post(TRI-1431): advance main with r26.06 upstream-tracking cherry-picks

Dockerfile.sdk

@@ -29,7 +29,7 @@
 #
 
 # Base image on the minimum Triton container
-ARG BASE_IMAGE=nvcr.io/nvidia/tritonserver:26.05-py3-min
+ARG BASE_IMAGE=nvcr.io/nvidia/tritonserver:26.06-py3-min
 
 ARG TRITON_CLIENT_REPO_SUBDIR=clientrepo
 ARG TRITON_REPO_ORGANIZATION=http://github.com/triton-inference-server

README.md

@@ -29,8 +29,8 @@
 
 >[!WARNING]
 >You are currently on the `main` branch which tracks under-development progress
->towards the next release. The current release is version [2.69.0](https://github.com/triton-inference-server/server/releases/latest)
->and corresponds to the 26.05 container release on NVIDIA GPU Cloud (NGC).
+>towards the next release. The current release is version [2.70.0](https://github.com/triton-inference-server/server/releases/latest)
+>and corresponds to the 26.06 container release on NVIDIA GPU Cloud (NGC).
 
 # Triton Inference Server
 
@@ -90,16 +90,16 @@ Inference Server with the
 
 ```bash
 # Step 1: Create the example model repository
-git clone -b r26.05 https://github.com/triton-inference-server/server.git
+git clone -b r26.06 https://github.com/triton-inference-server/server.git
 cd server/docs/examples
 ./fetch_models.sh
 
 # Step 2: Launch triton from the NGC Triton container
-docker run --gpus=1 --rm --net=host -v ${PWD}/model_repository:/models nvcr.io/nvidia/tritonserver:26.05-py3 tritonserver --model-repository=/models --model-control-mode explicit --load-model densenet_onnx
+docker run --gpus=1 --rm --net=host -v ${PWD}/model_repository:/models nvcr.io/nvidia/tritonserver:26.06-py3 tritonserver --model-repository=/models --model-control-mode explicit --load-model densenet_onnx
 
 # Step 3: Sending an Inference Request
 # In a separate console, launch the image_client example from the NGC Triton SDK container
-docker run -it --rm --net=host nvcr.io/nvidia/tritonserver:26.05-py3-sdk /workspace/install/bin/image_client -m densenet_onnx -c 3 -s INCEPTION /workspace/images/mug.jpg
+docker run -it --rm --net=host nvcr.io/nvidia/tritonserver:26.06-py3-sdk /workspace/install/bin/image_client -m densenet_onnx -c 3 -s INCEPTION /workspace/images/mug.jpg
 
 # Inference should return the following
 Image '/workspace/images/mug.jpg':

build.py

@@ -32,6 +32,7 @@
 import os.path
 import pathlib
 import platform
+import re
 import stat
 import subprocess
 import sys
@@ -75,8 +76,8 @@
     "triton_container_version": "26.07dev",
     "upstream_container_version": "26.05",
     "ort_version": "1.24.4",
-    "ort_openvino_version": "2026.1.0",
-    "standalone_openvino_version": "2026.1.0",
+    "ort_openvino_version": "2026.2.0",
+    "standalone_openvino_version": "2026.2.0",
     "dcgm_version": "4.5.3-1",
     "rhel_py_version": "3.12.3",
 }
@@ -921,6 +922,7 @@ def create_dockerfile_buildbase_rhel(ddir, dockerfile_name, argmap):
 
 RUN pip3 install --upgrade pip \\
       && pip3 install --upgrade \\
+          auditwheel \\
           build \\
           wheel \\
           setuptools \\
@@ -1034,6 +1036,7 @@ def create_dockerfile_buildbase(ddir, dockerfile_name, argmap):
       && rm -rf /var/lib/apt/lists/*
 
 RUN pip3 install --upgrade \\
+          auditwheel \\
           build \\
           docker \\
           virtualenv \\
@@ -1171,10 +1174,21 @@ def create_dockerfile_linux(
 
 WORKDIR /opt/tritonserver
 COPY NVIDIA_Deep_Learning_Container_License.pdf .
-RUN find /opt/tritonserver/python -maxdepth 1 -type f -name \\
-    "tritonserver-*.whl" | xargs -I {{}} pip install --upgrade {{}}[{FLAGS.triton_wheels_dependencies_group}] && \\
