or waiting for an existing maintainer branch to land?Describe alternatives you've considered
We can override the ONNX Runtime version downstream by passing
--ort-version 1.26.0tobuild.py, but that leaves us outside the documented support matrix unless the full Triton/ONNX Runtime/backend combination is validated upstream.We also considered opening a small PR to bump
server/build.py, but from the public repo history it looks like this may already be part of release sequencing rather than a missing one-line update.Additional context
Some public breadcrumbs I found while investigating:
triton-inference-server/server#8828prepared 26.06 and explicitly mentions reverting the ONNX Runtime version: ci(TRI-1406): prepare server for 26.06 (versions, TRT 11 QA compat, enroot env fix) #8828triton-inference-server/onnxruntime_backend#344prepared the ONNX Runtime backend for the 26.06 release train with OpenVINO 2026.2, TensorRT 11 support, and an ORT CCCL header-ordering cherry-pick: ci(TRI-1406): prepare ONNX Runtime backend for 26.06 (TRT 11, OV 2026.2, CCCL fix) onnxruntime_backend#344mchornyi/TRI-704/rubin-build, that appears to set ORT to 1.26.0 and OpenVINO to 2026.2.0.Happy to help validate CPU-only builds and provide a downstream test plan if that would be useful.