Skip to content

Commit 0c3ffbf

Browse files
author
pytorchbot
committed
2026-09-16 nightly release (e53ff9a)
1 parent e6be5d6 commit 0c3ffbf

1,400 files changed

Lines changed: 10865 additions & 2543 deletions

File tree

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

‎.github/scripts/filter-executorch-cuda-arches.py‎

Lines changed: 0 additions & 88 deletions
This file was deleted.

‎.github/scripts/filter-matrix.py‎

Lines changed: 19 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -1,8 +1,11 @@
11
#!/usr/bin/env python3
2+
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
3+
# SPDX-License-Identifier: BSD-3-Clause
24

35
import argparse
46
import json
57
import os
8+
import re
69
import sys
710
from typing import Any, Dict, List
811

@@ -13,16 +16,19 @@
1316
# jetpack 6.2 only officially supports python 3.10 and cu126
1417
jetpack_python_versions: List[str] = ["3.10"]
1518
jetpack_cuda_versions: List[str] = ["cu126"]
16-
# CUDA 12.6 wheels are published for x86_64 only. Keep the Arm matrices on
17-
# CUDA 13, including Windows Arm/AArch64.
18-
x86_cuda_versions: List[str] = ["cu126", "cu130", "cu132", "cu134"]
19+
# CUDA 13 wheels are supported on x86_64 and Arm, including Windows Arm/AArch64.
20+
x86_cuda_versions: List[str] = ["cu130", "cu132", "cu134"]
1921
arm_cuda_versions: List[str] = ["cu130", "cu132", "cu134"]
2022

2123
# For PRs we build/test a single representative config to keep cycle time short.
2224
# Full matrix runs on main / nightly / release branches.
2325
PR_PYTHON_VERSION: str = "3.12"
2426
PR_CUDA_VERSION: str = "cu134"
2527

28+
# The delegate links CUDA 13 libraries. This narrows the row lists above, it does not widen
29+
# them: a new minor still has to be added there first.
30+
EXECUTORCH_CUDA_MAJOR: str = "13"
31+
2632
jetpack_container_image: str = "nvcr.io/nvidia/l4t-jetpack:r36.4.0"
2733
sbsa_container_image: str = "quay.io/pypa/manylinux_2_39_aarch64"
2834

@@ -135,6 +141,12 @@ def main(args: list[str]) -> None:
135141
default="false",
136142
)
137143

144+
parser.add_argument(
145+
"--executorch-runtime",
146+
action="store_true",
147+
help="keep only CUDA 13 rows supported by the ExecuTorch delegate wheel",
148+
)
149+
138150
options = parser.parse_args(args)
139151
if options.matrix == "":
140152
raise ValueError("--matrix needs to be provided")
@@ -151,6 +163,10 @@ def main(args: list[str]) -> None:
151163
filtered_includes = []
152164

153165
for item in includes:
166+
if options.executorch_runtime and not re.fullmatch(
167+
rf"cu{EXECUTORCH_CUDA_MAJOR}\d+", item["desired_cuda"]
168+
):
169+
continue
154170
if filter_matrix_item(
155171
item,
156172
options.jetpack == "true",

‎.github/scripts/generate-release-matrix.py‎

Lines changed: 4 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -1,12 +1,14 @@
11
#!/usr/bin/env python3
2+
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
3+
# SPDX-License-Identifier: BSD-3-Clause
24

35
import argparse
46
import json
57
import sys
68

79
RELEASE_CUDA_VERSION = {
8-
"wheel": ["cu126", "cu130", "cu132", "cu134"],
9-
"tarball": ["cu126", "cu130", "cu132", "cu134"],
10+
"wheel": ["cu130", "cu132", "cu134"],
11+
"tarball": ["cu130", "cu132", "cu134"],
1012
}
1113
RELEASE_PYTHON_VERSION = {
1214
"wheel": ["3.10", "3.11", "3.12", "3.13", "3.14"],
@@ -49,13 +51,6 @@ def main(args: list[str]) -> None:
4951
includes = matrix_dict["include"]
5052
filtered_includes = []
5153
for item in includes:
52-
# CUDA 12.6 release artifacts are available only for x86_64. This
53-
# excludes both Linux AArch64 and Windows Arm/AArch64 matrix entries.
54-
if item["desired_cuda"] == "cu126" and item["gpu_arch_type"] in {
55-
"cuda-aarch64",
56-
"cuda-arm64",
57-
}:
58-
continue
5954
if (
6055
item["desired_cuda"] in cuda_versions
6156
and item["python_version"] in python_versions

‎.github/scripts/generate-tensorrt-test-matrix.py‎

Lines changed: 4 additions & 58 deletions
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,6 @@
11
#!/usr/bin/env python3
2+
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
3+
# SPDX-License-Identifier: BSD-3-Clause
24

