pax_global_header 0000666 0000000 0000000 00000000064 15207546147 0014525 g ustar 00root root 0000000 0000000 52 comment=9f9ed92ae3fd07bb43872acd4ab53953f8e957c3
torchcodec-0.14.0/ 0000775 0000000 0000000 00000000000 15207546147 0013724 5 ustar 00root root 0000000 0000000 torchcodec-0.14.0/.clang-format 0000664 0000000 0000000 00000006260 15207546147 0016303 0 ustar 00root root 0000000 0000000 ---
AccessModifierOffset: -1
AlignAfterOpenBracket: AlwaysBreak
AlignConsecutiveMacros: false
AlignConsecutiveAssignments: false
AlignConsecutiveBitFields: false
AlignConsecutiveDeclarations: false
AlignEscapedNewlines: Left
AlignOperands: DontAlign
AlignTrailingComments: false
AllowAllArgumentsOnNextLine: true
AllowAllConstructorInitializersOnNextLine: true
AllowAllParametersOfDeclarationOnNextLine: false
AllowShortEnumsOnASingleLine: true
AllowShortBlocksOnASingleLine: Never
AllowShortCaseLabelsOnASingleLine: false
AllowShortFunctionsOnASingleLine: Empty
AllowShortLambdasOnASingleLine: All
AllowShortIfStatementsOnASingleLine: Never
AllowShortLoopsOnASingleLine: false
AlwaysBreakAfterReturnType: None
AlwaysBreakBeforeMultilineStrings: true
AlwaysBreakTemplateDeclarations: Yes
BinPackArguments: false
BinPackParameters: false
BreakBeforeBinaryOperators: None
BreakBeforeBraces: Attach
BreakInheritanceList: BeforeColon
BreakBeforeTernaryOperators: true
BreakConstructorInitializers: BeforeColon
BreakAfterJavaFieldAnnotations: false
BreakStringLiterals: false
ColumnLimit: 80
CommentPragmas: '^ IWYU pragma:'
CompactNamespaces: false
ConstructorInitializerAllOnOneLineOrOnePerLine: true
ConstructorInitializerIndentWidth: 4
ContinuationIndentWidth: 4
Cpp11BracedListStyle: true
DeriveLineEnding: true
DerivePointerAlignment: false
DisableFormat: false
FixNamespaceComments: true
ForEachMacros:
- FOR_EACH
- FOR_EACH_R
- FOR_EACH_RANGE
IncludeBlocks: Preserve
IncludeCategories:
- Regex: '^<.*\.h(pp)?>'
Priority: 1
- Regex: '^<.*'
Priority: 2
- Regex: '.*'
Priority: 3
IndentCaseLabels: true
IndentCaseBlocks: false
IndentGotoLabels: true
IndentPPDirectives: None
IndentExternBlock: AfterExternBlock
IndentWidth: 2
IndentWrappedFunctionNames: false
InsertTrailingCommas: None
JavaScriptQuotes: Leave
JavaScriptWrapImports: true
KeepEmptyLinesAtTheStartOfBlocks: false
MacroBlockBegin: ''
MacroBlockEnd: ''
MaxEmptyLinesToKeep: 1
NamespaceIndentation: None
ObjCBinPackProtocolList: Auto
ObjCBlockIndentWidth: 2
ObjCBreakBeforeNestedBlockParam: true
ObjCSpaceAfterProperty: false
ObjCSpaceBeforeProtocolList: false
PenaltyBreakAssignment: 2
PenaltyBreakBeforeFirstCallParameter: 1
PenaltyBreakComment: 300
PenaltyBreakFirstLessLess: 120
PenaltyBreakString: 1000
PenaltyBreakTemplateDeclaration: 10
PenaltyExcessCharacter: 1000000
PenaltyReturnTypeOnItsOwnLine: 200
PointerAlignment: Left
ReflowComments: true
SeparateDefinitionBlocks: Always
SortIncludes: true
SortUsingDeclarations: true
SpaceAfterCStyleCast: false
SpaceAfterLogicalNot: false
SpaceAfterTemplateKeyword: true
SpaceBeforeAssignmentOperators: true
SpaceBeforeCpp11BracedList: false
SpaceBeforeCtorInitializerColon: true
SpaceBeforeInheritanceColon: true
SpaceBeforeParens: ControlStatements
SpaceBeforeRangeBasedForLoopColon: true
SpaceInEmptyBlock: false
SpaceInEmptyParentheses: false
SpacesBeforeTrailingComments: 1
SpacesInAngles: false
SpacesInConditionalStatement: false
SpacesInContainerLiterals: true
SpacesInCStyleCastParentheses: false
SpacesInParentheses: false
SpacesInSquareBrackets: false
SpaceBeforeSquareBrackets: false
Standard: Latest
TabWidth: 8
UseCRLF: false
UseTab: Never
...
torchcodec-0.14.0/.flake8 0000664 0000000 0000000 00000000230 15207546147 0015072 0 ustar 00root root 0000000 0000000 [flake8]
max-line-length = 120
ignore = E203, E402, W503, W504, F821, E501, B, C4, EXE, E251, E202
per-file-ignores =
__init__.py: F401, F403, F405
torchcodec-0.14.0/.github/ 0000775 0000000 0000000 00000000000 15207546147 0015264 5 ustar 00root root 0000000 0000000 torchcodec-0.14.0/.github/ISSUE_TEMPLATE/ 0000775 0000000 0000000 00000000000 15207546147 0017447 5 ustar 00root root 0000000 0000000 torchcodec-0.14.0/.github/ISSUE_TEMPLATE/bug-report.yml 0000664 0000000 0000000 00000003570 15207546147 0022265 0 ustar 00root root 0000000 0000000 name: 🐛 Bug Report
description: Create a report to help us reproduce and fix the bug
body:
- type: markdown
attributes:
value: >
#### Your bug may have already been reported! Please check [the existing and past issues](https://github.com/pytorch/torchcodec/issues?q=sort%3Aupdated-desc+is%3Aissue).
- type: textarea
attributes:
label: 🐛 Describe the bug
description: |
What broke? What behavior did you see versus what did you expect? Please provide any relevant code, error messages and exception tracebacks.
A minimal code example will help us help you faster! The ideal code example is a small chunk of code that we can copy-paste to see the same error you see. For example:
```python
# All necessary imports at the beginning
import torch
import torchcodec
from torchcodec.decoders import VideoDecoder
# A succinct reproducing example trimmed down to the essential parts:
decoder = VideoDecoder("path/to/video.mp4") # Help! This fails!
# ...
```
If the code is long, put it in a public gist and link it in the issue: https://gist.github.com. Please also paste any error messages and full exception tracebacks in ```` ```triple quotes blocks``` ````.
validations:
required: true
- type: textarea
attributes:
label: Versions
description: |
We support a wide variety of platforms and versions, and many bugs are verison-dependent. Knowing your setup will help us help you faster! Please run the following and paste the output below.
```sh
wget https://raw.githubusercontent.com/pytorch/pytorch/main/torch/utils/collect_env.py
# For security purposes, please check the contents of collect_env.py before running it.
python collect_env.py
```
validations:
required: true
- type: markdown
attributes:
value: >
Thanks for contributing 🎉!
torchcodec-0.14.0/.github/ISSUE_TEMPLATE/config.yml 0000664 0000000 0000000 00000000033 15207546147 0021433 0 ustar 00root root 0000000 0000000 blank_issues_enabled: true
torchcodec-0.14.0/.github/ISSUE_TEMPLATE/documentation.yml 0000664 0000000 0000000 00000000714 15207546147 0023045 0 ustar 00root root 0000000 0000000 name: 📚 Documentation
description: Report an issue related to the TorchCodec documentation
body:
- type: textarea
attributes:
label: 📚 The doc issue
description: >
Is something confusing or wrong? Let us know! Please provide URLs to the content in https://pytorch.org/torchcodec/stable/index.html that you're referring to.
validations:
required: true
- type: markdown
attributes:
value: >
Thanks for contributing 🎉!
torchcodec-0.14.0/.github/ISSUE_TEMPLATE/feature-request.yml 0000664 0000000 0000000 00000001037 15207546147 0023314 0 ustar 00root root 0000000 0000000 name: 🚀 Feature request
description: Submit a proposal/request for a new TorchCodec feature
body:
- type: textarea
attributes:
label: 🚀 The feature
description: >
What new functionality do you want?
validations:
required: true
- type: textarea
attributes:
label: Motivation, pitch
description: >
Why do you want it? If this is related to another GitHub issue, please link that here.
validations:
required: false
- type: markdown
attributes:
value: >
Thanks for contributing 🎉!
torchcodec-0.14.0/.github/workflows/ 0000775 0000000 0000000 00000000000 15207546147 0017321 5 ustar 00root root 0000000 0000000 torchcodec-0.14.0/.github/workflows/build_ffmpeg.yaml 0000664 0000000 0000000 00000007731 15207546147 0022640 0 ustar 00root root 0000000 0000000 # Taken and adapted from torchaudio.
# Ref: https://github.com/pytorch/audio/blob/main/.github/workflows/ffmpeg.yml
# This job is not directly related to regular CI pipeline.
# It is intended to create FFmpeg binaries that we upload on S3,
# which then will be used during all the build process in CI or local.
#
# This job does not include the uploading part.
# Upload needs to be done manually, and it should be done only once
# per new major release of FFmepg.
name: Build non-GPL FFmpeg from source
on:
workflow_dispatch:
pull_request:
paths:
- packaging/build_ffmpeg.sh
- .github/workflows/build_ffmpeg.yaml # self reference
schedule:
- cron: '0 0 * * 0' # on sunday
defaults:
run:
shell: bash -l -eo pipefail {0}
jobs:
LGPL-Linux-x86_64:
strategy:
fail-fast: false
matrix:
ffmpeg-version: ["4.4.4", "5.1.4", "6.1.1", "7.0.1", "8.0"]
uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@release/2.12
permissions:
id-token: write
contents: read
with:
job-name: Build
upload-artifact: ffmpeg-lgpl-linux_x86_64-${{ matrix.ffmpeg-version }}
repository: meta-pytorch/torchcodec
script: |
export FFMPEG_VERSION="${{ matrix.ffmpeg-version }}"
export FFMPEG_ROOT="${PWD}/ffmpeg"
packaging/build_ffmpeg.sh
tar -cf ffmpeg.tar.gz ffmpeg/include ffmpeg/lib
artifact_dir="${RUNNER_ARTIFACT_DIR}/$(date +%Y-%m-%d)/linux_x86_64"
mkdir -p "${artifact_dir}"
mv ffmpeg.tar.gz "${artifact_dir}/${FFMPEG_VERSION}.tar.gz"
LGPL-Linux-aarch64:
strategy:
fail-fast: false
matrix:
ffmpeg-version: ["4.4.4", "5.1.4", "6.1.1", "7.0.1", "8.0"]
uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@release/2.12
permissions:
id-token: write
contents: read
with:
job-name: Build
upload-artifact: ffmpeg-lgpl-linux_aarch64-${{ matrix.ffmpeg-version }}
repository: meta-pytorch/torchcodec
runner: linux.arm64.2xlarge
docker-image: pytorch/manylinux2_28_aarch64-builder:cpu-aarch64
script: |
export FFMPEG_VERSION="${{ matrix.ffmpeg-version }}"
export FFMPEG_ROOT="${PWD}/ffmpeg"
packaging/build_ffmpeg.sh
tar -cf ffmpeg.tar.gz ffmpeg/include ffmpeg/lib
artifact_dir="${RUNNER_ARTIFACT_DIR}/$(date +%Y-%m-%d)/linux_aarch64"
mkdir -p "${artifact_dir}"
mv ffmpeg.tar.gz "${artifact_dir}/${FFMPEG_VERSION}.tar.gz"
LGPL-macOS:
strategy:
fail-fast: false
matrix:
ffmpeg-version: ["4.4.4", "5.1.4", "6.1.1", "7.0.1", "8.0"]
uses: pytorch/test-infra/.github/workflows/macos_job.yml@release/2.12
with:
job-name: Build
upload-artifact: ffmpeg-lgpl-macos-${{ matrix.ffmpeg-version }}
repository: meta-pytorch/torchcodec
runner: macos-14-xlarge
script: |
export FFMPEG_VERSION="${{ matrix.ffmpeg-version }}"
export FFMPEG_ROOT="${PWD}/ffmpeg"
packaging/build_ffmpeg.sh
tar -cf ffmpeg.tar.gz ffmpeg/include ffmpeg/lib
artifact_dir="${RUNNER_ARTIFACT_DIR}/$(date +%Y-%m-%d)/macos_$(uname -m)"
mkdir -p "${artifact_dir}"
mv ffmpeg.tar.gz "${artifact_dir}/${FFMPEG_VERSION}.tar.gz"
LGPL-Windows:
strategy:
fail-fast: false
matrix:
ffmpeg-version: ["4.4.4", "5.1.4", "6.1.1", "7.0.1", "8.0"]
uses: pytorch/test-infra/.github/workflows/windows_job.yml@release/2.12
with:
job-name: Build
upload-artifact: ffmpeg-lgpl-windows_x86_64-${{ matrix.ffmpeg-version }}
repository: meta-pytorch/torchcodec
script: |
export FFMPEG_VERSION="${{ matrix.ffmpeg-version }}"
export FFMPEG_ROOT="${PWD}/ffmpeg"
packaging/build_ffmpeg.bat
tar -cf ffmpeg.tar.gz ffmpeg/include ffmpeg/bin
artifact_dir="${RUNNER_ARTIFACT_DIR}/$(date +%Y-%m-%d)/windows_$(uname -m)"
mkdir -p "${artifact_dir}"
mv ffmpeg.tar.gz "${artifact_dir}/${FFMPEG_VERSION}.tar.gz"
torchcodec-0.14.0/.github/workflows/cpp_tests.yaml 0000664 0000000 0000000 00000005651 15207546147 0022220 0 ustar 00root root 0000000 0000000 name: CPP tests
on:
push:
branches: [ main ]
pull_request:
concurrency:
group: unit-test${{ github.workflow }}-${{ github.ref == 'refs/heads/main' && github.run_number || github.ref }}
cancel-in-progress: true
defaults:
run:
shell: bash -l -eo pipefail {0}
jobs:
Cpp-tests:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
ffmpeg-version-for-tests: ['7']
steps:
- name: Check out repo
uses: actions/checkout@v6
- name: Setup conda env
uses: conda-incubator/setup-miniconda@v3
with:
auto-update-conda: true
# Using miniforge instead of miniconda ensures that the default
# conda channel is conda-forge instead of main/default. This ensures
# ABI consistency between dependencies:
# https://conda-forge.org/docs/user/transitioning_from_defaults/
miniforge-version: latest
activate-environment: test
python-version: '3.12'
- name: Update pip
run: python -m pip install --upgrade pip
- name: Install torch dependencies
run: bash packaging/install_pytorch.sh cpu "torch"
- name: Install ffmpeg, pkg-config and pybind11
run: |
conda install "ffmpeg=${{ matrix.ffmpeg-version-for-tests }}" pkg-config pybind11 -c conda-forge
ffmpeg -version
- name: Build and run C++ tests
run: |
# Note: we're not setting BUILD_AGAINST_ALL_FFMPEG_FROM_S3 here, so
# we're building libtorchcodec against the installed FFmpeg version
# (from conda-forge) instead of building against our pre-built non-GPL
# FFmpeg libraries.
# The reason we need this is because the C++ tests decode x264 files.
# x264 support is not LGPL, os it is not supported by our
# pre-built non-GPL FFmpeg libraries. And if we were to build against
# those, this is also what the tests would be loading at run time,
# then failing when we try to decode x264.
# To remediate that, we build against the FFmpeg that we installed
# from conda-forge (which is able to decode x264), and that's also
# what gets loaded at run time.
# The Python tests are also decoding x264 files, and are built against
# our non-GPL FFmpeg. And yet they pass. This is because in Python
# we're able to distinguish between build-time (non-GPL FFmpeg) and
# run time (conda-forge FFmpeg).
build_tests_dir="${PWD}/build_tests"
mkdir $build_tests_dir
pushd $build_tests_dir
TORCH_PATH=$(python -c "import pathlib, torch; print(pathlib.Path(torch.__path__[0]))")
Torch_DIR="${TORCH_PATH}/share/cmake/Torch"
cmake .. -DTorch_DIR=$Torch_DIR -DCMAKE_BUILD_TYPE=Debug -DBUILD_TESTS=ON -DCMAKE_VERBOSE_MAKEFILE=ON
cmake --build .
