8000 Update · pytorch/pytorch@fd183e3 · GitHub
[go: up one dir, main page]
Skip to content

Commit fd183e3

Browse files
xmfanxmfan
committed
Update
[ghstack-poisoned]
2 parents e77179b + 056327f commit fd183e3

File tree

259 files changed

+6858
-1721
lines changed

Some content is hidden

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

‎.ci/docker/common/install_cuda.sh

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -240,7 +240,7 @@ function prune_126 {
240240
}
241241

242242
function install_128 {
243-
CUDNN_VERSION=9.7.1.26
243+
CUDNN_VERSION=9.8.0.87
244244
echo "Installing CUDA 12.8.0 and cuDNN ${CUDNN_VERSION} and NCCL ${NCCL_VERSION} and cuSparseLt-0.6.3"
245245
rm -rf /usr/local/cuda-12.8 /usr/local/cuda
246246
# install CUDA 12.8.0 in the same container

‎.ci/docker/common/install_cuda_aarch64.sh

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -161,7 +161,7 @@ function prune_126 {
161161
}
162162

163163
function install_128 {
164-
CUDNN_VERSION=9.7.1.26
164+
CUDNN_VERSION=9.8.0.87
165165
echo "Installing CUDA 12.8.0 and cuDNN ${CUDNN_VERSION} and NCCL ${NCCL_VERSION} and cuSparseLt-0.6.3"
166166
rm -rf /usr/local/cuda-12.8 /usr/local/cuda
167167
# install CUDA 12.8.0 in the same container

‎.ci/docker/common/install_cudnn.sh

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -5,7 +5,7 @@ if [[ -n "${CUDNN_VERSION}" ]]; then
55
mkdir tmp_cudnn
66
pushd tmp_cudnn
77
if [[ ${CUDA_VERSION:0:4} == "12.8" ]]; then
8-
CUDNN_NAME="cudnn-linux-x86_64-9.7.1.26_cuda12-archive"
8+
CUDNN_NAME="cudnn-linux-x86_64-9.8.0.87_cuda12-archive"
99
elif [[ ${CUDA_VERSION:0:4} == "12.6" ]]; then
1010
CUDNN_NAME="cudnn-linux-x86_64-9.5.1.17_cuda12-archive"
1111
elif [[ ${CUDA_VERSION:0:2} == "12" ]]; then

‎.ci/docker/requirements-ci.txt

Lines changed: 4 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -102,10 +102,10 @@ networkx==2.8.8
102102
#Pinned versions: 2.8.8
103103
#test that import: functorch
104104

105-
#ninja
106-
#Description: build system. Note that it install from
107-
#here breaks things so it is commented out
108-
#Pinned versions: 1.10.0.post1
105+
ninja==1.11.1.3
106+
#Description: build system. Used in some tests. Used in build to generate build
107+
#time tracing information
108+
#Pinned versions: 1.11.1.3
109109
#test that import: run_test.py, test_cpp_extensions_aot.py,test_determination.py
110110

111111
numba==0.49.0 ; python_version < "3.9"
@@ -365,7 +365,6 @@ PyYAML
365365
pyzstd
366366
setuptools
367367

368-
ninja==1.11.1 ; platform_machine == "aarch64"
369368
scons==4.5.2 ; platform_machine == "aarch64"
370369

371370
pulp==2.9.0 ; python_version >= "3.8"

‎.ci/pytorch/check_binary.sh

Lines changed: 4 additions & 16 deletions
Original file line numberDiff line numberDiff line change
@@ -73,26 +73,14 @@ fi
7373
# Check GCC ABI
7474
###############################################################################
7575

76-
# NOTE [ Building libtorch with old vs. new gcc ABI ]
77-
#
78-
# Packages built with one version of ABI could not be linked against by client
79-
# C++ libraries that were compiled using the other version of ABI. Since both
80-
# gcc ABIs are still common in the wild, we need to support both ABIs. Currently:
81-
#
82-
# - All the nightlies built on CentOS 7 + devtoolset7 use the old gcc ABI.
83-
# - All the nightlies built on Ubuntu 16.04 + gcc 5.4 use the new gcc ABI.
76+
# NOTE: As of https://github.com/pytorch/pytorch/issues/126551 we only produce
77+
# wheels with cxx11-abi
8478