35
import argparse
46
import copy
@@ -12,8 +14,8 @@
1214
# channel: test if the future tensorRT version test workflow is triggered from the release branch(release/2.5 etc....)
1315
CUDA_VERSIONS_DICT = {
1416
"nightly": ["cu130"],
15-
"test": ["cu126", "cu128", "cu130"],
16-
"release": ["cu126", "cu128", "cu130"],
17+
"test": ["cu130"],
18+
"release": ["cu130"],
1719
}
1820

1921
# please update the python version you want to test with the future tensorRT version here
@@ -32,34 +34,6 @@
3234
"urls": "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.3.0/zip/TensorRT-10.3.0.26.Windows.win10.cuda-12.5.zip",
3335
"strip_prefix": "TensorRT-10.3.0.26",
3436
},
35-
"10.7.0": {
36-
"urls": "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.7.0/zip/TensorRT-10.7.0.23.Windows.win10.cuda-12.6.zip",
37-
"strip_prefix": "TensorRT-10.7.0.23",
38-
},
39-
"10.8.0": {
40-
"urls": "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.8.0/zip/TensorRT-10.8.0.43.Windows.win10.cuda-12.8.zip",
41-
"strip_prefix": "TensorRT-10.8.0.43",
42-
},
43-
"10.9.0": {
44-
"urls": "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.9.0/zip/TensorRT-10.9.0.34.Windows.win10.cuda-12.8.zip",
45-
"strip_prefix": "TensorRT-10.9.0.34",
46-
},
47-
"10.10.0": {
48-
"urls": "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.10.0/zip/TensorRT-10.10.0.31.Windows.win10.cuda-12.9.zip",
49-
"strip_prefix": "TensorRT-10.10.0.31",
50-
},
51-
"10.11.0": {
52-
"urls": "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.11.0/zip/TensorRT-10.11.0.33.Windows.win10.cuda-12.9.zip",
53-
"strip_prefix": "TensorRT-10.11.0.33",
54-
},
55-
"10.12.0": {
56-
"urls": "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.12.0/zip/TensorRT-10.12.0.36.Windows.win10.cuda-12.9.zip",
57-
"strip_prefix": "TensorRT-10.12.0.36",
58-
},
59-
"10.13.2": {
60-
"urls": "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.2/zip/TensorRT-10.13.2.6.Windows.win10.cuda-12.9.zip",
61-
"strip_prefix": "TensorRT-10.13.2.6",
62-
},
6337
"10.14.1": {
6438
"urls": "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.14.1/zip/TensorRT-10.14.1.48.Windows.win10.cuda-12.9.zip",
6539
"strip_prefix": "TensorRT-10.14.1.48",
@@ -70,34 +44,6 @@
7044
"urls": "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.3.0/tars/TensorRT-10.3.0.26.Linux.x86_64-gnu.cuda-12.5.tar.gz",
7145
"strip_prefix": "TensorRT-10.3.0.26",
7246
},
73-
"10.7.0": {
74-
"urls": "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.7.0/tars/TensorRT-10.7.0.23.Linux.x86_64-gnu.cuda-12.6.tar.gz",
75-
"strip_prefix": "TensorRT-10.7.0.23",
76-
},
77-
"10.8.0": {
78-
"urls": "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.8.0/tars/TensorRT-10.8.0.43.Linux.x86_64-gnu.cuda-12.8.tar.gz",
79-
"strip_prefix": "TensorRT-10.8.0.43",
80-
},
81-
"10.9.0": {
82-
"urls": "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.9.0/tars/TensorRT-10.9.0.34.Linux.x86_64-gnu.cuda-12.8.tar.gz",
83-
"strip_prefix": "TensorRT-10.9.0.34",
84-
},
85-
"10.10.0": {
86-
"urls": "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.10.0/tars/TensorRT-10.10.0.31.Linux.x86_64-gnu.cuda-12.9.tar.gz",
87-
"strip_prefix": "TensorRT-10.10.0.31",
88-
},
89-
"10.11.0": {
90-
"urls": "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.11.0/tars/TensorRT-10.11.0.33.Linux.x86_64-gnu.cuda-12.9.tar.gz",
91-
"strip_prefix": "TensorRT-10.11.0.33",
92-
},
93-
"10.12.0": {
94-
"urls": "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.12.0/tars/TensorRT-10.12.0.36.Linux.x86_64-gnu.cuda-12.9.tar.gz",
95-
"strip_prefix": "TensorRT-10.12.0.36",
96-
},
97-
"10.13.2": {
98-
"urls": "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.13.2/tars/TensorRT-10.13.2.6.Linux.x86_64-gnu.cuda-12.9.tar.gz",
99-
"strip_prefix": "TensorRT-10.13.2.6",
100-
},
10147
"10.14.1": {
10248
"urls": "https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.14.1/tars/TensorRT-10.14.1.48.Linux.x86_64-gnu.cuda-12.9.tar.gz",
10349
"strip_prefix": "TensorRT-10.14.1.48",

‎.github/scripts/install-cuda-aarch64.sh‎

Lines changed: 4 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -12,15 +12,7 @@ install_cuda_aarch64() {
1212
dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel8/sbsa/cuda-rhel8.repo
1313