ctest --output-on-failure
popd
torchcodec-0.14.0/.github/workflows/lint.yaml 0000664 0000000 0000000 00000004153 15207546147 0021156 0 ustar 00root root 0000000 0000000 name: Lint
on:
push:
branches: [ main ]
pull_request:
concurrency:
group: unit-test${{ github.workflow }}-${{ github.ref == 'refs/heads/main' && github.run_number || github.ref }}
cancel-in-progress: true
defaults:
run:
shell: bash -l -eo pipefail {0}
jobs:
pre-commit-checks:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
python-version: ['3.12']
steps:
- name: Check out repo
uses: actions/checkout@v6
- name: Setup conda env
uses: conda-incubator/setup-miniconda@v3
with:
auto-update-conda: true
miniforge-version: latest
activate-environment: test
python-version: ${{ matrix.python-version }}
- name: Update pip
run: python -m pip install --upgrade pip
- name: Install pre-commit
run: |
python -m pip install pre-commit
- name: Run pre-commit checks
run: |
pre-commit run --all-files
- name: Check to see what files pre-commit modified
run: |
git diff
mypy:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
python-version: ['3.12']
steps:
- name: Check out repo
uses: actions/checkout@v6
- name: Setup conda env
uses: conda-incubator/setup-miniconda@v3
with:
auto-update-conda: true
miniforge-version: latest
activate-environment: test
python-version: ${{ matrix.python-version }}
- name: Update pip
run: python -m pip install --upgrade pip
- name: Install dependencies and FFmpeg
run: |
bash packaging/install_pytorch.sh cpu "torch torchvision"
conda install "ffmpeg=7.0.1" pkg-config pybind11 -c conda-forge
ffmpeg -version
- name: Build and install torchcodec
run: |
python -m pip install -e ".[dev]" --no-build-isolation -vvv
- name: Install mypy
run: |
python -m pip install mypy
- name: Run mypy
run: |
mypy --install-types --non-interactive --config-file mypy.ini
torchcodec-0.14.0/.github/workflows/linux_aarch64_wheel.yaml 0000664 0000000 0000000 00000007075 15207546147 0024051 0 ustar 00root root 0000000 0000000 name: Linux CPU aarch64
on:
pull_request:
push:
branches:
- nightly
- main
- release/*
tags:
- v[0-9]+.[0-9]+.[0-9]+-rc[0-9]+
workflow_dispatch:
concurrency:
group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref_name }}-${{ github.ref_type == 'branch' && github.sha }}-${{ github.event_name == 'workflow_dispatch' }}
cancel-in-progress: true
permissions:
id-token: write
contents: write
defaults:
run:
shell: bash -l -eo pipefail {0}
jobs:
generate-matrix:
uses: pytorch/test-infra/.github/workflows/generate_binary_build_matrix.yml@release/2.12
with:
package-type: wheel
os: linux-aarch64
test-infra-repository: pytorch/test-infra
test-infra-ref: release/2.12
with-xpu: disable
with-rocm: disable
with-cuda: disable
build-python-only: "disable"
build:
needs: generate-matrix
strategy:
fail-fast: false
name: Build and Upload Linux aarch64 wheel
uses: pytorch/test-infra/.github/workflows/build_wheels_linux.yml@release/2.12
with:
repository: meta-pytorch/torchcodec
ref: ""
test-infra-repository: pytorch/test-infra
test-infra-ref: release/2.12
build-matrix: ${{ needs.generate-matrix.outputs.matrix }}
pre-script: packaging/pre_build_script.sh
post-script: packaging/post_build_script.sh
smoke-test-script: packaging/fake_smoke_test.py
package-name: torchcodec
trigger-event: ${{ github.event_name }}
architecture: aarch64
build-platform: "python-build-package"
build-command: "BUILD_AGAINST_ALL_FFMPEG_FROM_S3=1 python -m build --wheel -vvv --no-isolation"
install-and-test:
runs-on: linux.arm64.2xlarge
strategy:
fail-fast: false
matrix:
python-version: ['3.10', '3.11', '3.12', '3.13', '3.14']
ffmpeg-version-for-tests: ['4', '5', '6', '7', '8']
needs: build
steps:
- name: Check out repo
uses: actions/checkout@v6
- name: Remove src/ folder
run: bash packaging/remove_src.sh
- name: Setup conda env
uses: conda-incubator/setup-miniconda@v3
with:
auto-update-conda: true
# Using miniforge instead of miniconda ensures that the default
# conda channel is conda-forge instead of main/default. This ensures
# ABI consistency between dependencies:
# https://conda-forge.org/docs/user/transitioning_from_defaults/
miniforge-version: latest
activate-environment: test
python-version: ${{ matrix.python-version }}
- name: Update pip
run: python -m pip install --upgrade pip
- name: Install PyTorch
run: bash packaging/install_pytorch.sh cpu "torch torchvision"
- uses: actions/download-artifact@v4
with:
name: meta-pytorch_torchcodec__${{ matrix.python-version }}_cpu_aarch64
path: dist/
- name: Install torchcodec from the wheel
run: bash packaging/install_torchcodec_wheel.sh
- name: Install ffmpeg, post build
run: bash packaging/install_ffmpeg.sh ${{ matrix.ffmpeg-version-for-tests }}
- name: Install test dependencies
run: bash packaging/install_test_dependencies.sh
- name: Run Python smoke tests
run: |
pytest --override-ini="addopts=-v" test/smoke_test.py --tb=short
- name: Run Python tests
if: ${{ matrix.ffmpeg-version-for-tests == '7' && !startsWith(github.ref, 'refs/heads/release/') }}
run: |
pytest --override-ini="addopts=-v" test
torchcodec-0.14.0/.github/workflows/linux_cuda_aarch64_wheel.yaml 0000664 0000000 0000000 00000010220 15207546147 0025027 0 ustar 00root root 0000000 0000000 name: Linux CUDA aarch64
on:
pull_request:
push:
branches:
- nightly
- main
- release/*
tags:
- v[0-9]+.[0-9]+.[0-9]+-rc[0-9]+
workflow_dispatch:
concurrency:
group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref_name }}-${{ github.ref_type == 'branch' && github.sha }}-${{ github.event_name == 'workflow_dispatch' }}
cancel-in-progress: true
permissions:
id-token: write
contents: write
defaults:
run:
shell: bash -l -eo pipefail {0}
jobs:
generate-matrix:
uses: pytorch/test-infra/.github/workflows/generate_binary_build_matrix.yml@release/2.12
with:
package-type: wheel
os: linux-aarch64
test-infra-repository: pytorch/test-infra
test-infra-ref: release/2.12
with-cpu: disable
with-xpu: disable
with-rocm: disable
with-cuda: enable
build-python-only: "disable"
build:
needs: generate-matrix
strategy:
fail-fast: false
name: Build and Upload wheel
uses: pytorch/test-infra/.github/workflows/build_wheels_linux.yml@release/2.12
with:
repository: meta-pytorch/torchcodec
ref: ""
test-infra-repository: pytorch/test-infra
test-infra-ref: release/2.12
build-matrix: ${{ needs.generate-matrix.outputs.matrix }}
pre-script: packaging/pre_build_script.sh
post-script: packaging/post_build_script.sh
smoke-test-script: packaging/fake_smoke_test.py
package-name: torchcodec
trigger-event: ${{ github.event_name }}
architecture: aarch64
build-platform: "python-build-package"
build-command: "BUILD_AGAINST_ALL_FFMPEG_FROM_S3=1 ENABLE_CUDA=1 python -m build --wheel -vvv --no-isolation"
install-and-test:
runs-on: linux.arm64.2xlarge
strategy:
fail-fast: false
matrix:
python-version: ['3.10', '3.11', '3.12', '3.13', '3.14']
cuda-version: ['12.6', '13.0', '13.2']
ffmpeg-version-for-tests: ['4', '5', '6', '7', '8']
container:
image: "pytorch/manylinuxaarch64-builder:cuda${{ matrix.cuda-version }}"
needs: build
steps:
- name: Setup env vars
run: |
cuda_version_without_periods=$(echo "${{ matrix.cuda-version }}" | sed 's/\.//g')
echo cuda_version_without_periods=${cuda_version_without_periods} >> $GITHUB_ENV
- name: Check out repo
uses: actions/checkout@v6
- name: Remove src/ folder
run: bash packaging/remove_src.sh
- name: Setup conda env
uses: conda-incubator/setup-miniconda@v3
with:
auto-update-conda: true
# Using miniforge instead of miniconda ensures that the default
# conda channel is conda-forge instead of main/default. This ensures
# ABI consistency between dependencies:
# https://conda-forge.org/docs/user/transitioning_from_defaults/
miniforge-version: latest
activate-environment: test
python-version: ${{ matrix.python-version }}
- name: Update pip
run: python -m pip install --upgrade pip
- name: Install PyTorch
run: ${CONDA_RUN} bash packaging/install_pytorch.sh cu${{ env.cuda_version_without_periods }} "torch torchvision"
- uses: actions/download-artifact@v4
with:
name: meta-pytorch_torchcodec__${{ matrix.python-version }}_cu${{ env.cuda_version_without_periods }}_aarch64
path: dist/
- name: Install torchcodec from the wheel
run: bash packaging/install_torchcodec_wheel.sh
- name: Install ffmpeg, post build
run: |
bash packaging/install_ffmpeg.sh ${{ matrix.ffmpeg-version-for-tests }}
echo LD_LIBRARY_PATH=$CONDA_PREFIX/lib:/usr/local/cuda/lib64/:${LD_LIBRARY_PATH} >> $GITHUB_ENV
- name: Install test dependencies
run: bash packaging/install_test_dependencies.sh
- name: Run Python smoke tests
run: |
pytest --override-ini="addopts=-v" test/smoke_test.py --tb=short
- name: Run Python tests
if: ${{ matrix.ffmpeg-version-for-tests == '8' && !startsWith(github.ref, 'refs/heads/release/') }}
run: |
pytest --override-ini="addopts=-v" test
torchcodec-0.14.0/.github/workflows/linux_cuda_wheel.yaml 0000664 0000000 0000000 00000033210 15207546147 0023523 0 ustar 00root root 0000000 0000000 name: Linux CUDA x86
on:
pull_request:
push:
branches:
- nightly
- main
- release/*
tags:
- v[0-9]+.[0-9]+.[0-9]+-rc[0-9]+
- v[0-9]+.[0-9]+.[0-9]+
workflow_dispatch:
concurrency:
group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref_name }}-${{ github.ref_type == 'branch' && github.sha }}-${{ github.event_name == 'workflow_dispatch' }}
cancel-in-progress: true
permissions:
id-token: write
contents: write
defaults:
run:
shell: bash -l -eo pipefail {0}
jobs:
generate-matrix:
uses: pytorch/test-infra/.github/workflows/generate_binary_build_matrix.yml@release/2.12
with:
package-type: wheel
os: linux
test-infra-repository: pytorch/test-infra
test-infra-ref: release/2.12
with-cpu: disable
with-xpu: disable
with-rocm: disable
with-cuda: enable
build-python-only: "disable"
build:
needs: generate-matrix
strategy:
fail-fast: false
name: Build and Upload wheel
uses: pytorch/test-infra/.github/workflows/build_wheels_linux.yml@release/2.12
with:
repository: meta-pytorch/torchcodec
ref: ""
test-infra-repository: pytorch/test-infra
test-infra-ref: release/2.12
build-matrix: ${{ needs.generate-matrix.outputs.matrix }}
pre-script: packaging/pre_build_script.sh
post-script: packaging/post_build_script.sh
smoke-test-script: packaging/fake_smoke_test.py
package-name: torchcodec
trigger-event: ${{ github.event_name }}
build-platform: "python-build-package"
build-command: "BUILD_AGAINST_ALL_FFMPEG_FROM_S3=1 ENABLE_CUDA=1 python -m build --wheel -vvv --no-isolation"
install-and-test:
runs-on: linux.g5.4xlarge.nvidia.gpu
strategy:
fail-fast: false
matrix:
# 3.10 corresponds to the minimum python version for which we build
# the wheel unless the label cliflow/binaries/all is present in the
# PR.
# For the actual release we should add that label and change this to
# include more python versions.
python-version: ['3.10', '3.11', '3.12', '3.13', '3.14']
# We test against 12.6 and 13.0 to avoid having too big of a CI matrix,
# but for releases we should add 12.8.
cuda-version: ['12.6', '13.0', '13.2']
# TODO: put back ffmpeg 5 https://github.com/pytorch/torchcodec/issues/325
ffmpeg-version-for-tests: ['4', '5', '6', '7', '8']
container:
image: "pytorch/manylinux2_28-builder:cuda${{ matrix.cuda-version }}"
options: "--gpus all -e NVIDIA_DRIVER_CAPABILITIES=video,compute,utility"
needs: build
steps:
- name: Setup env vars
run: |
cuda_version_without_periods=$(echo "${{ matrix.cuda-version }}" | sed 's/\.//g')
echo cuda_version_without_periods=${cuda_version_without_periods} >> $GITHUB_ENV
python_version_without_periods=$(echo "${{ matrix.python-version }}" | sed 's/\.//g')
echo python_version_without_periods=${python_version_without_periods} >> $GITHUB_ENV
- name: Check out repo
uses: actions/checkout@v6
- name: Remove src/ folder
run: bash packaging/remove_src.sh
- name: Setup miniconda using test-infra
uses: pytorch/test-infra/.github/actions/setup-miniconda@release/2.12
with:
python-version: ${{ matrix.python-version }}
# We install conda packages at the start because otherwise conda may have conflicts with dependencies.
# Note: xorg-libxau was addded to fix a problem with ffmpeg 4. We should consider removing it.
default-packages: "nvidia::cuda-nvrtc=${{ matrix.cuda-version }} nvidia::cuda-toolkit=${{ matrix.cuda-version }} nvidia::cuda-cudart=${{ matrix.cuda-version }} nvidia::cuda-driver-dev=${{ matrix.cuda-version }} conda-forge::ffmpeg=${{ matrix.ffmpeg-version-for-tests }} conda-forge::xorg-libxau"
- name: Check env, set LD_LIBRARY_PATH
run: |
${CONDA_RUN} env
${CONDA_RUN} conda info
${CONDA_RUN} nvidia-smi
${CONDA_RUN} conda list
echo LD_LIBRARY_PATH=$CONDA_PREFIX/lib:/usr/local/cuda/lib64/:${LD_LIBRARY_PATH} >> $GITHUB_ENV
- name: Assert ffmpeg exists
run: |
${CONDA_RUN} ffmpeg -buildconf
- name: Update pip
run: ${CONDA_RUN} python -m pip install --upgrade pip
- name: Install PyTorch
run: |
${CONDA_RUN} bash packaging/install_pytorch.sh cu${{ env.cuda_version_without_periods }} "torch torchvision"
${CONDA_RUN} python -c 'import torch; print(f"{torch.__version__}"); print(f"{torch.__file__}"); print(f"{torch.cuda.is_available()=}")'
- uses: actions/download-artifact@v4
with:
name: meta-pytorch_torchcodec__${{ matrix.python-version }}_cu${{ env.cuda_version_without_periods }}_x86_64
path: dist/
- name: Install torchcodec from the wheel
run: ${CONDA_RUN} bash packaging/install_torchcodec_wheel.sh "*cu${{ env.cuda_version_without_periods }}-cp${{ env.python_version_without_periods }}*.whl"
- name: Install test dependencies
run: ${CONDA_RUN} bash packaging/install_test_dependencies.sh
- name: Run Python tests
run: |
${CONDA_RUN} FAIL_WITHOUT_CUDA=1 pytest --override-ini="addopts=-v" test --tb=short
- name: Run Python benchmark
run: |
${CONDA_RUN} time python benchmarks/decoders/gpu_benchmark.py --devices=cuda:0,cpu --resize_devices=none
install-and-test-on-cpu-only-machine:
# This job tests that CUDA wheels work fine on CPU-only machines. Note that
# we still install a CUDA-enabled version of torch, and that's by design.
# Essentially, what we want to make sure is that
# `pip install torch torchcodec` works on CPU-only machines, and this
# command should install CUDA-enabled versions of both torch and torchcodec.
# It's critical that this job runs on a CPU-only machine.
runs-on: ubuntu-latest
needs: build
env:
PYTHON_VERSION: '3.10'
CUDA_VERSION: '12.6'
FFMPEG_VERSION: '7'
steps:
- name: Setup env vars
run: |
cuda_version_without_periods=$(echo "${{ env.CUDA_VERSION }}" | sed 's/\.//g')
echo cuda_version_without_periods=${cuda_version_without_periods} >> $GITHUB_ENV
python_version_without_periods=$(echo "${{ env.PYTHON_VERSION }}" | sed 's/\.//g')
echo python_version_without_periods=${python_version_without_periods} >> $GITHUB_ENV
- name: Check out repo
uses: actions/checkout@v6
- name: Remove src/ folder
run: bash packaging/remove_src.sh
- name: Setup conda env
uses: conda-incubator/setup-miniconda@v3
with:
auto-update-conda: true
miniforge-version: latest
activate-environment: test
python-version: ${{ env.PYTHON_VERSION }}
- name: Update pip
run: python -m pip install --upgrade pip
- name: Install PyTorch
run: bash packaging/install_pytorch.sh cu${{ env.cuda_version_without_periods }} "torch torchvision"
- uses: actions/download-artifact@v4
with:
name: meta-pytorch_torchcodec__${{ env.PYTHON_VERSION }}_cu${{ env.cuda_version_without_periods }}_x86_64
path: dist/
- name: Install torchcodec from the wheel
run: bash packaging/install_torchcodec_wheel.sh "*cu${{ env.cuda_version_without_periods }}-cp${{ env.python_version_without_periods }}*.whl"
- name: Install ffmpeg
run: bash packaging/install_ffmpeg.sh ${{ env.FFMPEG_VERSION }}
- name: Install test dependencies
run: bash packaging/install_test_dependencies.sh
- name: Assert CUDA is not available
run: |
python -c "import torch; assert not torch.cuda.is_available()"
- name: Run Python tests
run: |
pytest --override-ini="addopts=-v" test --tb=short
build-docs:
runs-on: linux.g5.4xlarge.nvidia.gpu
env:
PYTHON_VERSION: '3.10'
CUDA_VERSION: '12.6'
FFMPEG_VERSION: '7'
container:
image: "pytorch/manylinux2_28-builder:cuda12.6" # must be same as env!!
options: "--gpus all -e NVIDIA_DRIVER_CAPABILITIES=video,compute,utility"
needs: build
steps:
- name: Setup env vars
run: |
cuda_version_without_periods=$(echo "${{ env.CUDA_VERSION }}" | sed 's/\.//g')
echo cuda_version_without_periods=${cuda_version_without_periods} >> $GITHUB_ENV
python_version_without_periods=$(echo "${{ env.PYTHON_VERSION }}" | sed 's/\.//g')
echo python_version_without_periods=${python_version_without_periods} >> $GITHUB_ENV
- name: Check out repo
uses: actions/checkout@v6
- name: Remove src/ folder
run: bash packaging/remove_src.sh
- name: Setup miniconda using test-infra
uses: pytorch/test-infra/.github/actions/setup-miniconda@release/2.12
with:
python-version: ${{ env.PYTHON_VERSION }}
# We install conda packages at the start because otherwise conda may have conflicts with dependencies.
default-packages: "nvidia::cuda-nvrtc=${{ env.CUDA_VERSION }} nvidia::cuda-toolkit=${{ env.CUDA_VERSION }} nvidia::cuda-cudart=${{ env.CUDA_VERSION }} nvidia::cuda-driver-dev=${{ env.CUDA_VERSION }} conda-forge::ffmpeg=${{ env.FFMPEG_VERSION }}"
- name: Check env, set LD_LIBRARY_PATH
run: |
${CONDA_RUN} env
${CONDA_RUN} conda info
${CONDA_RUN} nvidia-smi
${CONDA_RUN} conda list
echo LD_LIBRARY_PATH=$CONDA_PREFIX/lib:/usr/local/cuda/lib64/:${LD_LIBRARY_PATH} >> $GITHUB_ENV
- name: Assert ffmpeg exists
run: |
${CONDA_RUN} ffmpeg -buildconf
- name: Update pip
run: ${CONDA_RUN} python -m pip install --upgrade pip
- name: Install PyTorch
run: |
${CONDA_RUN} bash packaging/install_pytorch.sh cu${{ env.cuda_version_without_periods }} "torch torchvision"
${CONDA_RUN} python -c 'import torch; print(f"{torch.__version__}"); print(f"{torch.__file__}"); print(f"{torch.cuda.is_available()=}")'
- uses: actions/download-artifact@v4
with:
name: meta-pytorch_torchcodec__${{ env.PYTHON_VERSION }}_cu${{ env.cuda_version_without_periods }}_x86_64
path: dist/
- name: Install torchcodec from the wheel
run: ${CONDA_RUN} bash packaging/install_torchcodec_wheel.sh "*cu${{ env.cuda_version_without_periods }}-cp${{ env.python_version_without_periods }}*.whl"
- name: Install doc dependencies
run: |
cd docs
${CONDA_RUN} python -m pip install -r requirements.txt
- name: Build docs
run: |
cd docs
${CONDA_RUN} make html
- uses: actions/upload-artifact@v4
with:
name: Built-Docs
path: docs/build/html/
doc-preview:
runs-on: [self-hosted, linux.2xlarge]
needs: build-docs
if: github.repository == 'meta-pytorch/torchcodec' && github.event_name == 'pull_request'
steps:
- uses: actions/download-artifact@v4
with:
name: Built-Docs
path: docs
# Update HTML to add the no-index tag so that search engines do not index these ephemeral docs
- name: Add no-index tag
run: |
find docs -name "*.html" -print0 | xargs -0 sed -i '//a \ \ ';
- name: Upload docs preview
uses: seemethere/upload-artifact-s3@v5
with:
retention-days: 14
s3-bucket: doc-previews
if-no-files-found: error
path: docs
s3-prefix: meta-pytorch/torchcodec/${{ github.event.pull_request.number }}
upload-docs:
# This job uploads built docs:
# - to the `main` folder of the gh-pages branch (https://meta-pytorch.org/torchcodec/main) on every commit to the `main` branch
# - to the (e.g.) `0.10` folder in the gh-pages branch whenever a corresponding tag is pushed, like `v0.10.0` (https://meta-pytorch.org/torchcodec/0.10).