-    find /opt/tritonserver/python -maxdepth 1 -type f -name \\
-    "tritonfrontend-*.whl" | xargs -I {{}} pip install --upgrade {{}}[{FLAGS.triton_wheels_dependencies_group}]
+# TRI-1118 — fail fast if either tritonserver or tritonfrontend wheel is
+# missing from /opt/tritonserver/python. The legacy `find | xargs -I` is
+# a silent no-op when find returns zero matches: xargs runs the command
+# zero times and the layer succeeds, masking the gap until something
+# downstream (a wheel publish job, or pip install tritonfrontend)
+# discovers nothing was actually installed. Check existence first.
+RUN set -e; \\
+    for pkg in tritonserver tritonfrontend; do \\
+      wheels=$(find /opt/tritonserver/python -maxdepth 1 -type f -name "${{pkg}}-*.whl"); \\
+      if [ -z "$wheels" ]; then \\
+        echo "ERROR: ${{pkg}}-*.whl missing from /opt/tritonserver/python -- build did not stage the wheel into the image" >&2; \\
+        exit 1; \\
+      fi; \\
+      printf '%s\\n' "$wheels" | xargs -I {{}} pip install --upgrade "{{}}[{FLAGS.triton_wheels_dependencies_group}]"; \\
+    done
 
 RUN pip3 install -r python/openai/requirements.txt
 
@@ -1301,6 +1315,7 @@ def dockerfile_prepare_container_linux(argmap, backends, enable_gpu, target_mach
               libgoogle-perftools-dev \\
               libjemalloc-dev \\
               libnuma-dev \\
+              libssl-dev \\
               wget \\
               {backend_dependencies} \\
               python3-pip \\
@@ -1664,6 +1679,25 @@ def create_docker_build_script(script_name, container_install_dir, container_ci_
                     ),
                 ]
 
+        # TRI-1118 — propagate TRITON_RELEASE_VERSION into the wheel build
+        # only when it was actually set in main() (release-semantic version
+        # or explicit --release-version). Dev / pre-release builds leave it
+        # unset so build_wheel.py reads the in-tree TRITON_VERSION file and
+        # takes the PEP 817 variant path.
+        if "TRITON_RELEASE_VERSION" in os.environ:
+            runargs += [
+                "-e",
+                f"TRITON_RELEASE_VERSION={os.environ['TRITON_RELEASE_VERSION']}",
+            ]
+        # TRI-1118 — forward PEP 427 build-tag and PEP 817 nv-part inputs so
+        # build_wheel.py inside the buildbase container can emit pipeline-correct
+        # wheel filenames. CUDA_VERSION is deliberately NOT forwarded -- the
+        # container's own CUDA base image defines it; host CUDA may differ
+        # (see core/python/build_wheel.py:_detect_cuda_version).
+        for var in ("CI_PIPELINE_ID", "NVIDIA_UPSTREAM_VERSION", "NVIDIA_BUILD_ID"):
+            if os.environ.get(var):
+                runargs += ["-e", f"{var}={os.environ[var]}"]
+
         runargs += ["tritonserver_buildbase"]
 
         runargs += ["./cmake_build"]
@@ -1787,6 +1821,18 @@ def core_build(
     # [FIXME] Placing the tritonserver and tritonfrontend wheel files in 'python' for now,
     # should be uploaded to pip registry to be able to install directly
     cmake_script.mkdir(os.path.join(install_dir, "python"))
+    # TRI-1118 — tritonfrontend wheel is built by the triton-server
+    # sub-build's `frontend-server-wheel` target, but its CMake
+    # `install(DIRECTORY ${WHEEL_OUT_DIR})` rule doesn't traverse the
+    # outer install pass, so the wheel never lands in
+    # repo_install_dir/python/ alongside the core's tritonserver wheel.
+    # Pull it in from the build tree directly so the subsequent
+    # `triton*.whl` glob picks it up. Safe no-op when the wheel is
+    # absent (e.g. downstream builds that disable the frontend).
+    cmake_script.cmd(