1414
# nccl version must match libtorch_cuda.so was built with
15-
if [[ ${CU_VERSION:0:4} == "cu12" ]]; then
16-
# cu12: https://github.com/pytorch/pytorch/blob/main/.ci/docker/ci_commit_pins/nccl-cu12.txt
17-
if [[ ${CU_VERSION} == "cu126" ]]; then
18-
nccl_version="2.24.3-1"
19-
else
20-
echo "Unsupported CUDA version: ${CU_VERSION}"
21-
exit 1
22-
fi
23-
elif [[ ${CU_VERSION:0:4} == "cu13" ]]; then
15+
if [[ ${CU_VERSION:0:4} == "cu13" ]]; then
2416
# cu13: https://github.com/pytorch/pytorch/blob/main/.ci/docker/ci_commit_pins/nccl.txt
2517
# NVIDIA RHEL/SBSA NCCL RPMs are published per CUDA minor.
2618
if [[ ${CU_VERSION} == "cu134" ]]; then
@@ -40,9 +32,11 @@ install_cuda_aarch64() {
4032
echo "Unsupported CUDA version: ${CU_VERSION}"
4133
exit 1
4234
fi
35+
else
36+
echo "Unsupported CUDA version: ${CU_VERSION}"
37+
exit 1
4338
fi
4439

45-
# CUDA 12 NCCL packages use the same minor version as the toolkit.
4640
nccl_cuda_version=${nccl_cuda_version:-${CU_DOT_VER}}
4741

4842
# Keep these transactions separate so an invalid NCCL pin cannot silently

‎.github/scripts/install-torch-tensorrt.sh‎

Lines changed: 7 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -43,16 +43,10 @@ python -m pip uninstall -y torch torchvision
4343
python -m pip install --force-reinstall --pre ${TORCHVISION} --index-url ${INDEX_URL} --extra-index-url https://pypi.org/simple
4444
python -m pip install --force-reinstall --pre ${TORCH} --index-url ${INDEX_URL} --extra-index-url https://pypi.org/simple
4545

46-
# Prepend the venv's NVIDIA CUDA runtime libs to LD_LIBRARY_PATH. The two majors ship
47-
# different layouts: nvidia-cuda-runtime-cu12 installs nvidia/cuda_runtime/lib/libcudart.so.12
48-
# while the CUDA 13 line installs nvidia/cu13/lib/libcudart.so.13. Naming only the cu13 path
49-
# left every CUDA 12 row without a CUDA runtime on the search path, so a binary linked against
50-
# libcudart died at startup with "libcudart.so.12: cannot open shared object file" even though
51-
# the package was installed.
46+
# Prepend the venv's NVIDIA CUDA runtime libs to LD_LIBRARY_PATH.
5247
SITE_PACKAGES="$(python -c 'import sysconfig; print(sysconfig.get_path("platlib"))')"
5348
case "${CU_VERSION}" in
5449
cu13*) CUDA_RUNTIME_LIB_DIR="${SITE_PACKAGES}/nvidia/cu13/lib" ;;
55-
cu12*) CUDA_RUNTIME_LIB_DIR="${SITE_PACKAGES}/nvidia/cuda_runtime/lib" ;;
5650
*) CUDA_RUNTIME_LIB_DIR="" ;;
5751
esac
5852
if [[ -n "${CUDA_RUNTIME_LIB_DIR}" ]]; then
@@ -61,9 +55,13 @@ fi
6155

6256
# Install Torch-TensorRT
6357
if [[ ${PLATFORM} == win32 ]]; then
64-
python -m pip install ${RUNNER_ARTIFACT_DIR}/torch_tensorrt*.whl
58+
# pin-check: no-nightly -- Windows installs only the main wheel, without the Linux companion.
59+
python -m pip install ${RUNNER_ARTIFACT_DIR}/torch_tensorrt*.whl || exit 1
6560
else
66-
python -m pip install /opt/torch-tensorrt-builds/torch_tensorrt*.whl --use-deprecated=legacy-resolver
61+
# The companion requires the nightly ExecuTorch channel even on test/release jobs.
62+
# Exit explicitly: the caller appends its test script and this file does not use set -e.
63+
python -m pip install /opt/torch-tensorrt-builds/torch_tensorrt*.whl --use-deprecated=legacy-resolver \
64+
--extra-index-url "https://download.pytorch.org/whl/nightly/${CU_VERSION}" || exit 1
6765
fi
6866

6967
echo -e "Running test script";

‎.github/scripts/run_cpp_linter.py‎

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,3 +1,6 @@
1+
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
2+
# SPDX-License-Identifier: BSD-3-Clause
3+
14
import os
25
import json
36

‎.github/scripts/run_py_linter.py‎

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,3 +1,6 @@
1+
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
2+
# SPDX-License-Identifier: BSD-3-Clause
3+
14
import os
25
import json
36
from github import Github

0 commit comments

Comments
 (0)