needs: build-docs
if: github.repository == 'meta-pytorch/torchcodec' && github.event_name == 'push' &&
((github.ref_type == 'branch' && github.ref_name == 'main') || github.ref_type == 'tag')
runs-on: ubuntu-latest
steps:
- name: Check out gh-pages branch
uses: actions/checkout@v6
with:
ref: gh-pages
- uses: actions/download-artifact@v4
with:
name: Built-Docs
path: docs-artifact/
- name: Update docs and push
run: |
set -euo pipefail
REF_TYPE=${{ github.ref_type }}
REF_NAME=${{ github.ref_name }}
if [[ "${REF_TYPE}" == branch ]]; then
TARGET_FOLDER="${REF_NAME}"
elif [[ "${REF_TYPE}" == tag ]]; then
case "${REF_NAME}" in
*-rc*)
echo "Aborting upload since this is an RC tag: ${REF_NAME}"
exit 0
;;
*)
# Strip the leading "v" as well as the trailing patch version. For example:
# 'v0.10.2' -> '0.10'
TARGET_FOLDER=$(echo "${REF_NAME}" | sed 's/v\([0-9]\+\)\.\([0-9]\+\)\.[0-9]\+/\1.\2/')
;;
esac
fi
echo "Target Folder: ${TARGET_FOLDER}"
mkdir -p "${TARGET_FOLDER}"
rm -rf "${TARGET_FOLDER}"/*
cp -r docs-artifact/* "${TARGET_FOLDER}/"
git add "${TARGET_FOLDER}"
git config user.name 'pytorchbot'
git config user.email 'soumith+bot@pytorch.org'
git commit -m "auto-generating sphinx docs for ${TARGET_FOLDER}" || echo "No changes to commit"
git push
torchcodec-0.14.0/.github/workflows/linux_wheel.yaml 0000664 0000000 0000000 00000012153 15207546147 0022532 0 ustar 00root root 0000000 0000000 name: Linux CPU x86
on:
pull_request:
push:
branches:
- nightly
- main
- release/*
tags:
- v[0-9]+.[0-9]+.[0-9]+-rc[0-9]+
workflow_dispatch:
concurrency:
group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref_name }}-${{ github.ref_type == 'branch' && github.sha }}-${{ github.event_name == 'workflow_dispatch' }}
cancel-in-progress: true
permissions:
id-token: write
contents: write
defaults:
run:
shell: bash -l -eo pipefail {0}
jobs:
generate-matrix:
uses: pytorch/test-infra/.github/workflows/generate_binary_build_matrix.yml@release/2.12
with:
package-type: wheel
os: linux
test-infra-repository: pytorch/test-infra
test-infra-ref: release/2.12
with-xpu: disable
with-rocm: disable
with-cuda: disable
build-python-only: "disable"
build:
needs: generate-matrix
strategy:
fail-fast: false
name: Build and Upload wheel
uses: pytorch/test-infra/.github/workflows/build_wheels_linux.yml@release/2.12
with:
repository: meta-pytorch/torchcodec
ref: ""
test-infra-repository: pytorch/test-infra
test-infra-ref: release/2.12
build-matrix: ${{ needs.generate-matrix.outputs.matrix }}
pre-script: packaging/pre_build_script.sh
post-script: packaging/post_build_script.sh
smoke-test-script: packaging/fake_smoke_test.py
package-name: torchcodec
trigger-event: ${{ github.event_name }}
build-platform: "python-build-package"
build-command: "BUILD_AGAINST_ALL_FFMPEG_FROM_S3=1 python -m build --wheel -vvv --no-isolation"
also-upload-to-gpu-arch-versions: "xpu"
install-and-test:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
python-version: ['3.10', '3.11', '3.12', '3.13', '3.14']
ffmpeg-version-for-tests: ['4', '5', '6', '7', '8']
needs: build
steps:
- name: Check out repo
uses: actions/checkout@v6
- name: Remove src/ folder
run: bash packaging/remove_src.sh
- name: Setup conda env
uses: conda-incubator/setup-miniconda@v3
with:
auto-update-conda: true
# Using miniforge instead of miniconda ensures that the default
# conda channel is conda-forge instead of main/default. This ensures
# ABI consistency between dependencies:
# https://conda-forge.org/docs/user/transitioning_from_defaults/
miniforge-version: latest
activate-environment: test
python-version: ${{ matrix.python-version }}
- name: Update pip
run: python -m pip install --upgrade pip
- name: Install PyTorch
run: bash packaging/install_pytorch.sh cpu "torch torchvision"
- uses: actions/download-artifact@v4
with:
name: meta-pytorch_torchcodec__${{ matrix.python-version }}_cpu_x86_64
path: dist/
- name: Install torchcodec from the wheel
run: bash packaging/install_torchcodec_wheel.sh
- name: Install ffmpeg, post build
run: bash packaging/install_ffmpeg.sh ${{ matrix.ffmpeg-version-for-tests }}
- name: Install test dependencies
run: bash packaging/install_test_dependencies.sh
- name: Run Python tests
run: |
pytest --override-ini="addopts=-v" test
install-and-test-third-party-interface:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
python-version: ['3.10', '3.11', '3.12', '3.13', '3.14']
ffmpeg-version-for-tests: ['4', '5', '6', '7', '8']
needs: build
steps:
- name: Check out repo
uses: actions/checkout@v3
- name: Remove src/ folder
run: bash packaging/remove_src.sh
- name: Setup conda env
uses: conda-incubator/setup-miniconda@v3
with:
auto-update-conda: true
# Using miniforge instead of miniconda ensures that the default
# conda channel is conda-forge instead of main/default. This ensures
# ABI consistency between dependencies:
# https://conda-forge.org/docs/user/transitioning_from_defaults/
miniforge-version: latest
activate-environment: test
python-version: ${{ matrix.python-version }}
- name: Update pip
run: python -m pip install --upgrade pip
- name: Install PyTorch
run: bash packaging/install_pytorch.sh cpu "torch torchvision"
- uses: actions/download-artifact@v4
with:
name: meta-pytorch_torchcodec__${{ matrix.python-version }}_cpu_x86_64
path: dist/
- name: Install torchcodec from the wheel
run: bash packaging/install_torchcodec_wheel.sh
- name: Install ffmpeg, post build
run: bash packaging/install_ffmpeg.sh ${{ matrix.ffmpeg-version-for-tests }}
- name: Install pkg-config
run: |
conda install pkg-config -c conda-forge
- name: Install test dependencies
run: bash packaging/install_test_dependencies.sh
- name: Run Python tests
run: |
pytest --override-ini="addopts=-v" test/third-party-interface
torchcodec-0.14.0/.github/workflows/macos_wheel.yaml 0000664 0000000 0000000 00000007217 15207546147 0022502 0 ustar 00root root 0000000 0000000 name: MacOS
on:
pull_request:
push:
branches:
- nightly
- main
- release/*
tags:
- v[0-9]+.[0-9]+.[0-9]+-rc[0-9]+
workflow_dispatch:
concurrency:
group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref_name }}-${{ github.ref_type == 'branch' && github.sha }}-${{ github.event_name == 'workflow_dispatch' }}
cancel-in-progress: true
permissions:
id-token: write
contents: write
defaults:
run:
shell: bash -l -eo pipefail {0}
jobs:
generate-matrix:
uses: pytorch/test-infra/.github/workflows/generate_binary_build_matrix.yml@release/2.12
with:
package-type: wheel
os: macos-arm64
test-infra-repository: pytorch/test-infra
test-infra-ref: release/2.12
with-xpu: disable
with-rocm: disable
with-cuda: disable
build-python-only: "disable"
python-versions: '["3.10", "3.11", "3.12", "3.13", "3.14"]'
build:
needs: generate-matrix
strategy:
fail-fast: false
name: Build and Upload wheel
uses: pytorch/test-infra/.github/workflows/build_wheels_macos.yml@release/2.12
with:
repository: meta-pytorch/torchcodec
ref: ""
test-infra-repository: pytorch/test-infra
test-infra-ref: release/2.12
build-matrix: ${{ needs.generate-matrix.outputs.matrix }}
pre-script: packaging/pre_build_script.sh
post-script: packaging/post_build_script.sh
smoke-test-script: packaging/fake_smoke_test.py
runner-type: macos-14
setup-miniconda: true
package-name: torchcodec
trigger-event: ${{ github.event_name }}
build-platform: "python-build-package"
build-command: "BUILD_AGAINST_ALL_FFMPEG_FROM_S3=1 python -m build --wheel -vvv --no-isolation"
install-and-test:
runs-on: macos-14-xlarge
strategy:
fail-fast: false
matrix:
python-version: ['3.10', '3.11', '3.12', '3.13', '3.14']
ffmpeg-version-for-tests: ['4', '5', '6', '7', '8']
needs: build
steps:
- name: Check out torchcodec repo
uses: actions/checkout@v6
- name: Remove src/ folder
run: bash packaging/remove_src.sh
- name: Setup conda env
uses: conda-incubator/setup-miniconda@v3
with:
auto-update-conda: true
# Using miniforge instead of miniconda ensures that the default
# conda channel is conda-forge instead of main/default. This ensures
# ABI consistency between dependencies:
# https://conda-forge.org/docs/user/transitioning_from_defaults/
miniforge-version: latest
activate-environment: test
python-version: ${{ matrix.python-version }}
- name: Update pip
run: python -m pip install --upgrade pip
- name: Install PyTorch
run: bash packaging/install_pytorch.sh cpu "torch torchvision"
- name: Download wheel
uses: actions/download-artifact@v4
with:
name: meta-pytorch_torchcodec__${{ matrix.python-version }}_cpu_
path: dist/
- name: Install torchcodec from the wheel
run: bash packaging/install_torchcodec_wheel.sh
- name: Install ffmpeg
run: bash packaging/install_ffmpeg.sh ${{ matrix.ffmpeg-version-for-tests }}
- name: Install test dependencies
run: bash packaging/install_test_dependencies.sh
- name: Run Python smoke tests
run: |
pytest --override-ini="addopts=-v" test/smoke_test.py --tb=short
- name: Run Python tests
if: ${{ matrix.ffmpeg-version-for-tests == '8' && !startsWith(github.ref, 'refs/heads/release/') }}
run: |
pytest --override-ini="addopts=-v" test
torchcodec-0.14.0/.github/workflows/reference_resources.yaml 0000664 0000000 0000000 00000005647 15207546147 0024251 0 ustar 00root root 0000000 0000000 name: Reference resource generation tests
on:
workflow_dispatch:
pull_request:
paths:
- test/generate_reference_resources.py
- .github/workflows/reference_resources.yaml # self reference
schedule:
- cron: '0 0 * * 0' # on sunday
defaults:
run:
shell: bash -l -eo pipefail {0}
jobs:
generate-matrix:
uses: pytorch/test-infra/.github/workflows/generate_binary_build_matrix.yml@release/2.12
with:
package-type: wheel
os: linux
test-infra-repository: pytorch/test-infra
test-infra-ref: release/2.12
with-xpu: disable
with-rocm: disable
with-cuda: disable
build-python-only: "disable"
build:
needs: generate-matrix
strategy:
fail-fast: false
name: Build and Upload Linux wheel
uses: pytorch/test-infra/.github/workflows/build_wheels_linux.yml@release/2.12
with:
repository: meta-pytorch/torchcodec
ref: ""
test-infra-repository: pytorch/test-infra
test-infra-ref: release/2.12
build-matrix: ${{ needs.generate-matrix.outputs.matrix }}
pre-script: packaging/pre_build_script.sh
post-script: packaging/post_build_script.sh
smoke-test-script: packaging/fake_smoke_test.py
package-name: torchcodec
trigger-event: ${{ github.event_name }}
build-platform: "python-build-package"
build-command: "BUILD_AGAINST_ALL_FFMPEG_FROM_S3=1 python -m build --wheel -vvv --no-isolation"
test-reference-resource-generation:
needs: build
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
python-version: ['3.10']
# Traditionally we generate the resources locally on FFmpeg 4 or 6.
# The exact version shouln't matter as long as the unit tests pass
# across all version for a given generated resource.
ffmpeg-version-for-tests: ['6']
steps:
- name: Check out repo
uses: actions/checkout@v3
- name: Setup conda env
uses: conda-incubator/setup-miniconda@v3
with:
auto-update-conda: true
miniforge-version: latest
activate-environment: test
python-version: ${{ matrix.python-version }}
- name: Update pip
run: python -m pip install --upgrade pip
- name: Install PyTorch
run: bash packaging/install_pytorch.sh cpu "torch"
- uses: actions/download-artifact@v4
with:
name: meta-pytorch_torchcodec__${{ matrix.python-version }}_cpu_x86_64
path: dist/
- name: Install torchcodec from the wheel
run: bash packaging/install_torchcodec_wheel.sh
- name: Install ffmpeg, post build
run: bash packaging/install_ffmpeg.sh ${{ matrix.ffmpeg-version-for-tests }}
- name: Install test dependencies
run: bash packaging/install_test_dependencies.sh
- name: Run generation reference resources
run: |
python -m test.generate_reference_resources
torchcodec-0.14.0/.github/workflows/windows_cuda_wheel.yaml 0000664 0000000 0000000 00000017342 15207546147 0024066 0 ustar 00root root 0000000 0000000 name: Windows CUDA
on:
pull_request:
push:
branches:
- nightly
- main
- release/*
tags:
- v[0-9]+.[0-9]+.[0-9]+-rc[0-9]+
workflow_dispatch:
concurrency:
group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref_name }}-${{ github.ref_type == 'branch' && github.sha }}-${{ github.event_name == 'workflow_dispatch' }}
cancel-in-progress: true
permissions:
id-token: write
contents: write
defaults:
run:
shell: bash -l -eo pipefail {0}
jobs:
generate-matrix:
uses: pytorch/test-infra/.github/workflows/generate_binary_build_matrix.yml@release/2.12
with:
package-type: wheel
os: windows
test-infra-repository: pytorch/test-infra
test-infra-ref: release/2.12
with-cpu: disable
with-xpu: disable
with-rocm: disable
with-cuda: enable
build-python-only: "disable"
python-versions: '["3.10", "3.11", "3.12", "3.13", "3.14"]'
build:
needs: generate-matrix
strategy:
fail-fast: false
name: Build and Upload wheel
uses: pytorch/test-infra/.github/workflows/build_wheels_windows.yml@release/2.12
with:
repository: meta-pytorch/torchcodec
ref: ""
test-infra-repository: pytorch/test-infra
test-infra-ref: release/2.12
build-matrix: ${{ needs.generate-matrix.outputs.matrix }}
pre-script: packaging/pre_build_script.sh
# post-script: packaging/post_build_script.sh TODO: consider enabling post-build checks for Windows
env-script: packaging/vc_env_helper.bat
smoke-test-script: packaging/fake_smoke_test.py
package-name: torchcodec
trigger-event: ${{ github.event_name }}
build-platform: "python-build-package"
# The BUILD_AGAINST_ALL_FFMPEG_FROM_S3 and ENABLE_CUDA vars, needed to
# build the wheel, are set in vc_env_helper.bat. Couldn't find a way to
# set them from here.
build-command: "python -m build --wheel -vvv --no-isolation"
install-and-test:
runs-on: windows.g5.4xlarge.nvidia.gpu
strategy:
fail-fast: false
matrix:
# 3.10 corresponds to the minimum python version for which we build
# the wheel unless the label cliflow/binaries/all is present in the
# PR.
# For the actual release we should add that label and change this to
# include more python versions.
python-version: ['3.10', '3.11', '3.12', '3.13', '3.14']
cuda-version: ['12.6', '13.0', '13.2']
# TODO: FFmpeg 5 on Windows segfaults in avcodec_open2() when passing
# bad parameters.
# See https://github.com/pytorch/torchcodec/pull/806
ffmpeg-version-for-tests: ['4', '5', '6', '7', '8']
needs: build
steps:
- name: Setup env vars
run: |
cuda_version_without_periods=$(echo "${{ matrix.cuda-version }}" | sed 's/\.//g')
echo cuda_version_without_periods=${cuda_version_without_periods} >> $GITHUB_ENV
python_version_without_periods=$(echo "${{ matrix.python-version }}" | sed 's/\.//g')
echo python_version_without_periods=${python_version_without_periods} >> $GITHUB_ENV
- name: Check out repo
uses: actions/checkout@v6
- name: Update NVIDIA driver
uses: pytorch/test-infra/.github/actions/setup-nvidia-windows@release/2.12
- name: Remove src/ folder
run: bash packaging/remove_src.sh
- name: Setup conda env
uses: conda-incubator/setup-miniconda@v3
with:
auto-update-conda: true
miniforge-version: latest
activate-environment: test
python-version: ${{ matrix.python-version }}
- name: Install CUDA and FFmpeg conda packages
run: |
conda install -y \
"nvidia::cuda-nvrtc=${{ matrix.cuda-version }}" \
"nvidia::cuda-cudart=${{ matrix.cuda-version }}" \
"conda-forge::ffmpeg=${{ matrix.ffmpeg-version-for-tests }}"
- name: Check env
run: |
env
conda info
conda list
- name: Assert ffmpeg exists
run: |
ffmpeg -buildconf
- name: Check FFmpeg CUDA support
run: |
ffmpeg -decoders | grep -i nvidia
- name: Update pip
run: python -m pip install --upgrade pip
- name: Install PyTorch
run: |
bash packaging/install_pytorch.sh cu${{ env.cuda_version_without_periods }} "torch torchvision"
python -c 'import torch; print(f"{torch.__version__}"); print(f"{torch.__file__}"); print(f"{torch.cuda.is_available()=}")'
- uses: actions/download-artifact@v4
with:
name: meta-pytorch_torchcodec__${{ matrix.python-version }}_cu${{ env.cuda_version_without_periods }}_x64
path: dist/
- name: Install torchcodec from the wheel
run: bash packaging/install_torchcodec_wheel.sh "*cu${{ env.cuda_version_without_periods }}-cp${{ env.python_version_without_periods }}*.whl"
- name: Install test dependencies
run: bash packaging/install_test_dependencies.sh
- name: Run Python tests
run: |
FAIL_WITHOUT_CUDA=1 pytest --override-ini="addopts=-v" test/smoke_test.py --tb=short
- name: Run Python benchmark
run: |
time python benchmarks/decoders/gpu_benchmark.py --devices=cuda:0,cpu --resize_devices=none
install-and-test-on-cpu-only-machine:
# This job tests that CUDA wheels work fine on CPU-only machines. Note that
# we still install a CUDA-enabled version of torch, and that's by design.
# Essentially, what we want to make sure is that
# `pip install torch torchcodec` works on CPU-only machines, and this
# command should install CUDA-enabled versions of both torch and torchcodec.
# It's critical that this job runs on a CPU-only machine.
runs-on: windows-latest
needs: build
env:
PYTHON_VERSION: '3.10'
CUDA_VERSION: '12.6'
FFMPEG_VERSION: '7'
steps:
- name: Setup env vars
run: |
cuda_version_without_periods=$(echo "${{ env.CUDA_VERSION }}" | sed 's/\.//g')
echo cuda_version_without_periods=${cuda_version_without_periods} >> $GITHUB_ENV
python_version_without_periods=$(echo "${{ env.PYTHON_VERSION }}" | sed 's/\.//g')
echo python_version_without_periods=${python_version_without_periods} >> $GITHUB_ENV
- name: Check out repo
uses: actions/checkout@v6
- name: Remove src/ folder
run: bash packaging/remove_src.sh
- name: Setup conda env
uses: conda-incubator/setup-miniconda@v3
with:
auto-update-conda: true
miniforge-version: latest
activate-environment: test
python-version: ${{ env.PYTHON_VERSION }}
- name: Update pip
run: python -m pip install --upgrade pip
- name: Install PyTorch
run: bash packaging/install_pytorch.sh cu${{ env.cuda_version_without_periods }} "torch torchvision"
- uses: actions/download-artifact@v4
with:
name: meta-pytorch_torchcodec__${{ env.PYTHON_VERSION }}_cu${{ env.cuda_version_without_periods }}_x64
path: dist/
- name: Install torchcodec from the wheel
run: bash packaging/install_torchcodec_wheel.sh "*cu${{ env.cuda_version_without_periods }}-cp${{ env.python_version_without_periods }}*.whl"
- name: Install ffmpeg
run: bash -l packaging/install_ffmpeg.sh ${{ env.FFMPEG_VERSION }}
- name: Install test dependencies
run: bash packaging/install_test_dependencies.sh
- name: Assert CUDA is not available
run: |
python -c "import torch; assert not torch.cuda.is_available()"
- name: Run Python tests
run: |
pytest --override-ini="addopts=-v" test/smoke_test.py --tb=short
torchcodec-0.14.0/.github/workflows/windows_wheel.yaml 0000664 0000000 0000000 00000007762 15207546147 0023077 0 ustar 00root root 0000000 0000000 name: Windows CPU
on:
pull_request:
push:
branches:
- nightly
- main
- release/*
tags:
- v[0-9]+.[0-9]+.[0-9]+-rc[0-9]+
workflow_dispatch:
concurrency:
group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref_name }}-${{ github.ref_type == 'branch' && github.sha }}-${{ github.event_name == 'workflow_dispatch' }}
cancel-in-progress: true
permissions:
id-token: write
contents: write
defaults:
run:
shell: bash -l -eo pipefail {0}
jobs:
generate-matrix:
uses: pytorch/test-infra/.github/workflows/generate_binary_build_matrix.yml@release/2.12
with:
package-type: wheel
os: windows
test-infra-repository: pytorch/test-infra
test-infra-ref: release/2.12
with-xpu: disable
with-rocm: disable
with-cuda: disable
build-python-only: "disable"
python-versions: '["3.10", "3.11", "3.12", "3.13", "3.14"]'
build:
needs: generate-matrix
strategy:
fail-fast: false
name: Build and Upload wheel
uses: pytorch/test-infra/.github/workflows/build_wheels_windows.yml@release/2.12
with:
repository: meta-pytorch/torchcodec
ref: ""
test-infra-repository: pytorch/test-infra
test-infra-ref: release/2.12
build-matrix: ${{ needs.generate-matrix.outputs.matrix }}
pre-script: packaging/pre_build_script.sh
# post-script: packaging/post_build_script.sh TODO: consider enabling post-build checks for Windows
env-script: packaging/vc_env_helper.bat
smoke-test-script: packaging/fake_smoke_test.py
package-name: torchcodec
trigger-event: ${{ github.event_name }}
build-platform: "python-build-package"
# The BUILD_AGAINST_ALL_FFMPEG_FROM_S3 var, needed to build the wheel, is
# set in vc_env_helper.bat Couldn't find a way to set it from here.
build-command: "python -m build --wheel -vvv --no-isolation"
install-and-test:
runs-on: windows-latest
strategy:
fail-fast: false
matrix:
python-version: ['3.10', '3.11', '3.12', '3.13', '3.14']