+        f"find {repo_build_dir} -path '*/wheel/dist/tritonfrontend-*.whl' "
+        f"-exec cp {{}} {os.path.join(repo_install_dir, 'python')}/ \\;"
+    )
     cmake_script.cp(
         os.path.join(repo_install_dir, "python", "triton*.whl"),
         os.path.join(install_dir, "python"),
@@ -2466,8 +2512,12 @@ def enable_all():
     parser.add_argument(
         "--release-version",
         required=False,
-        default=DEFAULT_TRITON_VERSION_MAP["release_version"],
-        help="This flag sets the release version for Triton Inference Server to be built. Default: the latest released version.",
+        default=None,
+        help="Override the wheel base version (TRI-1118). When set, exported "
+        "as TRITON_RELEASE_VERSION so build_wheel.py uses it as the bare "
+        "PEP 440 version. When unset, build.py falls back to --version when "
+        "it matches X.Y.Z release-semantic, otherwise leaves the env var "
+        "unset and lets the in-tree TRITON_VERSION file rule (PEP 817 path).",
     )
     parser.add_argument(
         "--triton-container-version",
@@ -2596,6 +2646,22 @@ def enable_all():
     if FLAGS.version is None:
         FLAGS.version = DEFAULT_TRITON_VERSION_MAP["release_version"]
 
+    # TRI-1118 — choose whether to export TRITON_RELEASE_VERSION based on
+    # the resolved Triton version:
+    #   - --release-version explicitly passed     -> use it (override)
+    #   - FLAGS.version matches X.Y.Z             -> propagate FLAGS.version
+    #   - anything else (dev / pre-release / etc) -> leave unset so the
+    #                                                wheel build reads the
+    #                                                in-tree TRITON_VERSION
+    #                                                file (PEP 817 variant).
+    # The container build forwards this env var via `docker run -e`
+    # (create_docker_build_script). For --no-container-build, the
+    # cmake_build subprocess inherits the host env directly.
+    if FLAGS.release_version is not None:
+        os.environ.setdefault("TRITON_RELEASE_VERSION", FLAGS.release_version)
+    elif re.match(r"^\d+\.\d+\.\d+$", FLAGS.version):
+        os.environ.setdefault("TRITON_RELEASE_VERSION", FLAGS.version)
+
     if FLAGS.build_parallel is None:
         FLAGS.build_parallel = multiprocessing.cpu_count() * 2
 

deploy/aws/values.yaml

@@ -27,7 +27,7 @@
 replicaCount: 1
 
 image:
-  imageName: nvcr.io/nvidia/tritonserver:26.05-py3
+  imageName: nvcr.io/nvidia/tritonserver:26.06-py3
   pullPolicy: IfNotPresent
   modelRepositoryPath: s3://triton-inference-server-repository/model_repository
   numGpus: 1

deploy/fleetcommand/Chart.yaml

@@ -26,7 +26,7 @@
 
 apiVersion: v1
 # appVersion is the Triton version; update when changing release
-appVersion: 2.69.0
+appVersion: 2.70.0
 description: Triton Inference Server (Fleet Command)
 name: triton-inference-server
 # version is the Chart version; update when changing anything in the chart

deploy/fleetcommand/values.yaml

@@ -27,7 +27,7 @@
 replicaCount: 1
 
 image:
-  imageName: nvcr.io/nvidia/tritonserver:26.05-py3
+  imageName: nvcr.io/nvidia/tritonserver:26.06-py3
   pullPolicy: IfNotPresent
   numGpus: 1
   serverCommand: tritonserver
@@ -47,13 +47,13 @@ image:
     #
     # To set model control mode, uncomment and configure below
     # TODO: Fix the following url, it is invalid
-    # See https://github.com/triton-inference-server/server/blob/r26.05/docs/user_guide/model_management.md
+    # See https://github.com/triton-inference-server/server/blob/r26.06/docs/user_guide/model_management.md
     #  for more details
     #- --model-control-mode=explicit|poll|none