8579
echo "Checking that the gcc ABI is what we expect"
8680
if [[ "$(uname)" != 'Darwin' ]]; then
8781
function is_expected() {
88-
if [[ "$DESIRED_DEVTOOLSET" == *"cxx11-abi"* || "$DESIRED_CUDA" == *"rocm"* ]]; then
89-
if [[ "$1" -gt 0 || "$1" == "ON " ]]; then
90-
echo 1
91-
fi
92-
else
93-
if [[ -z "$1" || "$1" == 0 || "$1" == "OFF" ]]; then
94-
echo 1
95-
fi
82+
if [[ "$1" -gt 0 || "$1" == "ON " ]]; then
83+
echo 1
9684
fi
9785
}
9886

‎.ci/pytorch/smoke_test/check_binary_symbols.py

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -121,9 +121,9 @@ def main() -> None:
121121
else:
122122
install_root = Path(distutils.sysconfig.get_python_lib()) / "torch"
123123

124-
libtorch_cpu_path = install_root / "lib" / "libtorch_cpu.so"
125-
pre_cxx11_abi = "cxx11-abi" not in os.getenv("DESIRED_DEVTOOLSET", "")
126-
check_lib_symbols_for_abi_correctness(libtorch_cpu_path, pre_cxx11_abi)
124+
libtorch_cpu_path = str(install_root / "lib" / "libtorch_cpu.so")
125+
# NOTE: All binaries are built with cxx11abi now
126+
check_lib_symbols_for_abi_correctness(libtorch_cpu_path, False)
127127

128128

129129
if __name__ == "__main__":

‎.ci/pytorch/test.sh

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -483,6 +483,8 @@ elif [[ "${TEST_CONFIG}" == *aot_eager* ]]; then
483483
DYNAMO_BENCHMARK_FLAGS+=(--backend aot_eager)
484484
elif [[ "${TEST_CONFIG}" == *aot_inductor* ]]; then
485485
DYNAMO_BENCHMARK_FLAGS+=(--export-aot-inductor)
486+
elif 10000 [[ "${TEST_CONFIG}" == *max_autotune_inductor* ]]; then
487+
DYNAMO_BENCHMARK_FLAGS+=(--inductor --inductor-compile-mode max-autotune)
486488
elif [[ "${TEST_CONFIG}" == *inductor* && "${TEST_CONFIG}" != *perf* ]]; then
487489
DYNAMO_BENCHMARK_FLAGS+=(--inductor)
488490
fi
@@ -760,6 +762,8 @@ test_dynamo_benchmark() {
760762
fi
761763
elif [[ "${TEST_CONFIG}" == *aot_inductor* ]]; then
762764
test_single_dynamo_benchmark "inference" "$suite" "$shard_id" --inference --bfloat16 "$@"
765+
elif [[ "${TEST_CONFIG}" == *max_autotune_inductor* ]]; then
766+
test_single_dynamo_benchmark "inference" "$suite" "$shard_id" --inference --bfloat16 "$@"
763767
else
764768
test_single_dynamo_benchmark "inference" "$suite" "$shard_id" --inference --bfloat16 "$@"
765769
test_single_dynamo_benchmark "training" "$suite" "$shard_id" --training --amp "$@"
@@ -1615,6 +1619,7 @@ elif [[ "${TEST_CONFIG}" == *inductor_cpp_wrapper* ]]; then
16151619
install_torchvision
16161620
checkout_install_torchbench hf_T5 llama moco
16171621
PYTHONPATH=$(pwd)/torchbench test_inductor_cpp_wrapper_shard "$SHARD_NUMBER"
1622+
test_inductor_aoti
16181623
elif [[ "${TEST_CONFIG}" == *inductor* ]]; then
16191624
install_torchvision
16201625
test_inductor_shard "${SHARD_NUMBER}"