# TODO: FFmpeg 5 on Windows segfaults in avcodec_open2() when passing
# bad parameters.
# See https://github.com/pytorch/torchcodec/pull/806
ffmpeg-version-for-tests: ['4', '5', '6', '7', '8']
needs: build
steps:
- name: Check out repo
uses: actions/checkout@v6
- name: Remove src/ folder
run: bash packaging/remove_src.sh
- name: Setup conda env
uses: conda-incubator/setup-miniconda@v3
with:
auto-update-conda: true
# Using miniforge instead of miniconda ensures that the default
# conda channel is conda-forge instead of main/default. This ensures
# ABI consistency between dependencies:
# https://conda-forge.org/docs/user/transitioning_from_defaults/
miniforge-version: latest
activate-environment: test
python-version: ${{ matrix.python-version }}
- name: Update pip
run: python -m pip install --upgrade pip
- name: Install PyTorch
run: bash packaging/install_pytorch.sh cpu "torch torchvision"
- uses: actions/download-artifact@v4
with:
name: meta-pytorch_torchcodec__${{ matrix.python-version }}_cpu_x64
path: dist/
- name: Install torchcodec from the wheel
run: bash packaging/install_torchcodec_wheel.sh
- name: Install ffmpeg, post build
# need -l for conda to be exposed
run: bash -l packaging/install_ffmpeg.sh ${{ matrix.ffmpeg-version-for-tests }}
- name: Install test dependencies
run: bash packaging/install_test_dependencies.sh
- name: Run Python smoke tests
run: |
pytest --override-ini="addopts=-v" test/smoke_test.py --tb=short
- name: Run Python tests
if: ${{ matrix.ffmpeg-version-for-tests == '8' && !startsWith(github.ref, 'refs/heads/release/') }}
run: |
pytest --override-ini="addopts=-v" test
torchcodec-0.14.0/.gitignore 0000664 0000000 0000000 00000001134 15207546147 0015713 0 ustar 00root root 0000000 0000000 build/
dist/
src/TorchCodec.egg-info/
*/**/__pycache__
*/__pycache__
*/*.pyc
*/**/*.pyc
*/**/**/*.pyc
*/**/*~
*~
frame180.* # output from smoke test
src/torchcodec/version.py
docs/build
# sphinx-gallery
docs/source/generated_examples/
docs/source/gen_modules/
docs/source/generated/
docs/source/models/generated/
docs/source/sg_execution_times.rst
# pytorch-sphinx-theme gets installed here
docs/src
.coverage
htmlcov
.*.swp
*.so*
*.dylib*
*/*.so*
*/*.dylib*
*.swp
*.swo
gen.yml
.mypy_cache
.vscode/
.idea/
*.orig
*-checkpoint.ipynb
*.venv
## Xcode User settings
xcuserdata/
# direnv
.direnv
.envrc
torchcodec-0.14.0/.pre-commit-config.yaml 0000664 0000000 0000000 00000002074 15207546147 0020210 0 ustar 00root root 0000000 0000000 repos:
- repo: https://github.com/pre-commit/pre-commit-hooks
rev: v4.6.0
hooks:
- id: check-docstring-first
- id: trailing-whitespace
- id: check-toml
- id: check-yaml
args:
- --allow-multiple-documents
- id: mixed-line-ending
args: [--fix=lf]
- id: end-of-file-fixer
- id: check-added-large-files
args: ['--maxkb=1000']
- repo: https://github.com/asottile/pyupgrade
rev: v3.21.2
hooks:
- id: pyupgrade
args: [--py310-plus]
files: ^(test|src)/
exclude: ^examples/
- repo: https://github.com/omnilib/ufmt
rev: v2.6.0
hooks:
- id: ufmt
additional_dependencies:
- black == 24.4.2
- usort == 1.0.5
- repo: https://github.com/PyCQA/flake8
rev: 7.1.0
hooks:
- id: flake8
args: [--config=.flake8]
- repo: https://github.com/pre-commit/mirrors-clang-format
rev: v18.1.3
hooks:
- id: clang-format
name: clang-format
files: \.(cpp|hpp|c|h)$
types: [file]
torchcodec-0.14.0/CMakeLists.txt 0000664 0000000 0000000 00000000514 15207546147 0016464 0 ustar 00root root 0000000 0000000 cmake_minimum_required(VERSION 3.18)
project(TorchCodec)
# Define LINUX platform variable globally
if (UNIX AND NOT APPLE)
set(LINUX TRUE)
else()
set(LINUX FALSE)
endif()
add_subdirectory(src/torchcodec/_core)
option(BUILD_TESTS "Build tests" OFF)
if(BUILD_TESTS)
enable_testing()
add_subdirectory(test)
endif()
torchcodec-0.14.0/CODE_OF_CONDUCT.md 0000664 0000000 0000000 00000006417 15207546147 0016533 0 ustar 00root root 0000000 0000000 # Code of Conduct
## Our Pledge
In the interest of fostering an open and welcoming environment, we as
contributors and maintainers pledge to make participation in our project and
our community a harassment-free experience for everyone, regardless of age, body
size, disability, ethnicity, sex characteristics, gender identity and expression,
level of experience, education, socio-economic status, nationality, personal
appearance, race, religion, or sexual identity and orientation.
## Our Standards
Examples of behavior that contributes to creating a positive environment
include:
* Using welcoming and inclusive language
* Being respectful of differing viewpoints and experiences
* Gracefully accepting constructive criticism
* Focusing on what is best for the community
* Showing empathy towards other community members
Examples of unacceptable behavior by participants include:
* The use of sexualized language or imagery and unwelcome sexual attention or
advances
* Trolling, insulting/derogatory comments, and personal or political attacks
* Public or private harassment
* Publishing others' private information, such as a physical or electronic
address, without explicit permission
* Other conduct which could reasonably be considered inappropriate in a
professional setting
## Our Responsibilities
Project maintainers are responsible for clarifying the standards of acceptable
behavior and are expected to take appropriate and fair corrective action in
response to any instances of unacceptable behavior.
Project maintainers have the right and responsibility to remove, edit, or
reject comments, commits, code, wiki edits, issues, and other contributions
that are not aligned to this Code of Conduct, or to ban temporarily or
permanently any contributor for other behaviors that they deem inappropriate,
threatening, offensive, or harmful.
## Scope
This Code of Conduct applies within all project spaces, and it also applies when
an individual is representing the project or its community in public spaces.
Examples of representing a project or community include using an official
project e-mail address, posting via an official social media account, or acting
as an appointed representative at an online or offline event. Representation of
a project may be further defined and clarified by project maintainers.
## Enforcement
Instances of abusive, harassing, or otherwise unacceptable behavior may be
reported by contacting the project team at . All
complaints will be reviewed and investigated and will result in a response that
is deemed necessary and appropriate to the circumstances. The project team is
obligated to maintain confidentiality with regard to the reporter of an incident.
Further details of specific enforcement policies may be posted separately.
Project maintainers who do not follow or enforce the Code of Conduct in good
faith may face temporary or permanent repercussions as determined by other
members of the project's leadership.
## Attribution
This Code of Conduct is adapted from the [Contributor Covenant][homepage], version 1.4,
available at https://www.contributor-covenant.org/version/1/4/code-of-conduct.html
[homepage]: https://www.contributor-covenant.org
For answers to common questions about this code of conduct, see
https://www.contributor-covenant.org/faq
torchcodec-0.14.0/CONTRIBUTING.md 0000664 0000000 0000000 00000010621 15207546147 0016155 0 ustar 00root root 0000000 0000000 # Contributing to TorchCodec
You can contribute to this project by writing code, fixing issues or simply by
using the library and reporting your feedback.
Below are instructions to build TorchCodec from source, as well as the usual
contribution guidelines (code formatting, testing, etc). To submit a PR, please
follow the [official GitHub
guidelines](https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/proposing-changes-to-your-work-with-pull-requests/creating-a-pull-request-from-a-fork).
## Building TorchCodec from source
### Installing dependencies
The instructions below assume you are using a conda environment, but the steps
are easily adaptable to other kind of virtual environments. To build, run and
test locally you will need the following dependencies:
- A C++ compiler+linker. This is typically available on a baseline Linux
installation already.
- cmake
- pkg-config
- pybind11
- FFmpeg
- PyTorch nightly
Start by installing the **nightly** build of PyTorch following the
[official instructions](https://pytorch.org/get-started/locally/).
Then, the easiest way to install the rest of the dependencies is to run:
```bash
conda install cmake pkg-config pybind11 "ffmpeg" -c conda-forge
```
### Clone and build
To clone and install the repo, run:
```bash
git clone git@github.com:pytorch/torchcodec.git
# Or, using https instead of ssh: git clone https://github.com/pytorch/torchcodec.git
cd torchcodec
# Optional, but recommended: define a persistent build directory which speeds-up
# subsequent builds.
export TORCHCODEC_CMAKE_BUILD_DIR="${PWD}/build"
pip install -e ".[dev]" --no-build-isolation -vv
# Or, for cuda support: ENABLE_CUDA=1 pip install -e ".[dev]" --no-build-isolation -vv
```
### Running unit tests
To run python tests run:
```bash
pytest
```
Some tests are marked as 'slow' and aren't run by default. You can use `pytest
-m slow` to run those, or `pytest -m ""` to run all tests, slow or not.
To run the C++ tests run:
```bash
mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Debug -DBUILD_TESTS=1 -DCMAKE_PREFIX_PATH=$(python3 -c 'import torch;print(torch.utils.cmake_prefix_path)') ..
cmake --build . -- VERBOSE=1
ctest --rerun-failed --output-on-failure
```
### Code formatting and type checking
We use `pre-commit` to enforce code formatting and `mypy` for type checking.
Install both with
```bash
pip install pre-commit mypy
```
To run pre-commit hooks before each commit, run `pre-commit install`. You may
prefer to run these checks manually, in which case you can just use `pre-commit
run --all-files`.
For `mypy` we recommend the following command:
```bash
mypy --install-types --non-interactive --config-file mypy.ini
```
### Building the docs
First install from source, then install the doc dependencies:
```bash
cd docs
pip install -r requirements.txt
```
Then, still from within the `docs` directory:
```bash
make html
```
The built docs will be in `build/html`. Open in your browser to view them.
To avoid building the examples (which execute python code and can take time) you
can use `make html-noplot`. To build a subset of specific examples instead of
all of them, you can use a regex like
`EXAMPLES_PATTERN="plot_the_best_example*" make html`.
Run `make clean` from time to time if you encounter issues.
### Serving docs locally (if building from a GPU env)
If you're developing locally, you can just open the generated `index.html`file
in your browser.
If instead you're using a remote machine, you can use a combination of a simple
python HTTP server and port forwarding to serve the docs locally. This allows
you to iterate on the documentation much more quickly than relying on
PR previews.
To do so, after following the above doc build steps, run the following from
the `docs/build/html` folder:
```
python -m http.server 8000 # or any free port
```
This will open up a simple HTTP server serving the files in the build directory.
If this is done on a remote machine, you can set up port forwarding from your
local machine to access the server, for example:
```
ssh -L 9000:localhost:8000 $REMOTE_DEV_HOST
```
Now, you can navigate to `localhost:9000` on your local machine to view the
rendered documentation.
## License
By contributing to TorchCodec, you agree that your contributions will be
licensed under the LICENSE file in the root directory of this source tree.
Contributors are also required to
[sign our Contributor License Agreement](https://code.facebook.com/cla).
torchcodec-0.14.0/LICENSE 0000664 0000000 0000000 00000002702 15207546147 0014732 0 ustar 00root root 0000000 0000000 BSD 3-Clause License
Copyright 2024 Meta
Redistribution and use in source and binary forms, with or without modification,
are permitted provided that the following conditions are met:
1. Redistributions of source code must retain the above copyright notice,this list
of conditions and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright notice, this
list of conditions and the following disclaimer in the documentation
and/or other materials provided with the distribution.
3. Neither the name of the copyright holder nor the names of its contributors may
be used to endorse or promote products derived from this software without specific
prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” AND ANY
EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT
SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
DAMAGE.
torchcodec-0.14.0/MANIFEST.in 0000664 0000000 0000000 00000000255 15207546147 0015464 0 ustar 00root root 0000000 0000000 include README.md
include LICENSE
include CMakeLists.txt
recursive-include src *
recursive-exclude * __pycache__
recursive-exclude src *.py[co]
recursive-exclude src *.so
torchcodec-0.14.0/README.md 0000664 0000000 0000000 00000017004 15207546147 0015205 0 ustar 00root root 0000000 0000000 [**Installation**](#installing-torchcodec) | [**Simple Example**](#using-torchcodec) | [**Detailed Example**](https://meta-pytorch.org/torchcodec/stable/generated_examples/) | [**Documentation**](https://meta-pytorch.org/torchcodec) | [**Contributing**](CONTRIBUTING.md) | [**License**](#license)
# TorchCodec
TorchCodec is a Python library for decoding video and audio data into PyTorch
tensors, on CPU and CUDA GPU. It also supports video and audio encoding on CPU!
It aims to be fast, easy to use, and well integrated
into the PyTorch ecosystem. If you want to use PyTorch to train ML models on
videos and audio, or run inference, TorchCodec is how you turn these into data.
We achieve these capabilities through:
* Pythonic APIs that mirror Python and PyTorch conventions.
* Relying on [FFmpeg](https://www.ffmpeg.org/) to do the decoding and encoding.
TorchCodec uses the version of FFmpeg you already have installed. FFmpeg is a
mature library with broad coverage available on most systems. It is, however,
not easy to use. TorchCodec abstracts FFmpeg's complexity to ensure it is used
correctly and efficiently.
* Returning data as PyTorch tensors, ready to be fed into PyTorch transforms
or used directly to train models.
## Usage Examples
Below are some examples of what you can do with TorchCodec. For more detailed
examples and more use-cases, [check out our
documentation](https://meta-pytorch.org/torchcodec/stable/generated_examples/)!
#### Video Decoding
```python
from torchcodec.decoders import VideoDecoder
device = "cpu" # or e.g. "cuda" !
decoder = VideoDecoder("path/to/video.mp4", device=device)
decoder.metadata
# VideoStreamMetadata:
# num_frames: 250
# duration_seconds: 10.0
# bit_rate: 31315.0
# codec: h264
# average_fps: 25.0
# ... (truncated output)
# Simple Indexing API
decoder[0] # uint8 tensor of shape [C, H, W]
decoder[0 : -1 : 20] # uint8 stacked tensor of shape [N, C, H, W]
# Indexing, with PTS and duration info:
decoder.get_frames_at(indices=[2, 100])
# FrameBatch:
# data (shape): torch.Size([2, 3, 270, 480])
# pts_seconds: tensor([0.0667, 3.3367], dtype=torch.float64)
# duration_seconds: tensor([0.0334, 0.0334], dtype=torch.float64)
# Time-based indexing with PTS and duration info
decoder.get_frames_played_at(seconds=[0.5, 10.4])
# FrameBatch:
# data (shape): torch.Size([2, 3, 270, 480])
# pts_seconds: tensor([ 0.4671, 10.3770], dtype=torch.float64)
# duration_seconds: tensor([0.0334, 0.0334], dtype=torch.float64)
```
You can use the following snippet to generate a video with FFmpeg and try out
the `VideoDecoder`:
```bash
ffmpeg -f lavfi -i testsrc2=size=640x400:duration=10:rate=25 /tmp/output_video.mp4
```
#### Encoding
```python
from torchcodec.encoders import Encoder
encoder = Encoder()
video_stream = encoder.add_video(
height=height, width=width, frame_rate=frame_rate,
)
audio_stream = encoder.add_audio(
sample_rate=sample_rate, num_channels=num_channels,
)
with encoder.open_file("output.mp4"):
video_stream.add_frames(frames_batch_0)
audio_stream.add_samples(samples_batch_0)
video_stream.add_frames(frames_batch_1)
audio_stream.add_samples(samples_batch_1)
# ...
```
## Installing TorchCodec
1. Install FFmpeg, if it's not already installed. TorchCodec supports all major
FFmpeg versions in [4, 8]. Linux distributions usually come with FFmpeg
pre-installed. You'll need FFmpeg that comes with separate shared libraries.
This is especially relevant for Windows users: these are usually called the
"shared" releases.
If FFmpeg is not already installed, or you need a more recent version, an
easy way to install it is to use `conda`:
```bash
conda install "ffmpeg"
# or
conda install "ffmpeg" -c conda-forge
```
2. Install PyTorch and TorchCodec:
```bash
pip install torch torchcodec
```
That's it! On Linux x86 and aarch64, this will install CUDA-enabled wheels by
default (matching the default behavior of `pip install torch`). These wheels
should *still* work even if you do not have a GPU on your machine. On macOS
and Windows this will install CPU-only wheels. CPU wheels are available for
Linux (x86_64 and aarch64), macOS, and Windows.
For other versions of PyTorch, refer to the compatibility table below.
### CUDA support
CUDA-enabled wheels are installed by default on Linux. For Windows, you'll need
to pass `--index-url` as described below.
Make sure you have a GPU with NVDEC hardware that can decode the format you
want. Refer to Nvidia's GPU support matrix
[here](https://developer.nvidia.com/video-encode-and-decode-gpu-support-matrix-new).
You will need the `libnvrtc` CUDA library, which is usually part of the CUDA
Toolkit.