     #
     # Additional server args
     #
-    # see https://github.com/triton-inference-server/server/blob/r26.05/README.md
+    # see https://github.com/triton-inference-server/server/blob/r26.06/README.md
     #  for more details
 
 service:

deploy/gcp/values.yaml

@@ -27,7 +27,7 @@
 replicaCount: 1
 
 image:
-  imageName: nvcr.io/nvidia/tritonserver:26.05-py3
+  imageName: nvcr.io/nvidia/tritonserver:26.06-py3
   pullPolicy: IfNotPresent
   modelRepositoryPath: gs://triton-inference-server-repository/model_repository
   numGpus: 1

deploy/gke-marketplace-app/benchmark/perf-analyzer-script/triton_client.yaml

@@ -33,7 +33,7 @@ metadata:
   namespace: default
 spec:
   containers:
-  - image: nvcr.io/nvidia/tritonserver:26.05-py3-sdk
+  - image: nvcr.io/nvidia/tritonserver:26.06-py3-sdk
     imagePullPolicy: Always
     name: nv-triton-client
     securityContext:

deploy/gke-marketplace-app/server-deployer/build_and_push.sh

@@ -28,8 +28,8 @@
 export REGISTRY=gcr.io/$(gcloud config get-value project | tr ':' '/')
 export APP_NAME=tritonserver
 export MAJOR_VERSION=2.67
-export MINOR_VERSION=2.69.0
-export NGC_VERSION=26.05-py3
+export MINOR_VERSION=2.70.0
+export NGC_VERSION=26.06-py3
 
 docker pull nvcr.io/nvidia/$APP_NAME:$NGC_VERSION
 

deploy/gke-marketplace-app/server-deployer/chart/triton/Chart.yaml

@@ -28,4 +28,4 @@ apiVersion: v1
 appVersion: "2.68"
 description: Triton Inference Server
 name: triton-inference-server
-version: 2.69.0
+version: 2.70.0

deploy/gke-marketplace-app/server-deployer/chart/triton/values.yaml

@@ -31,14 +31,14 @@ maxReplicaCount: 3
 tritonProtocol: HTTP
 # HPA GPU utilization autoscaling target
 HPATargetAverageValue: 85
-modelRepositoryPath: gs://triton_sample_models/26.05
-publishedVersion: '2.69.0'
+modelRepositoryPath: gs://triton_sample_models/26.06
+publishedVersion: '2.70.0'
 gcpMarketplace: true
 
 image:
   registry: gcr.io
   repository: nvidia-ngc-public/tritonserver
-  tag: 26.05-py3
+  tag: 26.06-py3
   pullPolicy: IfNotPresent
   # modify the model repository here to match your GCP storage bucket
   numGpus: 1

deploy/gke-marketplace-app/server-deployer/data-test/schema.yaml

@@ -27,7 +27,7 @@
 x-google-marketplace:
   schemaVersion: v2
   applicationApiVersion: v1beta1
-  publishedVersion: '2.69.0'
+  publishedVersion: '2.70.0'
   publishedVersionMetadata:
     releaseNote: >-
       Initial release.

deploy/gke-marketplace-app/server-deployer/schema.yaml

@@ -27,7 +27,7 @@
 x-google-marketplace:
   schemaVersion: v2
   applicationApiVersion: v1beta1
-  publishedVersion: '2.69.0'
+  publishedVersion: '2.70.0'
   publishedVersionMetadata:
     releaseNote: >-
       Initial release.
@@ -89,7 +89,7 @@ properties:
   modelRepositoryPath:
     type: string
     title: Bucket where models are stored. Please make sure the user/service account to create the GKE app has permission to this GCS bucket. Read Triton documentation on configs and formatting details, supporting TensorRT, TensorFlow, Pytorch, Onnx ... etc.
-    default: gs://triton_sample_models/26.05
+    default: gs://triton_sample_models/26.06
   image.ldPreloadPath:
     type: string
     title: Leave this empty by default. Triton allows users to create custom layers for backend such as TensorRT plugin, the compiled shared library must be provided via LD_PRELOAD environment variable.

deploy/gke-marketplace-app/trt-engine/README.md

@@ -33,7 +33,7 @@
 ```
 docker run --gpus all -it --network host \
     --shm-size=1g --ulimit memlock=-1 --ulimit stack=67108864 \
-    -v ~:/scripts nvcr.io/nvidia/tensorrt:26.05-py3
+    -v ~:/scripts nvcr.io/nvidia/tensorrt:26.06-py3
 