‎.clang-tidy

Lines changed: 0 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -48,7 +48,6 @@ misc-*,
4848
-misc-no-recursion,
4949
-misc-non-private-member-variables-in-classes,
5050
-misc-unused-using-decls,
51-
-misc-use-internal-linkage,
5251
modernize-*,
5352
-modernize-macro-to-enum,
5453
-modernize-return-braced-init-list,

‎.github/actionlint.yaml

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -3,8 +3,11 @@ self-hosted-runner:
33
# GitHub hosted runner that actionlint doesn't recognize because actionlint version (1.6.21) is too old
44
- ubuntu-24.04
55
# GitHub hosted x86 Linux runners
6+
# TODO: Cleanup mentions of linux.20_04 when upgrade to linux.24_04 is complete
67
- linux.20_04.4x
78
- linux.20_04.16x
9+
- linux.24_04.4x
10+
- linux.24_04.16x
811
# Organization-wide AWS Linux Runners
912
- linux.large
1013
- linux.2xlarge

‎.github/actions/checkout-pytorch/action.yml

Lines changed: 38 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -24,8 +24,12 @@ runs:
2424
run: echo "IN_CONTAINER_RUNNER=$(if [ -f /.inarc ] || [ -f /.incontainer ]; then echo true ; else echo false; fi)" >> "$GITHUB_OUTPUT"
2525

2626
- name: Set up parallel fetch and clean workspace
27+
id: first-clean
28+
continue-on-error: true
2729
shell: bash
2830
if: ${{ steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'false' }}
31+
env:
32+
NO_SUDO: ${{ inputs.no-sudo }}
2933
run: |
3034
# Use all available CPUs for fetching
3135
cd "${GITHUB_WORKSPACE}"
@@ -35,14 +39,47 @@ runs:
3539
# Clean workspace. The default checkout action should also do this, but
3640
# do it here as well just in case
3741
if [[ -d .git ]]; then
38-
git clean -ffdx
42+
if [ -z "${NO_SUDO}" ]; then
43+
sudo git clean -ffdx
44+
else
45+
git clean -ffdx
46+
fi
3947
fi
4048
4149
- name: Checkout PyTorch
50+
id: first-checkout-attempt
51+
continue-on-error: true
4252
uses: actions/checkout@v4
4353
with:
4454
ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
4555
# --depth=1 for speed, manually fetch history and other refs as necessary
4656
fetch-depth: ${{ inputs.fetch-depth }}
4757
submodules: ${{ inputs.submodules }}
4858
show-progress: false
59+
60+
- name: Clean workspace (try again)
61+
if: ${{ steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'false' &&
62+
(steps.first-clean.outcome != 'success' || steps.first-checkout-attempt.outcome != 'success') }}
63+
shell: bash
64+
env:
65+
NO_SUDO: ${{ inputs.no-sudo }}
66+
run: |
67+
retry () {
68+
$* || (sleep 1 && $*) || (sleep 2 && $*) || (sleep 4 && $*) || (sleep 8 && $*)
69+
}
70+
echo "${GITHUB_WORKSPACE}"
71+
if [ -z "${NO_SUDO}" ]; then
72+
retry sudo rm -rf "${GITHUB_WORKSPACE}"
73+
else
74+
retry rm -rf "${GITHUB_WORKSPACE}"
75+
fi
76+
mkdir "${GITHUB_WORKSPACE}"
77+
78+
- name: Checkout PyTorch (try again)
79+
uses: actions/checkout@v4
80+
if: ${{ steps.first-clean.outcome != 'success' || steps.first-checkout-attempt.outcome != 'success' }}
81+
with:
82+
ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
83+
fetch-depth: ${{ inputs.fetch-depth }}
84+
submodules: ${{ inputs.submodules }}
85+
show-progress: false

0 commit comments

Comments
 (0)
0