To select a specific CUDA Toolkit version, use `--index-url`. Make sure to
install the corresponding PyTorch version as well (refer to the
[official instructions](https://pytorch.org/get-started/locally/)):
```bash
# This corresponds to CUDA Toolkit version 13.0.
pip install torch torchcodec --index-url=https://download.pytorch.org/whl/cu130
```
Make sure your FFmpeg has NVDEC support:
```bash
ffmpeg -decoders | grep -i nvidia
# This should show a line like this:
# V..... h264_cuvid Nvidia CUVID H264 decoder (codec h264)
```
To check that FFmpeg libraries work with NVDEC correctly you can decode a
generated test video:
```bash
ffmpeg -hwaccel cuda -hwaccel_output_format cuda -f lavfi -i testsrc2=duration=1 -f null -
```
### CPU-only installation
To install CPU-only wheels explicitly (e.g. on Linux where CUDA wheels are the
default):
```bash
pip install torchcodec --index-url=https://download.pytorch.org/whl/cpu
```
### Compatibility
The following table indicates the compatibility between versions of
`torchcodec`, `torch` and Python.
| `torchcodec` | `torch` | Python |
| ------------------ | ------------------ | ------------------- |
| `main` / `nightly` | `main` / `nightly` | `>=3.10`, `<=3.14` |
| `0.13` | `>=2.11` | `>=3.10`, `<=3.14` |
| `0.12` | `>=2.11` | `>=3.10`, `<=3.14` |
| `0.11` | `2.11` | `>=3.10`, `<=3.14` |
| `0.10` | `2.10` | `>=3.10`, `<=3.14` |
older versions
| `torchcodec` | `torch` | Python |
| ------------------ | ------------------ | ------------------- |
| `0.9` | `2.9` | `>=3.10`, `<=3.14` |
| `0.8` | `2.9` | `>=3.10`, `<=3.13` |
| `0.7` | `2.8` | `>=3.9`, `<=3.13` |
| `0.6` | `2.8` | `>=3.9`, `<=3.13` |
| `0.5` | `2.7` | `>=3.9`, `<=3.13` |
| `0.4` | `2.7` | `>=3.9`, `<=3.13` |
| `0.3` | `2.7` | `>=3.9`, `<=3.13` |
| `0.2` | `2.6` | `>=3.9`, `<=3.13` |
| `0.1` | `2.5` | `>=3.9`, `<=3.12` |
| `0.0.3` | `2.4` | `>=3.8`, `<=3.12` |
## Contributing
We welcome contributions to TorchCodec! Please see our [contributing
guide](CONTRIBUTING.md) for more details.
## License
TorchCodec is released under the [BSD 3 license](./LICENSE).
However, TorchCodec may be used with code not written by Meta which may be
distributed under different licenses.
For example, if you build TorchCodec with ENABLE_CUDA=1 or use the CUDA-enabled
release of torchcodec, please review CUDA's license here:
[Nvidia licenses](https://docs.nvidia.com/cuda/eula/index.html).
torchcodec-0.14.0/benchmarks/ 0000775 0000000 0000000 00000000000 15207546147 0016041 5 ustar 00root root 0000000 0000000 torchcodec-0.14.0/benchmarks/decoders/ 0000775 0000000 0000000 00000000000 15207546147 0017631 5 ustar 00root root 0000000 0000000 torchcodec-0.14.0/benchmarks/decoders/benchmark_audio_decoders.py 0000664 0000000 0000000 00000005363 15207546147 0025175 0 ustar 00root root 0000000 0000000 from argparse import ArgumentParser
from datetime import timedelta
from pathlib import Path
from time import perf_counter_ns
import torch
import torchaudio
from torch import Tensor
from torchaudio.io import StreamReader
from torchcodec.decoders._audio_decoder import AudioDecoder
DEFAULT_NUM_EXP = 30
def bench(f, *args, num_exp=DEFAULT_NUM_EXP, warmup=1, **kwargs) -> Tensor:
for _ in range(warmup):
f(*args, **kwargs)
times = []
for _ in range(num_exp):
start = perf_counter_ns()
f(*args, **kwargs)
end = perf_counter_ns()
times.append(end - start)
return torch.tensor(times).float()
def report_stats(times: Tensor, unit: str = "ms", prefix: str = "") -> float:
mul = {
"ns": 1,
"µs": 1e-3,
"ms": 1e-6,
"s": 1e-9,
}[unit]
times = times * mul
std = times.std().item()
med = times.median().item()
mean = times.mean().item()
min = times.min().item()
max = times.max().item()
print(
f"{prefix:<40} {med = :.2f}, {mean = :.2f} +- {std:.2f}, {min = :.2f}, {max = :.2f} - in {unit}"
)
def decode_with_torchcodec(path: Path) -> None:
AudioDecoder(path).get_all_samples()
def decode_with_torchaudio_StreamReader(path: Path) -> None:
reader = StreamReader(path)
reader.add_audio_stream(frames_per_chunk=1024)
for _ in reader.stream():
pass
def decode_with_torchaudio_load(path: Path, backend: str) -> None:
torchaudio.load(str(path), backend=backend)
parser = ArgumentParser()
parser.add_argument("--path", type=str, help="path to file", required=True)
parser.add_argument(
"--num-exp",
type=int,
default=DEFAULT_NUM_EXP,
help="number of runs to average over",
)
args = parser.parse_args()
path = Path(args.path)
metadata = AudioDecoder(path).metadata
duration = str(timedelta(seconds=metadata.duration_seconds_from_header)).split(".")[0]
print(
f"Benchmarking {path.name}, duration: {duration}, codec: {metadata.codec}, format: {metadata.sample_format}, averaging over {args.num_exp} runs:"
)
for decode_f, kwargs, prefix in (
(decode_with_torchcodec, {}, "torchcodec.AudioDecoder"),
(
decode_with_torchaudio_load,
{"backend": "ffmpeg"},
"torchaudio.load(backend='ffmpeg')",
),
(decode_with_torchaudio_load, {"backend": "sox"}, "torchaudio.load(backend='sox')"),
(
decode_with_torchaudio_load,
{"backend": "soundfile"},
"torchaudio.load(backend='soundfile')",
),
(decode_with_torchaudio_StreamReader, {}, "torchaudio.StreamReader"),
):
try:
times = bench(decode_f, path, **kwargs, num_exp=args.num_exp)
report_stats(times, prefix=prefix)
except RuntimeError:
print(f"{prefix:<40} Not supported")
torchcodec-0.14.0/benchmarks/decoders/benchmark_decoders.py 0000664 0000000 0000000 00000012744 15207546147 0024015 0 ustar 00root root 0000000 0000000 # Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the BSD-style license found in the
# LICENSE file in the root directory of this source tree.
import argparse
import importlib.resources
import json
import os
import platform
from pathlib import Path
import torch
from benchmark_decoders_library import (
decoder_registry,
plot_data,
run_benchmarks,
verify_outputs,
)
def in_fbcode() -> bool:
return "FB_PAR_RUNTIME_FILES" in os.environ
def get_test_resource_path(filename: str) -> str:
if in_fbcode():
resource = importlib.resources.files(__package__).joinpath(filename)
with importlib.resources.as_file(resource) as path:
return os.fspath(path)
return str(Path(__file__).parent / f"../../test/resources/{filename}")
def parse_options_code(options_code: str) -> dict[str, str]:
options = {}
for item in options_code.split("+"):
if item.strip() == "":
continue
k, v = item.split("=")
options[k] = v
return options
def main() -> None:
"""Benchmarks the performance of a few video decoders"""
parser = argparse.ArgumentParser()
parser.add_argument(
"--bm-video-creation",
help="Benchmark large video creation memory",
default=False,
action=argparse.BooleanOptionalAction,
)
parser.add_argument(
"--verbose",
help="Show verbose output",
default=False,
action=argparse.BooleanOptionalAction,
)
parser.add_argument(
"--min-run-seconds",
help="Benchmark minimum run time, in seconds, to wait per datapoint",
type=float,
default=2.0,
)
parser.add_argument(
"--video-paths",
help="Comma-separated paths to videos that you want to benchmark.",
type=str,
default=get_test_resource_path("nasa_13013.mp4"),
)
parser.add_argument(
"--decoders",
help=(
"Comma-separated list of decoders to benchmark. "
"Choices are: " + ", ".join(decoder_registry.keys()) + ". "
"To specify options, append a ':' and then value pairs seperated by a '+'. "
"For example, torchcodec_core:num_threads=1+color_conversion_library=filtergraph."
),
type=str,
default=(
"decord,decord_batch,"
"torchvision,"
"torchaudio,"
"torchcodec_core,torchcodec_core:num_threads=1,torchcodec_core_batch,torchcodec_core_nonbatch,"
"torchcodec_public"
),
)
parser.add_argument(
"--video-dir",
help="Directory where video files reside. We will run benchmarks on all .mp4 files in this directory.",
type=str,
default="",
)
parser.add_argument(
"--plot-path",
help="Path where the generated plot is stored, if non-empty",
type=str,
default="benchmarks.png",
)
parser.add_argument(
"--verify-outputs",
help="Verify that the outputs of the decoders are the same",
default=False,
action=argparse.BooleanOptionalAction,
)
parser.add_argument(
"--output-json",
help="Output the results to a JSON file",
type=str,
default="",
)
args = parser.parse_args()
specified_decoders = set(args.decoders.split(","))
# These are the PTS values we want to extract from the small video.
num_uniform_samples = 10
decoders_to_run = {}
for decoder in specified_decoders:
if ":" in decoder:
decoder, _, options_code = decoder.partition(":")
assert decoder in decoder_registry, f"Unknown decoder: {decoder}"
display = decoder_registry[decoder].display_name + ":" + options_code
options = parse_options_code(options_code)
else:
assert decoder in decoder_registry, f"Unknown decoder: {decoder}"
display = decoder_registry[decoder].display_name
options = decoder_registry[decoder].default_options
kind = decoder_registry[decoder].kind
decoders_to_run[display] = kind(**options)
video_paths = args.video_paths.split(",")
if args.video_dir:
video_paths = []
for entry in os.scandir(args.video_dir):
if entry.is_file() and entry.name.endswith(".mp4"):
video_paths.append(entry.path)
if args.verify_outputs:
verify_outputs(decoders_to_run, video_paths, num_uniform_samples)
else:
results = run_benchmarks(
decoders_to_run,
video_paths,
num_uniform_samples,
num_sequential_frames_from_start=[1, 10, 100],
min_runtime_seconds=args.min_run_seconds,
benchmark_video_creation=args.bm_video_creation,
)
if args.output_json:
with open(args.output_json, "w") as f:
json.dump(results, f, indent=2)
data = {
"experiments": results,
"system_metadata": {
"cpu_count": os.cpu_count(),
"system": platform.system(),
"machine": platform.machine(),
"python_version": str(platform.python_version()),
"cuda": (
torch.cuda.get_device_properties(0).name
if torch.cuda.is_available()
else "not available"
),
},
}
plot_data(data, args.plot_path)
if __name__ == "__main__":
main()
torchcodec-0.14.0/benchmarks/decoders/benchmark_decoders_library.py 0000664 0000000 0000000 00000110633 15207546147 0025535 0 ustar 00root root 0000000 0000000 import abc
import subprocess
import typing
import urllib.request
from concurrent.futures import ThreadPoolExecutor, wait
from dataclasses import dataclass, field
from itertools import product
from pathlib import Path
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import torch
import torch.utils.benchmark as benchmark
from torchcodec._core import get_frames_at_indices, get_frames_by_pts, get_next_frame
from torchcodec._core.ops import _add_video_stream, create_from_file, seek_to_pts
from torchcodec._frame import FrameBatch
from torchcodec.decoders import VideoDecoder, VideoStreamMetadata
torch._dynamo.config.cache_size_limit = 100
torch._dynamo.config.capture_dynamic_output_shape_ops = True
class AbstractDecoder:
def __init__(self):
pass
@abc.abstractmethod
def decode_frames(self, video_file, pts_list):
pass
def decode_frames_description(self, num_frames: int, kind: str) -> str:
return f"decode {num_frames} {kind} frames"
@abc.abstractmethod
def decode_first_n_frames(self, video_file, n):
pass
def decode_first_n_frames_description(self, n) -> str:
return f"first {n} frames"
def init_decode_and_resize(self):
pass
@abc.abstractmethod
def decode_and_resize(self, video_file, pts_list, height, width, device):
pass
def decode_and_resize_description(
self, num_frames: int, height: int, width: int
) -> str:
return f"decode {num_frames} -> {height}x{width}"
class DecordAccurate(AbstractDecoder):
def __init__(self):
import decord # noqa: F401
self.decord = decord
self.decord.bridge.set_bridge("torch")
def decode_frames(self, video_file, pts_list):
decord_vr = self.decord.VideoReader(video_file, ctx=self.decord.cpu())
frames = []
fps = decord_vr.get_avg_fps()
for pts in pts_list:
decord_vr.seek_accurate(int(pts * fps))
frame = decord_vr.next()
frames.append(frame)
return frames
def decode_first_n_frames(self, video_file, n):
decord_vr = self.decord.VideoReader(video_file, ctx=self.decord.cpu())
frames = []
for _ in range(n):
frame = decord_vr.next()
frames.append(frame)
return frames
class DecordAccurateBatch(AbstractDecoder):
def __init__(self):
import decord # noqa: F401
self.decord = decord
self.decord.bridge.set_bridge("torch")
def decode_frames(self, video_file, pts_list):
decord_vr = self.decord.VideoReader(video_file, ctx=self.decord.cpu())
average_fps = decord_vr.get_avg_fps()
indices_list = [int(pts * average_fps) for pts in pts_list]
return decord_vr.get_batch(indices_list)
def decode_first_n_frames(self, video_file, n):
decord_vr = self.decord.VideoReader(video_file, ctx=self.decord.cpu())
indices_list = list(range(n))
return decord_vr.get_batch(indices_list)
class TorchVision(AbstractDecoder):
def __init__(self, backend):
self._backend = backend
self._print_each_iteration_time = False
import torchvision # noqa: F401
self.torchvision = torchvision
def init_decode_and_resize(self):
from torchvision.transforms import v2 as transforms_v2
self.transforms_v2 = transforms_v2
def decode_frames(self, video_file, pts_list):
self.torchvision.set_video_backend(self._backend)
reader = self.torchvision.io.VideoReader(video_file, "video", num_threads=0)
frames = []
for pts in pts_list:
reader.seek(pts)
frame = next(reader)
frames.append(frame["data"].permute(1, 2, 0))
return frames
def decode_first_n_frames(self, video_file, n):
self.torchvision.set_video_backend(self._backend)
reader = self.torchvision.io.VideoReader(video_file, "video", num_threads=0)
frames = []
for _ in range(n):
frame = next(reader)
frames.append(frame["data"].permute(1, 2, 0))
return frames
def decode_and_resize(self, video_file, pts_list, height, width, device):
self.torchvision.set_video_backend(self._backend)
reader = self.torchvision.io.VideoReader(video_file, "video", num_threads=1)
frames = []
for pts in pts_list:
reader.seek(pts)
frame = next(reader)
frames.append(frame["data"].permute(1, 2, 0))
frames = [
self.transforms_v2.functional.resize(frame.to(device), (height, width))
for frame in frames
]
return frames
class OpenCVDecoder(AbstractDecoder):
def __init__(self, backend):
import cv2
self.cv2 = cv2
self._available_backends = {"FFMPEG": cv2.CAP_FFMPEG}
self._backend = self._available_backends.get(backend)
self._print_each_iteration_time = False
def decode_frames(self, video_file, pts_list):
cap = self.cv2.VideoCapture(video_file, self._backend)
if not cap.isOpened():
raise ValueError("Could not open video stream")
fps = cap.get(self.cv2.CAP_PROP_FPS)
approx_frame_indices = [int(pts * fps) for pts in pts_list]
current_frame = 0
frames = []
while True:
ok = cap.grab()
if not ok:
raise ValueError("Could not grab video frame")
if current_frame in approx_frame_indices: # only decompress needed
ret, frame = cap.retrieve()
if ret:
frame = self.convert_frame_to_rgb_tensor(frame)
frames.append(frame)
if len(frames) == len(approx_frame_indices):
break
current_frame += 1
cap.release()
assert len(frames) == len(approx_frame_indices)
return frames
def decode_first_n_frames(self, video_file, n):
cap = self.cv2.VideoCapture(video_file, self._backend)
if not cap.isOpened():
raise ValueError("Could not open video stream")
frames = []
for i in range(n):
ok = cap.grab()
if not ok:
raise ValueError("Could not grab video frame")
ret, frame = cap.retrieve()
if ret:
frame = self.convert_frame_to_rgb_tensor(frame)
frames.append(frame)
cap.release()
assert len(frames) == n
return frames
def decode_and_resize(self, *args, **kwargs):
raise ValueError(
"OpenCV doesn't apply antialias while pytorch does by default, this is potentially an unfair comparison"
)
def convert_frame_to_rgb_tensor(self, frame):
# OpenCV uses BGR, change to RGB
frame = self.cv2.cvtColor(frame, self.cv2.COLOR_BGR2RGB)
# Update to C, H, W
frame = np.transpose(frame, (2, 0, 1))
# Convert to tensor
frame = torch.from_numpy(frame)
return frame
class TorchCodecCore(AbstractDecoder):
def __init__(
self,
num_threads: str | None = None,
color_conversion_library=None,
device="cpu",
):
self._num_threads = int(num_threads) if num_threads else None
self._color_conversion_library = color_conversion_library
self._device = device
def decode_frames(self, video_file, pts_list):
decoder = create_from_file(video_file, seek_mode="exact")
_add_video_stream(
decoder,
num_threads=self._num_threads,
color_conversion_library=self._color_conversion_library,
device=self._device,
)
frames, *_ = get_frames_by_pts(decoder, timestamps=pts_list)
return frames
def decode_first_n_frames(self, video_file, n):
decoder = create_from_file(video_file, seek_mode="approximate")
_add_video_stream(
decoder,
num_threads=self._num_threads,
color_conversion_library=self._color_conversion_library,
device=self._device,
)
frames = []
for _ in range(n):
frame = get_next_frame(decoder)
frames.append(frame)
return frames
class TorchCodecCoreNonBatch(AbstractDecoder):
def __init__(
self,
num_threads: str | None = None,
color_conversion_library=None,
device="cpu",
):
self._num_threads = num_threads
self._color_conversion_library = color_conversion_library
self._device = device
def init_decode_and_resize(self):
from torchvision.transforms import v2 as transforms_v2
self.transforms_v2 = transforms_v2
def decode_frames(self, video_file, pts_list):
decoder = create_from_file(video_file, seek_mode="approximate")
num_threads = int(self._num_threads) if self._num_threads else 0
_add_video_stream(
decoder,
num_threads=num_threads,
color_conversion_library=self._color_conversion_library,
device=self._device,
)
frames = []
for pts in pts_list:
seek_to_pts(decoder, pts)
frame = get_next_frame(decoder)
frames.append(frame)
return frames
def decode_first_n_frames(self, video_file, n):
num_threads = int(self._num_threads) if self._num_threads else 0
decoder = create_from_file(video_file, seek_mode="approximate")
_add_video_stream(
decoder,
num_threads=num_threads,
color_conversion_library=self._color_conversion_library,
device=self._device,
)
frames = []
for _ in range(n):
frame = get_next_frame(decoder)
frames.append(frame)
return frames
def decode_and_resize(self, video_file, pts_list, height, width, device):
num_threads = int(self._num_threads) if self._num_threads else 1
decoder = create_from_file(video_file, seek_mode="approximate")
_add_video_stream(
decoder,
num_threads=num_threads,
color_conversion_library=self._color_conversion_library,
device=self._device,
)
frames = []
for pts in pts_list:
seek_to_pts(decoder, pts)
frame, *_ = get_next_frame(decoder)
frames.append(frame)
frames = [
self.transforms_v2.functional.resize(frame.to(device), (height, width))
for frame in frames
]
return frames
class TorchCodecCoreBatch(AbstractDecoder):
def __init__(
self,
num_threads: str | None = None,
color_conversion_library=None,
device="cpu",
):
self._print_each_iteration_time = False
self._num_threads = int(num_threads) if num_threads else None
self._color_conversion_library = color_conversion_library
self._device = device
def decode_frames(self, video_file, pts_list):