 pip install onnx six torch tf2onnx tensorflow
 
@@ -57,7 +57,7 @@ mkdir -p engines
 
 python3 builder.py -m models/fine-tuned/bert_tf_ckpt_large_qa_squad2_amp_128_v19.03.1/model.ckpt -o engines/bert_large_int8_bs1_s128.engine -b 1 -s 128 -c models/fine-tuned/bert_tf_ckpt_large_qa_squad2_amp_128_v19.03.1/ -v models/fine-tuned/bert_tf_ckpt_large_qa_squad2_amp_128_v19.03.1/vocab.txt --int8 --fp16 --strict --calib-num 1 -iln -imh
 
-gsutil cp bert_large_int8_bs1_s128.engine gs://triton_sample_models/26.05/bert/1/model.plan
+gsutil cp bert_large_int8_bs1_s128.engine gs://triton_sample_models/26.06/bert/1/model.plan
 ```
 
-For each Triton upgrade, container version used to generate the model, and the model path in GCS `gs://triton_sample_models/26.05/` should be updated accordingly with the correct version.
+For each Triton upgrade, container version used to generate the model, and the model path in GCS `gs://triton_sample_models/26.06/` should be updated accordingly with the correct version.

deploy/k8s-onprem/values.yaml

@@ -30,7 +30,7 @@ tags:
   openshift: false
 
 image:
-  imageName: nvcr.io/nvidia/tritonserver:26.05-py3
+  imageName: nvcr.io/nvidia/tritonserver:26.06-py3
   pullPolicy: IfNotPresent
   modelRepositoryServer: < Replace with the IP Address of your file server >
   modelRepositoryPath: /srv/models

deploy/oci/values.yaml

@@ -27,7 +27,7 @@
 replicaCount: 1
 
 image:
-  imageName: nvcr.io/nvidia/tritonserver:26.05-py3
+  imageName: nvcr.io/nvidia/tritonserver:26.06-py3
   pullPolicy: IfNotPresent
   modelRepositoryPath: s3://https://<OCI_NAMESPACE>.compat.objectstorage.<OCI_REGION>.oraclecloud.com:443/triton-inference-server-repository
   numGpus: 1

docs/customization_guide/compose.md

@@ -46,8 +46,8 @@ The `compose.py` script can be found in the
 Simply clone the repository and run `compose.py` to create a custom container.
 Note: Created container version will depend on the branch that was cloned.
 For example branch
- [r26.05](https://github.com/triton-inference-server/server/tree/r26.05)
-should be used to create a image based on the NGC 26.05 Triton release.
+ [r26.06](https://github.com/triton-inference-server/server/tree/r26.06)
+should be used to create a image based on the NGC 26.06 Triton release.
 
 `compose.py` provides `--backend`, `--repoagent` options that allow you to
 specify which backends and repository agents to include in the custom image.
@@ -78,20 +78,20 @@ For example, running
 ```
 python3 compose.py --backend pytorch --repoagent checksum
 ```
-on branch [r26.05](https://github.com/triton-inference-server/server/tree/r26.05) pulls:
-- `min` container `nvcr.io/nvidia/tritonserver:26.05-py3-min`
-- `full` container `nvcr.io/nvidia/tritonserver:26.05-py3`
+on branch [r26.06](https://github.com/triton-inference-server/server/tree/r26.06) pulls:
+- `min` container `nvcr.io/nvidia/tritonserver:26.06-py3-min`
+- `full` container `nvcr.io/nvidia/tritonserver:26.06-py3`
 
 Alternatively, users can specify the version of Triton container to pull from
 any branch by either:
 1. Adding flag `--container-version <container version>` to branch
 ```
-python3 compose.py --backend pytorch --repoagent checksum --container-version 26.05
+python3 compose.py --backend pytorch --repoagent checksum --container-version 26.06
 ```
 2. Specifying `--image min,<min container image name> --image full,<full container image name>`.
    The user is responsible for specifying compatible `min` and `full` containers.
 ```
-python3 compose.py --backend pytorch --repoagent checksum --image min,nvcr.io/nvidia/tritonserver:26.05-py3-min --image full,nvcr.io/nvidia/tritonserver:26.05-py3
+python3 compose.py --backend pytorch --repoagent checksum --image min,nvcr.io/nvidia/tritonserver:26.06-py3-min --image full,nvcr.io/nvidia/tritonserver:26.06-py3
 ```
 Method 1 and 2 will result in the same composed container. Furthermore,
 `--image` flag overrides the `--container-version` flag when both are specified.
@@ -102,8 +102,8 @@ Note:
 2. vLLM and TensorRT-LLM backends are currently not supported backends for
 `compose.py`. If you want to build additional backends on top of these backends,
 it would be better to [build it yourself](#build-it-yourself) by using
-`nvcr.io/nvidia/tritonserver:26.05-vllm-python-py3` or
-`nvcr.io/nvidia/tritonserver:26.05-trtllm-python-py3` as a `min` container.
+`nvcr.io/nvidia/tritonserver:26.06-vllm-python-py3` or
+`nvcr.io/nvidia/tritonserver:26.06-trtllm-python-py3` as a `min` container.
 
 
 ### CPU-only container composition