decoder = create_from_file(video_file, seek_mode="exact")
_add_video_stream(
decoder,
num_threads=self._num_threads,
color_conversion_library=self._color_conversion_library,
device=self._device,
)
frames, *_ = get_frames_by_pts(decoder, timestamps=pts_list)
return frames
def decode_first_n_frames(self, video_file, n):
decoder = create_from_file(video_file, seek_mode="exact")
_add_video_stream(
decoder,
num_threads=self._num_threads,
color_conversion_library=self._color_conversion_library,
device=self._device,
)
indices_list = list(range(n))
frames, *_ = get_frames_at_indices(decoder, frame_indices=indices_list)
return frames
class TorchCodecPublic(AbstractDecoder):
def __init__(
self,
num_ffmpeg_threads: str | None = None,
device="cpu",
seek_mode="exact",
stream_index: str | None = None,
):
self._num_ffmpeg_threads = num_ffmpeg_threads
self._device = device
self._seek_mode = seek_mode
self._stream_index = int(stream_index) if stream_index else None
def init_decode_and_resize(self):
from torchvision.transforms import v2 as transforms_v2
self.transforms_v2 = transforms_v2
def decode_frames(self, video_file, pts_list):
num_ffmpeg_threads = (
int(self._num_ffmpeg_threads) if self._num_ffmpeg_threads else 0
)
decoder = VideoDecoder(
video_file,
num_ffmpeg_threads=num_ffmpeg_threads,
device=self._device,
seek_mode=self._seek_mode,
stream_index=self._stream_index,
)
return decoder.get_frames_played_at(pts_list)
def decode_first_n_frames(self, video_file, n):
num_ffmpeg_threads = (
int(self._num_ffmpeg_threads) if self._num_ffmpeg_threads else 0
)
decoder = VideoDecoder(
video_file,
num_ffmpeg_threads=num_ffmpeg_threads,
device=self._device,
seek_mode=self._seek_mode,
stream_index=self._stream_index,
)
frames = []
count = 0
for frame in decoder:
frames.append(frame)
count += 1
if count == n:
break
return frames
def decode_and_resize(self, video_file, pts_list, height, width, device):
num_ffmpeg_threads = (
int(self._num_ffmpeg_threads) if self._num_ffmpeg_threads else 1
)
decoder = VideoDecoder(
video_file,
num_ffmpeg_threads=num_ffmpeg_threads,
device=self._device,
seek_mode=self._seek_mode,
stream_index=self._stream_index,
)
frames = decoder.get_frames_played_at(pts_list)
frames = self.transforms_v2.functional.resize(frames.data, (height, width))
return frames
class TorchCodecPublicNonBatch(AbstractDecoder):
def __init__(
self,
num_ffmpeg_threads: str | None = None,
device="cpu",
seek_mode="approximate",
):
self._num_ffmpeg_threads = num_ffmpeg_threads
self._device = device
self._seek_mode = seek_mode
def init_decode_and_resize(self):
from torchvision.transforms import v2 as transforms_v2
self.transforms_v2 = transforms_v2
def decode_frames(self, video_file, pts_list):
num_ffmpeg_threads = (
int(self._num_ffmpeg_threads) if self._num_ffmpeg_threads else 0
)
decoder = VideoDecoder(
video_file,
num_ffmpeg_threads=num_ffmpeg_threads,
device=self._device,
seek_mode=self._seek_mode,
)
frames = []
for pts in pts_list:
frame = decoder.get_frame_played_at(pts)
frames.append(frame)
return frames
def decode_first_n_frames(self, video_file, n):
num_ffmpeg_threads = (
int(self._num_ffmpeg_threads) if self._num_ffmpeg_threads else 0
)
decoder = VideoDecoder(
video_file,
num_ffmpeg_threads=num_ffmpeg_threads,
device=self._device,
seek_mode=self._seek_mode,
)
frames = []
count = 0
for frame in decoder:
frames.append(frame)
count += 1
if count == n:
break
return frames
def decode_and_resize(self, video_file, pts_list, height, width, device):
num_ffmpeg_threads = (
int(self._num_ffmpeg_threads) if self._num_ffmpeg_threads else 1
)
decoder = VideoDecoder(
video_file,
num_ffmpeg_threads=num_ffmpeg_threads,
device=self._device,
seek_mode=self._seek_mode,
)
frames = []
for pts in pts_list:
frame = decoder.get_frame_played_at(pts)
frames.append(frame)
frames = [
self.transforms_v2.functional.resize(frame.to(device), (height, width))
for frame in frames
]
return frames
@torch.compile(fullgraph=True, backend="eager")
def compiled_seek_and_next(decoder, pts):
seek_to_pts(decoder, pts)
return get_next_frame(decoder)
@torch.compile(fullgraph=True, backend="eager")
def compiled_next(decoder):
return get_next_frame(decoder)
class TorchCodecCoreCompiled(AbstractDecoder):
def __init__(self):
pass
def decode_frames(self, video_file, pts_list):
decoder = create_from_file(video_file)
_add_video_stream(decoder)
frames = []
for pts in pts_list:
frame = compiled_seek_and_next(decoder, pts)
frames.append(frame)
return frames
def decode_first_n_frames(self, video_file, n):
decoder = create_from_file(video_file)
_add_video_stream(decoder)
frames = []
for _ in range(n):
frame = compiled_next(decoder)
frames.append(frame)
return frames
class TorchAudioDecoder(AbstractDecoder):
def __init__(self, stream_index: str | None = None):
import torchaudio # noqa: F401
self.torchaudio = torchaudio
self._stream_index = int(stream_index) if stream_index else None
def init_decode_and_resize(self):
from torchvision.transforms import v2 as transforms_v2
self.transforms_v2 = transforms_v2
def decode_frames(self, video_file, pts_list):
stream_reader = self.torchaudio.io.StreamReader(src=video_file)
stream_reader.add_basic_video_stream(
frames_per_chunk=1,
decoder_option={"threads": "0"},
stream_index=self._stream_index,
)
frames = []
for pts in pts_list:
stream_reader.seek(pts)
stream_reader.fill_buffer()
clip = stream_reader.pop_chunks()
frames.append(clip[0][0])
return frames
def decode_first_n_frames(self, video_file, n):
stream_reader = self.torchaudio.io.StreamReader(src=video_file)
stream_reader.add_basic_video_stream(
frames_per_chunk=1,
decoder_option={"threads": "0"},
stream_index=self._stream_index,
)
frames = []
frame_cnt = 0
for vframe in stream_reader.stream():
if frame_cnt >= n:
break
frames.append(vframe[0][0])
frame_cnt += 1
return frames
def decode_and_resize(self, video_file, pts_list, height, width, device):
stream_reader = self.torchaudio.io.StreamReader(src=video_file)
stream_reader.add_basic_video_stream(
frames_per_chunk=1,
decoder_option={"threads": "1"},
stream_index=self._stream_index,
)
frames = []
for pts in pts_list:
stream_reader.seek(pts)
stream_reader.fill_buffer()
clip = stream_reader.pop_chunks()
frames.append(clip[0][0])
frames = [
self.transforms_v2.functional.resize(frame.to(device), (height, width))
for frame in frames
]
return frames
def create_torchcodec_core_decode_first_frame(video_file):
video_decoder = create_from_file(video_file)
_add_video_stream(video_decoder)
get_next_frame(video_decoder)
return video_decoder
def generate_video(command):
print(command)
print(" ".join(command))
subprocess.check_call(command)
return True
def generate_videos(
resolutions,
encodings,
patterns,
fpses,
gop_sizes,
durations,
pix_fmts,
ffmpeg_cli,
output_dir,
):
executor = ThreadPoolExecutor(max_workers=20)
video_count = 0
futures = []
for resolution, duration, fps, gop_size, encoding, pattern, pix_fmt in product(
resolutions, durations, fpses, gop_sizes, encodings, patterns, pix_fmts
):
outfile = f"{output_dir}/{pattern}_{resolution}_{duration}s_{fps}fps_{gop_size}gop_{encoding}_{pix_fmt}.mp4"
command = [
ffmpeg_cli,
"-y",
"-f",
"lavfi",
"-i",
f"{pattern}=s={resolution}",
"-t",
str(duration),
"-c:v",
encoding,
"-r",
str(fps),
"-g",
str(gop_size),
"-pix_fmt",
pix_fmt,
outfile,
]
futures.append(executor.submit(generate_video, command))
video_count += 1
wait(futures)
for f in futures:
assert f.result()
executor.shutdown(wait=True)
print(f"Generated {video_count} videos")
def retrieve_videos(urls_and_dest_paths):
for url, path in urls_and_dest_paths:
urllib.request.urlretrieve(url, path)
def plot_data(json_data, plot_path):
plt.rcParams["font.size"] = 18
# Creating the DataFrame
df = pd.DataFrame(json_data["experiments"])
# Sorting by video, type, and frame_count
df_sorted = df.sort_values(by=["video", "type", "frame_count"])
# Group by video first
grouped_by_video = df_sorted.groupby("video")
# Define colors (consistent across decoders)
colors = plt.get_cmap("tab10")
# Find the unique combinations of (type, frame_count) per video
video_type_combinations = {
video: video_group.groupby(["type", "frame_count"]).ngroups
for video, video_group in grouped_by_video
}
# Get the unique videos and the maximum number of (type, frame_count) combinations per video
unique_videos = list(video_type_combinations.keys())
max_combinations = max(video_type_combinations.values())
# Create subplots: each row is a video, and each column is for a unique (type, frame_count)
fig, axes = plt.subplots(
nrows=len(unique_videos),
ncols=max_combinations,
figsize=(max_combinations * 6, len(unique_videos) * 4),
sharex=False,
sharey=True,
)
# Handle cases where there's only one row or column
if len(unique_videos) == 1:
axes = np.array([axes]) # Make sure axes is a list of lists
if max_combinations == 1:
axes = np.expand_dims(axes, axis=1) # Ensure a 2D array for axes
# Loop through each video and its sub-groups
for row, (video, video_group) in enumerate(grouped_by_video):
sub_group = video_group.groupby(["type", "frame_count"])
# Loop through each (type, frame_count) group for this video
for col, ((vtype, vcount), group) in enumerate(sub_group):
ax = axes[row, col] # Select the appropriate axis
# Set the title for the subplot
base_video = Path(video).name.removesuffix(".mp4")
ax.set_title(f"{base_video}\n{vtype}", fontsize=11)
# Plot bars with error bars
ax.barh(
group["decoder"],
group["fps_median"],
xerr=[
group["fps_median"] - group["fps_p75"],
group["fps_p25"] - group["fps_median"],
],
color=[colors(i) for i in range(len(group))],
align="center",
capsize=5,
label=group["decoder"],
)
# Set the labels
ax.set_xlabel("FPS")
# Remove any empty subplots for videos with fewer combinations
for row in range(len(unique_videos)):
for col in range(video_type_combinations[unique_videos[row]], max_combinations):
fig.delaxes(axes[row, col])
# Stamp the metadata for the experimental system on the chart.
plt.gcf().text(
0.005,
0.013,
"\n".join([f"{k}: {v}" for k, v in json_data["system_metadata"].items()]),
fontsize=11,
bbox=dict(facecolor="white"),
)
# Adjust layout to avoid overlap
plt.tight_layout()
# Show plot
plt.savefig(
plot_path,
)
def get_metadata(video_file_path: str) -> VideoStreamMetadata:
return VideoDecoder(video_file_path).metadata
@dataclass
class BatchParameters:
num_threads: int
batch_size: int
@dataclass
class DataLoaderInspiredWorkloadParameters:
batch_parameters: BatchParameters
resize_height: int
resize_width: int
resize_device: str
def run_batch_using_threads(
function,
*args,
batch_parameters: BatchParameters = BatchParameters(num_threads=8, batch_size=40),
):
executor = ThreadPoolExecutor(max_workers=batch_parameters.num_threads)
futures = []
for _ in range(batch_parameters.batch_size):
futures.append(executor.submit(function, *args))
for f in futures:
# TODO: Add a stronger check here based on arguments to the function.
assert len(f.result()) > 0
executor.shutdown(wait=True)
def convert_result_to_df_item(
result, decoder_name, video_file_path, num_samples, decode_pattern
):
df_item = {}
df_item["decoder"] = decoder_name
df_item["video"] = str(video_file_path)
df_item["description"] = result.description
df_item["frame_count"] = num_samples
df_item["median"] = result.median
df_item["iqr"] = result.iqr
df_item["type"] = decode_pattern
df_item["fps_median"] = num_samples / result.median
df_item["fps_p75"] = num_samples / result._p75
df_item["fps_p25"] = num_samples / result._p25
return df_item
@dataclass
class DecoderKind:
display_name: str
kind: typing.Type[AbstractDecoder]
default_options: dict[str, str] = field(default_factory=dict)
decoder_registry = {
"decord": DecoderKind("DecordAccurate", DecordAccurate),
"decord_batch": DecoderKind("DecordAccurateBatch", DecordAccurateBatch),
"torchcodec_core": DecoderKind("TorchCodecCore", TorchCodecCore),
"torchcodec_core_batch": DecoderKind("TorchCodecCoreBatch", TorchCodecCoreBatch),
"torchcodec_core_nonbatch": DecoderKind(
"TorchCodecCoreNonBatch", TorchCodecCoreNonBatch
),
"torchcodec_core_compiled": DecoderKind(
"TorchCodecCoreCompiled", TorchCodecCoreCompiled
),
"torchcodec_public": DecoderKind("TorchCodecPublic", TorchCodecPublic),
"torchcodec_public_nonbatch": DecoderKind(
"TorchCodecPublicNonBatch", TorchCodecPublicNonBatch
),
"torchvision": DecoderKind(
# We don't compare against TorchVision's "pyav" backend because it doesn't support
# accurate seeks.
"TorchVision[backend=video_reader]",
TorchVision,
{"backend": "video_reader"},
),
"torchaudio": DecoderKind("TorchAudio", TorchAudioDecoder),
"opencv": DecoderKind(
"OpenCV[backend=FFMPEG]", OpenCVDecoder, {"backend": "FFMPEG"}
),
}
def run_benchmarks(
decoder_dict: dict[str, AbstractDecoder],
video_files_paths: list[Path],
num_samples: int,
num_sequential_frames_from_start: list[int],
min_runtime_seconds: float,
benchmark_video_creation: bool,
dataloader_parameters: DataLoaderInspiredWorkloadParameters = None,
) -> list[dict[str, str | float | int]]:
# Ensure that we have the same seed across benchmark runs.
torch.manual_seed(0)
print(f"video_files_paths={video_files_paths}")
results = []
df_data = []
verbose = False
for video_file_path in video_files_paths:
metadata = get_metadata(video_file_path)
metadata_label = f"{metadata.codec} {metadata.width}x{metadata.height}, {metadata.duration_seconds}s {metadata.average_fps}fps"
duration = metadata.duration_seconds
uniform_pts_list = [i * duration / num_samples for i in range(num_samples)]
# Note that we are using the same random pts values for all decoders for the same
# video. However, because we use the duration as part of this calculation, we
# are using different random pts values across videos.
random_pts_list = (torch.rand(num_samples) * duration).tolist()
# The decoder items are sorted to perform and display the benchmarks in a consistent order.
for decoder_name, decoder in sorted(decoder_dict.items(), key=lambda x: x[0]):
print(f"video={video_file_path}, decoder={decoder_name}")
if dataloader_parameters:
bp = dataloader_parameters.batch_parameters
decoder.init_decode_and_resize()
description = (
f"concurrency {bp.num_threads}"
f"batch {bp.batch_size}"
+ decoder.decode_and_resize_description(
num_samples,
dataloader_parameters.resize_height,
dataloader_parameters.resize_width,
)
)
dataloader_result = benchmark.Timer(
stmt="run_batch_using_threads(decoder.decode_and_resize, video_file, pts_list, height, width, device, batch_parameters=batch_parameters)",
globals={
"video_file": str(video_file_path),
"pts_list": uniform_pts_list,
"decoder": decoder,
"run_batch_using_threads": run_batch_using_threads,
"batch_parameters": dataloader_parameters.batch_parameters,
"height": dataloader_parameters.resize_height,
"width": dataloader_parameters.resize_width,
"device": dataloader_parameters.resize_device,
},
label=f"video={video_file_path} {metadata_label}",
sub_label=decoder_name,
description=description,
)
print(description)
results.append(
dataloader_result.blocked_autorange(
min_run_time=min_runtime_seconds
)
)
df_data.append(
convert_result_to_df_item(
results[-1],
decoder_name,
video_file_path,
num_samples * dataloader_parameters.batch_parameters.batch_size,
description,
)
)
for kind, pts_list in [
("uniform", uniform_pts_list),
("random", random_pts_list),
]:
if verbose:
print(
f"video={video_file_path}, decoder={decoder_name}, pts_list={pts_list}"
)
seeked_result = benchmark.Timer(
stmt="decoder.decode_frames(video_file, pts_list)",
globals={
"video_file": str(video_file_path),
"pts_list": pts_list,
"decoder": decoder,
},
label=f"video={video_file_path} {metadata_label}",
sub_label=decoder_name,
description=decoder.decode_frames_description(num_samples, kind),
)
print(
f"{decoder_name} {decoder.decode_frames_description(num_samples, kind)}"
)
results.append(
seeked_result.blocked_autorange(min_run_time=min_runtime_seconds)
)
df_data.append(
convert_result_to_df_item(
results[-1],
decoder_name,
video_file_path,
num_samples,
decoder.decode_frames_description(num_samples, kind),
)
)
for num_frames in num_sequential_frames_from_start:
consecutive_frames_result = benchmark.Timer(
stmt="decoder.decode_first_n_frames(video_file, n)",
globals={
"video_file": str(video_file_path),
"n": num_frames,
"decoder": decoder,
},
label=f"video={video_file_path} {metadata_label}",
sub_label=decoder_name,
description=decoder.decode_first_n_frames_description(num_frames),
)
print(
f"{decoder_name} {decoder.decode_first_n_frames_description(num_frames)}"
)
results.append(
consecutive_frames_result.blocked_autorange(
min_run_time=min_runtime_seconds
)
)
df_data.append(
convert_result_to_df_item(
results[-1],
decoder_name,
video_file_path,
num_frames,
decoder.decode_first_n_frames_description(num_frames),
)
)
if benchmark_video_creation:
metadata = get_metadata(video_file_path)
metadata_label = f"{metadata.codec} {metadata.width}x{metadata.height}, {metadata.duration_seconds}s {metadata.average_fps}fps"
creation_result = benchmark.Timer(
stmt="create_torchcodec_core_decode_first_frame(video_file)",
globals={
"video_file": str(video_file_path),
"create_torchcodec_core_decode_first_frame": create_torchcodec_core_decode_first_frame,
},
label=f"video={video_file_path} {metadata_label}",
sub_label="TorchCodecCore",
description="create decode first",
)
results.append(
creation_result.blocked_autorange(
min_run_time=2.0,
)
)
compare = benchmark.Compare(results)
compare.print()
return df_data
def verify_outputs(decoders_to_run, video_paths, num_samples):
# Reuse TorchCodecPublic decoder stream_index option, if provided.
options = decoder_registry["torchcodec_public"].default_options
if torchcodec_decoder := next(
(
decoder
for name, decoder in decoders_to_run.items()
if "TorchCodecPublic" in name
),
None,
):
options["stream_index"] = (
str(torchcodec_decoder._stream_index)
if torchcodec_decoder._stream_index is not None
else ""
)
# Create default TorchCodecPublic decoder to use as a baseline
torchcodec_public_decoder = TorchCodecPublic(**options)
# Get frames using each decoder
for video_file_path in video_paths:
metadata = get_metadata(video_file_path)
metadata_label = f"{metadata.codec} {metadata.width}x{metadata.height}, {metadata.duration_seconds}s {metadata.average_fps}fps"
print(f"{metadata_label=}")
# Generate uniformly spaced PTS
duration = metadata.duration_seconds
pts_list = [i * duration / num_samples for i in range(num_samples)]
# Get the frames from TorchCodecPublic as the baseline
torchcodec_public_results = decode_and_adjust_frames(
torchcodec_public_decoder,
video_file_path,
num_samples=num_samples,
pts_list=pts_list,
)
for decoder_name, decoder in decoders_to_run.items():
print(f"video={video_file_path}, decoder={decoder_name}")
frames = decode_and_adjust_frames(
decoder,
video_file_path,
num_samples=num_samples,
pts_list=pts_list,
)
for f1, f2 in zip(torchcodec_public_results, frames):
torch.testing.assert_close(f1, f2)
print(f"Results of baseline TorchCodecPublic and {decoder_name} match!")
def decode_and_adjust_frames(
decoder, video_file_path, *, num_samples: int, pts_list: list[float]
):
frames = []
# Decode non-sequential frames using decode_frames function
non_seq_frames = decoder.decode_frames(video_file_path, pts_list)
# TorchCodec's batch APIs return a FrameBatch, so we need to extract the frames
if isinstance(non_seq_frames, FrameBatch):
non_seq_frames = non_seq_frames.data
frames.extend(non_seq_frames)
# Decode sequential frames using decode_first_n_frames function
seq_frames = decoder.decode_first_n_frames(video_file_path, num_samples)
if isinstance(seq_frames, FrameBatch):
seq_frames = seq_frames.data
frames.extend(seq_frames)
# Check and convert frames to C,H,W for consistency with other decoders.
if frames[0].shape[-1] == 3:
frames = [frame.permute(-1, *range(frame.dim() - 1)) for frame in frames]
return frames
torchcodec-0.14.0/benchmarks/decoders/benchmark_readme_chart.png 0000664 0000000 0000000 00000613727 15207546147 0025007 0 ustar 00root root 0000000 0000000 PNG
IHDR ` q :tEXtSoftware Matplotlib version3.10.0, https://matplotlib.org/rXr pHYs a a?i IDATxw\S?WX%@A ETGjZ
jQZW½7'8*
'O`(Qjٳ1qD`ΝFnݰsNet۷oǼypUԭ[/^ڥoڵkxb߿ ݻw~۷/_lmmu:t(̘1[lArr24i8Mhh(ڷoʕ+c֭?>._f͚Rףҥ z}sq =z5kDҥu5kLÇc]pttDfʹ~3TFFڶmuaʔ)8p O33\ג>}aÆ
Ś5kN} 0 f!Cp̟? EQkץK 8p K,[nEHKKN} :tH+WdNĉؿ4-,,Ld2aÆ-<^|˭6իWN|wҴTQL1f{MT*Œ%K ~ >,Ĝ9s i^~/Rc <<<<6a 1db ` ` ` x0|21000y><
TZ ;,,,pDGGsΰFՆݽ{76m%JطoںW\ L/EP*puuի5>}: `ii :m3g*T \r_~k?? :@T3fP̙3W o _ԯ_ (Vh @xx8ʖ- ԩd2d2s(ĻwիWh֬t___\r > D/OJ wwwIӚ6m
;;;ٳ@搻Xh&%% hذ!222@Nivvvh֬= Gd24@ ۷ѹsgmt Frrr[' -333tAǏ sΰ9ʖ-ÇU ԩ ߩ@TV
ФId2"<< 2 VB߾}acc;;;1BW ߿?J*sss8;;kooom} [[[4mTjի {!N*ot ͛71bJ, sssT^ [nUA>w^TZU;!*WFaaaooo!mL6rtt4J*nݺ J%
:---g :p :Tz ?˗?4yzzJC"00PjcggT6]L&iEj]o^/@2ewa(U
ڶm[@&恬<<<<1000b ` ` EDTDFFbȑ8q"oܽ{=z@.]PZ5lj_- hݺ5֬Y͛7Ahٲ{͚5ömPF
{h"L4 ={͛Q\9Wc=Æ
ɓݻw# cǎŒ%Kro˗ǖ-[_cĉ˥{믱w^4k ϟGӦMaee`̚5 ;wD-GGGlٲ 0c pV2Lm]!ӧ ֮]u}W,YR$$$H?31` ZM6 Zj:uPaddq5p~ٳg SSSi;wVVV W."## """Ԗ;{ ?>]t
:}t!5 ɓ'B!J_s]! hٲؿ믿Dɒ%Ezr]aU_|IBB!T1lunraii)֭+v%l"*U$ʔ)#^~-}۷1| ,Y֮^zf͚cccc7n^@ې ٳ^^^Hܾ}[ ###ccch߾Xv ۗ/Jx4m\)RD,\P >|X:FYח(uKN :$ 7nHF!u?؈)SHkԨ!w֦\rj1O<VVVs:}EXYYXiÇz~
!<;B000y @<< ۘryyyEDT,YUT WP zڢDx̡TQT) Pu8\R <~XZӧOѾ}{e:vСC ???I>>>EJދWPho8&&Fo jժ-[#GĴi @NKI&j)8ebbb`gg6RJQ1x`jw{CEƍ1w\4j ;vݻqYFZ FH"HNN8cbb ɤZ6l&M`ĉpuuEŊ!sbkk L
ccc\~=Wd'
ٳ'ΝjժtXx^J*j7778999xgLL LMM5
{{{!+M{{]vy}kԵ
tHNNw}s}{7|I&_mI&̝;AAACll, D&_~Xf
Ұk.x /i]vEbb"v s8aԯ_ pIlڴ ?bcc$ @||YDr_y @ֺT&1y 0db ` Z7 @AEDTݹs/^ļyзo_xyyv@&?{L d2 ?gjyxxd ]DۯCٳgyShQ 8p O<իWc.]fffYZۭ[bРA:u*飱.m(B֭[4^d Μ9www 9ҠbŊxU7Pzu6NNN(V+K233Crr]_vuiu[K?˗/qK.w^P((S/P̟֒? r
5koVEHN".X"nݺ!F;:Tux"Cjj3d2=oKNEwL<QQQpuu5y ; SNr}UV(R)֭[F_6wވ®]QF([4 jˤj|isvvF
~zddd`ƍڵ }-f͚Rͪ>˗xEm<),<yy +yy@_1o?b (<<yy +恜Q!
֪_ ?u999[nUi&Ǫ QQQ]ƟUѶ%K
MÆ
m6I |QQQhٲ4ݻ5jƌ;wӧ?~|kn֬yfiZLL 8=\pAm ˗җrʡB
VkUI]կ_ jJMMŖ-[V122B:u0}t=W_}Ǐ;G Azz:jժeP-h;
UѶZj_ 7o{?_HiZXX ?~,퇾_-ÇKmn߾/Jl߾]m]۶mS{z`ԨQCmXٸ8ap!xzzu u5Ν;|vN8ٳgcܹXv- ; ƍhժ7n%Khݖ/LMM!U <(=/իWGHHڟ_d ~w_}~gݻWˋRRR$u
{]Sᄛ7oQرcd'O" ȑ#` ` @
3}1dyyyy0c ` @ "UTQ" gϪMwqqoNNN]ܹS[NTPA)SFm]+V ċ/! , ĨQȑ# BBBv}ӧO {m۶L2E(J})UQF ~ӹʹs焙v˗/ŋuiiiB!Ӆhٲ8v8{HNN !Eppԩ077 Km/ /R8;;OJmnܸ!lllDժUɓ';wHRRRDժUEjΝ;ņ
hժ&==]4h@xxxHb
!đ#GvRŋ͛7 g5` akk+O~%J*%bcc6 СCEժUxnڵB&ɓ'1p@abb"N8BϟKuV\YԪUKz(9sbOXYYw
!bѢEbݢyVÇV4kL[\R888
kB/jf߿/ %K o>1c abb"F -_={ػw8p믅q^ /ʔ)#jժ%m&6n(܄1bƍbǎ ZjCJm,X d27nؿ7n(] ] R{=z2LرCj/D2eĪUEzh{ fϞ-=bŊbƍqrvvoB ʗ//Zh!-7eann.n)ٳgI*UJ >|X~uF )E@@db}[ j^|)JhԨطoXlV077SLQksabb"J,)*Wv}%K۷oVnݺ>NTr<<QJann.>sqEaccJ/ 4ޯyB3p̙Q֭[kb << !
3e/y-UV ooW'L2x 2d wQi~Yr%zs)S"00AAA:X By .恏 GAb x00y .恏 @AXDD ˗/uwsspU$''_vV˹stΓV]c_p}K,%K~>Oӧ:-[E }v1|< !)))3f K(Jd2-8qՃ\\\0k,{C&aӦMyR[`` N8arm2L/,,L]\\ ;;;XYYcǎq.]L&ChhAdXr;ŋ6l<==!˳} v B
Xb;o]ܺu } *W BebРA+Wj=GƍX :%K9ʗ/s~uQy＞,Ͷm uDGG}(Rd2mۖQ1OA2 sQ6f8::G G&M`eeLK.wIDDT/(]4Ѯ];B&e{?}m۶
^mϝ;{RJ022W_}?J.
ԫWNhS
vvvׯ s߽>-&] }8III?PN |ؿvw܁/6mӧʕ+7n 1jԨR[PP,--Q~~]4hqL2jtׯcɒ%077ĉѢE ;w&&nz ֯_uvڸ|vǎ Cѯ_?̟? G A߾}l8z( >,*,, hӦ6B|Tُ
v
k ѵ WN _|!6Ȁio>.n݂:/o{ sC a ̜9#G _|*T9sHmmڴ ׯ_͛7TE/Μ9uꬃ߽>-_pI4k ְ' <(͏F>}PX1XXX~퍯
6m,--Ѹqcܽ{W]rr2~+Wr NNNXOVn'O?aرppp@% dw ,,,
`077GR0qDKT6ŋhѢJ%\]]zjc{n4h
E7.^T888`ĉt%۰گݻڵ4kɓ'&&&o߾ F4i~DDr0a5j=zt+5syRJppp@NVիWѰaC(
TZUڻw/lmmѴiSiW={>}: `ii :B`Μ9Pr9ʕ+_~E͛7ѡCAP֭y#Fn+WzulݺUc=ʔ)B&MΝ;Z3sW}. '''#$$:uRkWܼyӠ/?IѢE5^ 5j 9tn-[ }:9iߣH"x4066ҥK."* ɓ'իaaa+b L> eʔ\.GŊoߎ+ uٳgִ{nxzzŋǠA>ufJu ݻwcǎ>V^
N`gΜQ[^u Gjj*{}EBBF=֭+
ZnO8x YCj wwwѫ6m
;;;#Ȋ߽>= Ǐqϱl2ܹsťthìY`oo_M6ʼn'PV-i].]ٳO?!== #F_vAG ŋزeA/X 9/_֩hpuuł `ll Rya̘1>|8Ν7oJ9'Lj# uԑݰau놶mbڵ033ԨQXz5M&}."11 =BŊզWX2 aaaj0a5k7
0[>dʔ)prr5k?/DѩSwht1ڲR
^| cƌ1 ב(VڴiCT ,-- 3gѻwo 8p III/ ouQQQQW Druh =z4,X ~Ǯ]0p@ТE ̛7GΝ5nӦMҥ z DDD`ܸq6h7믿֭[?{#%%֭×_~+WݖG1~x
B͚5n:7Nd/˗3g0tP <@^0rH̜9 :tÇuןTOD1c ./>"##1tP7666?~<ڷowj K*WG /D_͛78q"pXYYIuMڬ qI 4iڨΝCxx8N"Eϟ?ʢE5kӘ2e
qAL8fff:t(Huǎ Ô)SPR%ٳ_5)-Z Zn
{{{,_ 666s(L4 +Wرce8::|~>{4i^x0 ŋSͪڮ]?|P+,,LL&CŊmM DDDYEʕEZZ۷o ľ}i)))tҢC4///T*ϥi+V ģGB 8p@ kYhժڴ/
"$$D@T\YddduqqE Ҵxaii)Ə vB|R~Mj
6mB $|}}u'd2سg4_"..NrYjyǏ bݺu(J?
! 6&M
BDGGknjjￅHLLgVk{-!Č3tǔ)S {ni5k\N!&O,/_._~zQvmajj*N8!\ #**JXXXI&/ Y8q9r /^,Mz mۖmVxyyM ĩS]655Uu4E899di +M۹s .]$?#)J믿f[NJy uzxx鱿F,qQi5~? >\(J%Mydjq]2VUajj* >|]ߩ[ZZj L&nܸ!M[p bbbB 9rD W[K.N:B]KرcGQغu]!yYi9}E*Ur]Bӧ \1=88X O<BghkժhڴiwoO oAHDDURRN:iT =zC/*իWWUj$կ > B]I-Z:ܽH;'N@BB:uꄴ4_ƵkԖo֬J...>ܺu ?F>}tg4mŊ-t>۫=ر#m?>*W G HKKý{]#G :TVFFF )S92((}_|.] 44%KĴiӲ]m ?ӧO : sϓHiÇg\ =z 4~ܥK\xHOOs)UQ="##1|ܾ}
6nۧXzk'O/5kEaĈXd """ dޚ *TUVI&8p ̙u 0zh /^ GG|
~??|07nsӧO#55Ug̋/pm]֭5ݾ} <@ΝjQdV4ݼy۷=ajj[nISz*^~=m.VX5"W:T-
OOOk֬A5`ii SSS4l 4-*T ###4nXgf5kbΜ9Xx!&"*_ɒ%QJףuQ4nvvvҴ+Cs_5)C>ks jrs}iӦj+T%K쬪Qu 8 ;;;4nX\iڴ).^t-Z...?~x~&8W111@ɒ%mwVVfkkX0y s}GGG aoo/_ ҟ+ ^#> @c|/_ŋ/UmoW<%%IIIj hϟ?טٳ\עE"""B͑6MuqmtU[l -[6U-
3 С CvtQ*pvvָw[ иݳg @'88mڴ̙367nЫE"--
ϟ?ϱRbE :tH
Cj o_u ;;;d2 ;vL l
@~IMMELL%P=L[[[D9w)_>^EvѢEZՙSg6hv;y$ ?~]vC ''\5 h;*y իcҥҴ'϶oFrttğe˖>}:t[n\ryVQA<ۻUG1}>UTՏ
+z}G vvv(^8٣uP
T8q&L@֭XZZY=DD~ip-k aaa(]4,,,v;~!p-Q D 'EDDRDzzjkmӰaCҴ4lݺU^U_>>>HJJƍuqrr.ĩdݶ!իBǏ vNUpssVXm;\={?ڵkѻw<8עE l߾ Ҵ
6WkʦMP(I_6MMM5:4n2,cWk.ԩSG֢E Ҵ۷oŋhٲ%sQJ5PTT
@yi&zJkEpp`ԨQJƨYf5幫\.GF4$ 4jQz*[8::{066ƦMc̞=[W֭ða0sLK}Z~= Yfj ^~
jĉU nj2 /9C雑TPN L> iii%}gh Puy4͛<*Wv6lÇuu\R`>uyy/^sm ƍC|||?d$"C{!wNammyRSSe?5._OvaDEE{WMTq,""z~'4n>>>oQH\pŊC>}ЪU+ԭ[_5~'c
r- l}ݻwhlڴ 6l t˗/ǐ!CЮ];8qBanbԩ3f ?~ ooo{ؾ};6o Bd2̙nݺz\.ɓ'QN |WRc066F@@^w/N ;w*Wʕ+ F ݺu÷~Wb 5 ܽ{{F]q̜9ÇFfjڴ),XEB
믿J*a/T*
*` &M )) wF``/qsѣ={6ڷo2eӧ;w."##վpիWӧΝ sssL8㌍1n8 6 hڴ) 8mVPFϞ=1zhxzz"55oFHHm 2e
vڅ
b̘1pttč71cH 1b^~
777_8qo.moĉh۶? k֬Q]]kƍHOO שS... I&~-:w]VzMܹsqYL0z2hthܸ1p)it!7:رaÆAZfʔ)_>z?~|8VXf͚aĉx
~.]Z:E;֭[yҵD\~ wep 9]tA3gL2KDT _R%Xn \]]QX1?}4zԋ/ ]nٲ%
1~xxV~wDEEaԨQҺ:v3f3f@RRF%}+MDDD EFBVVV? |R;;;!EzDhh:DVԦ]xQ !!!Ҵׯ_qƉҥK SSS$飶?,R:u X 1{l}qqqֺ֭uZԨQCL4I
!Xb ^xWc)ͅhܸx6+T Zh m\\\ )S;~r\8993g i b
/6lHIIڽzJ"E"E;w
ٳRG5k
"==]U
ѥK'bʔ)BTj짍>e ___ LMMhٲ8}FXѧO akk+,--E ē'Oddd QD P(D6mľ}4Ϋ pBQjUaff&DzļywuѦMamm-
^X~4?))I|A wwwyfڗ,Y"Or:wuv.i[EjՄ˳]ov"##
^Veʔ):kz :T
!EjĂ ^ * kr\888#G7o!HNNA"==]ZZ 6,_"#/ >|(z!섹X_bԩRruuK,Ѩi˖-B
B.ZjSNi=pJR(JQJ1rH>Uf&$$D#ݻWTRE wwwg oHNNC Z5k R;Cl+fնm[6_TREmZZ9r(^ ;vN DppN[NvT929e7oވ~ 777aaa!Dfę3gO" @!8wtU& :qFcLLƱ_~ ˢiӦ5 :p6&g6ڮGi B}G[ڎ<,033ŋ5WB3_r \.J(!۷O""0nݪ}W
}D]EѢE ϰT 1c $r'NXǏE }";Mi KDDDTܽ{_~CDDDDDDDDDDDDDa,""B(**
nԩSq-01ᝅ>4.roΝhذ!߿ G""""""""""""""""""2G""""""""""""""""""2;` dQ <}VVVd]Q%WPdIߜ0 ""<@DDD :QS8;;wDDD G ) Q*lY ` ""kDDD ;`VVV 2? ""+>> gka,@DD7
k
}<XDDj(akkk~"""m~f""Uز <@DDטCuc""""""""""""""""" XDDDDDDDDDDDDDDDDDDb,""""""""""""""""""@EDDDDDDDDDDDDDDDDDd v""""""""""""""""""2;`$ ""u :{]eDDDwL'"">~
}8@EDDDDDDDDDDDDDDDDDd v""""""""""""""""""2;` XL&L&Cxxʕ+!PLWDDDDDDDDDDDDDDDD BrJ!444J(lll
'kΈDdd$ :vtt ti߿t5*IL qҥ.?Ӊ>mDDDTXſ~n#"" @1Qa[2۷GXX >'[.'+LfΜ0̜93K! '"""zL'""1 Qa&o+(( XDDDDDDDDDDDDDDDDDD V\ L L .OOOǟƍXb(U:u !T̝;k׆-d2Ʋ7oQreXYYnnnڵ+6oތ z
e ڵQ
L5kf{T*!ɰfWL˝>}=zjQ*pqqMǏ
.=z4TR R*U`̘1xe՞ #}yxx@&aĈٶ;rd2Ciz@@ d2t.A
<==BUk7
pwwĉl{.
WWWXXX5kԩS ! ;voAڵ333(QXn]ɓ'0` !{Ν;zgJJ
~w4jŊ жm[ݻWr844ϟ?Lj#PB(l_7DDDDDDDDDDDDDDDi0=
R KKK6 8kN bll +++DDD`ӦMشiFٳgޛ7o'NVVV&f͚ &H͡P(p ܾ}6l@LL lmm5uϟ?B < 6l{Cj߳gO미x"_*Uh{HJJ%:tAVZ{K Yr9LLLC<|/pڵCll, @T nܸ7n`ٲeرc6ls =Fٳ'F uay6U6///.]Zu !зo_X@f' [[[;wgΜAHHry1~xR(HMMիWqUXwF54ݸq#zd RRRpE\x˖-QR%G V ݧǏm۶ckkkŋ8p 8[b02z &&@H\\ V\-[?v X]tAdd$ׯ 5j"## }Ehh(믿">> 111x) 3g,Ys{\+V >> x /Ƹq㐑6mŋx5+ 8p ]t)ꊳg"11 8x !C]6*W j#[M5CR$%%aȐ!B믿Ɲ;w!!!ΝѣQD ץ#Uʕq1$$$ !!/mɓ':דc =z8x6_͛ z{ .:_}wx9 @lذ۷o1vX(
㏈@bb"p94nhӦ
Ԗpk$''Ar
㑔 ;v =B֭5}9Ve
"..qqqGTT,X kkkcѢEuz
۷GLL J. 11^'ܯD4o ׯ_7BCCk"66̓%,Y \akkÇ#11u떁Gk$&&]
Q ui
1di$,_ ~~~ I&͛7Zדk" E111;v, kضmW.-P(дiS_ Zooov `ҥ GjgϞ 2ooֆOÁPZ5 ZnݻwwͶaneZy}[.,Y ג ˗ ~W/^ CӦMիCe{+y!,, ^^^8p lll0|p>}:Ҵ Bƍ*Tf/99jDDD` """"""@Æ
aii v
kÆ
'''OkiӦ ^|s*UuZmڴ ^)͛m']F )uVɪE033 \zUm^ =`ddǏ K RurrrBFCu{ĔDEEr/4 !qF 6Y999a \!H_lVoS$֮];XYY :: 0ydmƍsss6oތXԨQZۘ[n KcccǣG: fUF
鶔֭kk]I Z ܽ{jTVNrA:Qu1bkx%Ο?MϞ=̙3acc#F#""O 1 ǡv:w QF:oWR%*UJ۲'N jժGGG:ŋ 33h
j>ʗ/+"555k.]t}wEULiӴiS @TT߿!H_۷ڭU={ U璳3>3mlllPV- ? y& tM: i .@@~o.+̞=^^^(Q :e
-%%Eqc}Qms:>?~tKƸ8ޮ]%c(\.>=DDD<@DDDb Rz ""ªWu3䖃y
غu4]5U͚5Qr\aaaؼy3TZ_ơCߢbŊ SuT
UUXv= F ҥ 9ڲ1z}ݻ7 ?~2e 88QQQHLL'OHmU#公#ݼySڳvCCDDDDDDDDDDDDDDDTxXfj!888 }˱۷BXv-܍~:t耥Kիx,Y;;;rz.y/:`n;kڵk BBBMXXN:~ s!"""˿+UǠC!22R w'(Z( Yf .^ +[I: :tHڎ~2 _5ufS Z{M||<Ο?u^
ϟyUfM鵡qiaw٧ըQCk 2 t9reʔn]sN%"""""""""""""""ʍ|emm
:k 2GYl6'O +V >>>SNFZZ >ӇJ,tR͛xZWrrrCY';2 ]t ,]TEO>ҥK ݺuMyNՙƍ8w4!#5mE L6Mk_:SNEjj*Fy/ `HJJXؼy3welll|FW^a:ק:q-ϟ?ǒ%Kt.߿ qE :::DDDDDDDDDDDDDDDDo˷XUV ٳG֭ ??? !Ch"3Jdd$` ds]
~g
о}{\tIݻwm۶sbdd$t|rl۶
гg\khР.]{Iӱ~7 P^=sQN&L [[[DGGG ?~ >>>RbEiAѣG0666h$1 L4 j*|Ҩa6mf̘[[[b 2VZHny&Ν*U`]jO> s|}}qUi={e˖HKKC1`\!e*UҥK 6rɓ'퍘 4h͛RӧOG(y 0rHTV
oAFh"Qbccw^ _|EN!bom ///TR%K!""wt""O Vr4
GGG #L"=&""*Lk;w.ܹҥKxR'cǎ ˗/˗/?`Ȑ!>|8+u5j hp @tt4~l߾ ۷o,,,}ի̙7oҦM\G'NHr9,--#^dIz ۶mC۶mqu4h J (ջm66oW^8w Ν; s1C۰apEY ,
accxkXjk 6 {{$>> R[Ll͚5f ǎ ;7o sNXZZ0to}~:j
B Rubgmm[iӦ
+++,[L)mطo p) 2C
222:W~gyq
REl0oîQt""O VZAV```EDDGm,WWWM6(^8=G """""""""""""""# gϞjT
ѧ >]'
!+ammeQ>a """"""b ""*\ """"""""""""""""" XDDDDDDDDDDDDDDDDDDb,""""""""""""""""""@EDDDDDDDDDDDDDDDDDd .L q>WCDDD Z,@ }(DDT0¬CD~p,""""""""""""""""""@EDDDDDDDDDDDDDDDDDd v""""""""""""""""""2;` Hpd2d2w9jBCCզ
NDDDDDDDDDDDDDDDy$ \w~CDDDDDDDDDDDDDDDDy#`G+WDPPKT
nnn055ͷ: ooo\t)j """" .1
ϭ¥K.>.1
'""B)ky4[J*0TRVGdL8 \~=j """" >)
{ϯċDDD !}|Rf,~Qa5ӼOKDD@E \2 ((2L/<<\j?7Fb QT)t)z{{C&!00;w.j
[[[d2eorʰ%еkWl:3 6 e˖9ѵkW\ԩS;v, 1k,$$$h].<<\qz[2e ɰrJcbb0zh/_ pttDNpywqqq:u*j֬ kkkXXX
½{]?}x9
ذa իWOBB>}Oի%J|Ps)<< x IIIشiv`}u4o ?y,MMMq ܹs+V
??<"""""""""""""""*8{ХKDFF~ QF!22R зo_ +OO> 9s`ɒ%:oʕ+Xb /^ xb7hӦ
.^ׯ_#**
^ХKiKٳ'\]]qY$&&"!! #1d˵n6l@DD $lٲnnnq hq%== :uPHlܸ +n ?FR{n$&&">> .]d/_=#"""%{ׯ_wDDDD$%-ɩ%&& &EN>͛7 .\oF˗#..7oƤI sss$$$`ǎ hݺ4hѢ 2o7vX @]vZѱ
6mM >>>Xt)ڴiGprrR[nǎ 벰@Qn]/_ ۶mÇQt 6oތs ѤIi^ʕo>#666q}b߾}Z4 ;1qDڵK纒/>>>Q,P7K " @ˎ%-""yއ_v|s??\ DDE6l prrB~6m ˗8x6UTQ|զM+by&ȑ#WYhfff WjJ8qDk
4P|P(0f֭z:vJJZ{s]3g΄>
DDD<@DDDDDD v"GQ5j*UBRڿA:TV-8::T&&&(^8 ::Zc~FF֮]6mڠtҰL&Μ9 xAu:N7&yʕ+ Tedd
:ۍ ? qqqߣGr]^""0<DZc ""00|K_BB;1/i-@`/""k\HDDD f&rMgRg^"">:r?"$$_yQQQDdd4"_j$"""""""""""""""zWET#Wt >F \/.֯_ |ݟ
Z7NJ˗ lYk Q]t ;v _~ٳg\>;lӦMv -[Jǁ>NE UZ g [. C E |Gpp0 `ڴi077u
666R
6}t4?)) wF۶m Ҽys ӱe ߿ݻcƍj#][n ?;^~
sD~aÆ
P(ZF͛ y&Zh/^_
Bٲe-Z`eW^/߿\ӧ
"e*UzX+E^R
LGG.ZQ'""B)ky4wD 3̝;wAҥQxqձc˗˗0d > VVVF5j hp @tt4~l߾ ۷o,,,+u uDCL> ijg(J f̘ѺرceܸqƐ!C`mmXb7n WDDDDDDDDDDDDDDDD B kʕ
Bhhh~B3"##Cw9 ߿t.5*!""*P^!00ҴiDD4z #G}P/Ұ HK-"""*fw>>m>y 3r9лwoܹs'TRRR￣QF(Vmb9.i}
Qre]1o>^ >,I2LyfmYO=.]ZK4a;2oooѣ:r6nܨ2c%..N
ӧ%INNN
PRR{n9;;瞳kkFᘓܴ{n}72eJG i"W:r䈎 9h}֭[c{5c I???IҩSt)YF-?23e͝;W&_>Cڵk$w} ;vhŊWv
4|p]Vc$%%/ԗ_~CjժUruu$jJK,;oqܿ:~ h8jXJLLT.]$I&MRbbŭy撤ѣG+&&FnnnzwT={VF$͟?_˖-y%KСCZr222/]vKjԩ*..րt*99Yڶm"##d
֭[kVVVo߮PeddhiРAڽ{5g ]pATBB}m֢]\\ {1k:t222͛7+44TOVe633S:|"""\)--M ,-[~TiӦWFұcǔjӦMӧݻN>͛k͚5TrrrssuaEEE)((Ȯ/j֭2L5kRSSD=Ӛ3g G銌TnnnckڴOX(%%E?&L5w\=3:\͜9S&I[nՋ/hMn:͜9SJII3HV\w}f]ӦMڵke2+(99Y)))JJJҴi$IW+baÆiذa.u?G d2iժU
. Vvv]y ly| u^^qr3 VU^ӆ
$I|Ic]HHVXtmذAF!R͛տcY&M$]d!C?KzR^l ۷S⢞={׀̙3
3ڬZJ}L&6nܨ~YiӦK/BjJ۶m3FrrrRΝ;ײe4i$ҥKu)믿!뮻Ҏ ;J|]i:zwm۶#
Iy ԬY4n8\>~W͟?_ /hΜ95p@/˴iӔ&M3
ˣtmVӧOk߶={( zK/뮹{JKKիmcѢEs/tmСCڰa {
g1Ljٲ-Z"-YDK,˭#==]Æ
(/|͚5K&MMZm+_6wUZZ>#͘1CO>d݄oԩzm5j_]yyyZ`,Xg}V6{ ?~\Æ
Ӷm hܸq{?7FtiB ># cOI| (ӧoYրRF*iOXf̙.O}vtM WZ~233 5ĉ+Wnnnb݇ ~(Iׯ_+[Ҍ&Ol_]Cz%TϚ5k$I? CBBSO<+$] 5+
4H~~~JJJҏ?h,_jդIcZm+SN(NׯWQQ<==-
ݮe={ؽ [Jݻھ_}U'O,OOOVVߔ7%%ŮkwÆ
***Nju_~Y*,,-h͚5rssΝ;5sL
:TYYYС|Mٳoj u +66VԣGiF͚5Ϻvj{$ubt;.wqq1KII1idĚ8fIRϞ=mnW2ZסCTXX(I*((0 mmKHHP||BXS l/{euʈ5 gE*+I:umzoՈ#t7G&ɸ͝;Wtb?4i$uQrvv6]RW־7oVZY]秎 ;J}7[6F-isdddhС*,,-"ɤիW+:: V`9CݧSN]v6G 58n
ХKIRllΝ;WpqqQΝ%IyvmFܹv;v$zFڵk'rpc\o^f=:udLW5j$I6...6v\eu~fP1;t`ul|\MɾoyyyYݦ<~~w233?J__YY}sqol/˕oH.--x>}hΝFѾ}ԳgO#Iĉu-(//O=zŋlOKKӖ-[4|p}m[jɓ'KΝ'xb +00Pڵ֭[\IG_4yd_hWRv=
SO8Yf4n8IϿٳg+33StE=3裏b?ӧ$i˖-9s%]7xCǏW&Ml>'Od2d2)**b}''''G 5[Nw|}}mVZg}8?333oeƍSӦMms$%%gժU$I]nf͌|MM>(LKK+yI&LP3HyMW~GM6q Аhﻡ@]5*Y M(Ņ[ κ2 !8\SC;Uov)//ONNN
2
1٣0ko$]WiiiFФIB+EDDoKٳ/HTZZ,55U.lRt [oxxrJ1b]RR |A}$$Ӷ
8P6mhvZ
>OHG͛7֭[զMRƪW^F%___-_):{ ~"bVZҥKzgdc\]]bDh 9R^{N|EaV@@222T\\+##CV?hyyy߿13eVB{@@U\\_ :tЬYԽ{wX s9>uTBlRfң>*4QQQ1cw;Cs5ХK$I={^jĈFiڵW\iLYnAA
uI<ч"UVqK.U ?SX5zhL&} ~S Zs P9zƎUn[ GM( A. @CF.5Zn]vi
Rrr "IFvܩ+V(""BRHH ]vY-_|Q?ƌVZI\Ѻuk
:T7n'*00P111裏Էo_);;[^^^رN7xTH >|Xcǎu]|}1c~WԩS' :tHO<5k" W\\:w\fM6՞={zj
0@QAAտ-ZHw.Y/֞={裏E*,,lV۶m5zhcD{vm:r|MyUff G ).::Zoڷo/OOOh;e C[l1ڻl6ֺuolm۶i*٬Ν;kҥڼyH 5;,7?OuY*,,TV4m4 8pU6g YM۷oW^Z:t蠢"y{{뮻?OZTUEiڵZ~&M(33SM4Q߾}qF}F$M6MիK_IҨQMfYÇׅ r,j*2VC3k,+֭1VUwdWQF 4F:r?-ǪD> j А #`?du4 T-
Pd2d2}߫O> Nh"\1c 5 TppŲ@ES{+77a4Hޥƍ;( PS(BHGƌ1c8: PØ@EEEl6+&&Ʈ2͊Ҹ , S@-sÏst A na, X `'
N` ( ;Q v N9{:;: Ϗ rn@U `/r yP1 ؉, X `'
N` ( ;Q v ɓ'e2d2tIGSej㊎6)oaUHDDL&t=\1bD >` (8:rNNN
Vpp Ng V)L+TTTΝ;g,;w\e P*""B tt( (L+Oϫ0Xg- PWXeAڂ,5c
pU͛7WbbcGS+kFDFF:: j"͘1T֟@]oF vt( (J+.(Xg- PWXeAڂ, X `'
hL&}73fd2YN|XF҅ %Ip֮]-[hݺ[KKJJ{ncY@@8 ;vL6mhn:
>\$___:p/_[M6={*55U"==]ڸq>(IuaIdΟ?͛7kzꩧtRm/] &w1y{{EG Ց#GqF-qqq4hN>-IrrrR@@ݻwY=]&??_4
ǜݻwoє)Sl?rHMJMMՑ#GtEGG/Pnݪ,f \egg W\XߥY 5W2U=DڃH%&&K.I&)11ּysIѣ#777;PjjΞ=QFIϟe˖<ޒ%Kt!\RJIIŋծ];IҥK5uTk:prssLm۶Mrr~6L[,m߾]K)**ҠA{nkΜ9pRSSj۶E8=cW]u!edd(''G7oVhhN>+++ˢmff xEڶm{y;v~V>}taEDD(&&FJKKSZZ,X -[LovӦM3Fcǎ)++K)))JMMզMԧO /ӧջwo>}Z͛7ך5kdÊRPP]_֭[e24k,*55Uz5g >>Y | @8 w߱P he{+Q0m߾}ڰa$wՓO>i ъ+
6W^ш#Qj?YYYڼyo,kҤ$)55hȐ!-FR^ԫW/qk$gϞz5` }:s6Vw}'ɤ7_~lڴ-K/PZҶmیԿsߵl2M4htR:uJnnn-FȺ뮻cǎR_WZ` =ݻk۶mruu5Wxx |A͚5KƍK_%I/̙co
8PyL6MIIIjҤ;O$*Yf$ :=aHHz)_b$i WW4h Xj*I&UgggԩS-UTT$OOOB+EEEٽgggcD={ؽ+JOO7n%5 P_ٳGYYYʪӺ\
-_jUY 4tW2录|wͨM +Iѣڴif͚)!!AV vj{$uQvy wX]⢠ %$$(%%X^TTK0kd6%I={]^n: :tHruuUAAQv=l{=h٥'$$(>>^i!mdxoJW^VG)XJXVq%SN_o|Zb~9sbgΜ^% .q# /gg-! ]˔Wj
Pn.\$5k֬`lg\sͶk(%___J+)d WVLuXTT+%%EEEEnggϞ5'%%+ޜ ~U-%RӚ)ShܹjԨ1rZVVe )Qɞ)+ǬK. 9"|ۈ#6bJۍ⫿矕%&&*11Q?$# Pr+jӯ/k+[BBB$ɘ^&4nX>>)Yƍ,)D+k:<[J1}useb+,//6V< egYY#Iݻ,YoT{ H`
NNO[tIk.[ѣF^t"Iչs* ...ܹ$?/wn賝;wnǎ [o(rssSv$]O[X] nLWK,vGN)*[F$IOX6+]>wno,+9f 76 pQ$)--!CH<˭nꫯJճg
OEEEzklѣGK+}Wj{K͛ROڰa$iС"##%I| ;V l2 3f$iŊ:p@X=b9;;+99YӧO/mEyyyʛoiʚ[oU駟Z}WZejrvvVnnϟou^{%]~άYl2 ֹh
5Z u{3wu]7tC
K%Xg- PWXeAڂ,nfIMֹsg
}ZU``t-hctjذa˗+00P5RF4a ;Vy9skӦMSFԸqckɒ%2e B,T{WhҤ^z%GO?] /
`hݺvڥ(((HɊW||$] jΝZb"""뫬,hڵk͛WX^|EO3fZj%I*((P֭5tPmܸј2*&&F } eggK ;vԩSojÇkر딟/o^3f/6mX=fNt!=j֬ \qqqܹs17mT{ի5` *''G
Wh"T[ggg-^X{ѣ>-ZPfYm۶ѣqmȑ#z7uwW
RDD,XG yThj߾<==U\\l;yLmٲh&٬[֭ӛoimۦӧ믗f:wKjrvv? @d2y W #Gkɓ' N5bZJ?.##Cj=) /??s =5==ʋ P? j. y#` ( ;Q v D ؉, XV'Oddɓ' Nm+::ڈ־}{G` -[C2d =Zf_kݺu4h5k&wwwcǎ2e~~w=jݺ<==nM222|Ql뫂 8p@ ˵uViӦ1ԳgOJ\+::Z7n |` %)>>^ >,I2LyfmYO=.]ZK4a;2oooѣ:r6nhЩ,} h']xQ۶mӶm駟j͚5rr]go
'٬]'|ʚ{O+++ˢmff xEڶm{y;v~V>}taEDD(&&FJKKSZZ,X -[LovӦM3Fcǎ)++K)))JMMզMԧO /Ǝ ۷+==]Prr~mO>ŋ?.]ӕkתQF7]tjۋ/C=SN)55UZtu =G-X@SL畒l 9R_h 28p)++K?C顇 oԻ+Iz뭷m6cѣGJ^x{p(q B eg_ @Tr> @cw+KX}aIһﾫ'|X+V(==]6l+#Fã~yfX֤IIRjjL"I2d>cѱԫW/fھ}<==%I...ٳ}
0@~Μ90ͪUwd2iƍׯ>6mjxKK*,,TVm6c-'''_aaaܹ~w-[L&M2.]TNk.رTו,XG{ڶm\]]uz |PfҸqrR_5|I }̙co
8bܹ:w\jyƍCM6?wyG?l'88X_}sM7V^~mܸQ?p999([zꩧ'xB~߯oT,M0Ao]^{&Iw}6l`aÆ) @ {駟ꡇ 23j(m߾]k֬u!kȐ!QΝk]+??ߢjb5"H @! Vpշ !*z" Vv'PzW0V=vZIRXXx ̜9SonunɢJׯWff\]]*jĉFΕ+777I?l?$ׯT-iiiںu$iFՕ:t |Pzjuk֬$=V' SO=e+V$M0J
?WZb5iD3f(aV_+11v'O\ y嗭6rH V{''' :#G()R"??l2lRϟ?I&駟~V^m9ٳoJF
UJzas6mڨYfJHHPllB]<{%I ;vThh]qq V(((H JII1id0̚8fI lի֭[CP*((0 m ͞=K<%S }SϽz:JYUԲe_ȑ#JKKSaaaΜ9Zݻ8/Ew}uNNNG }d}Vt5X]WEݻwc^VիW[n믍K._Nj/ &gdd
+ j .3KClaK箺խ[7;? PrG> N,V7PU\pAԬY2 SBB*jdT2͛=|}}m+Bd WVTTT+%%EEEEnggϞ5'%%+ޜ
W\_U{Μ9c,R@@QwyIRvv7%lgrww}YiI u;C>z-IRdd }Q{qy 5\ PxV}}[]! 8B^ϗ ;X+K@ edϔcV֥KG э7XG 3g(<<\=ܣƍ/]dÕ"V۔uV4}'qqqʒO |uSJFrT#kJ֗5-%SѕLW7n,WW
WVsqq17nl$$$lkkݕSWϧO68\z z!+#p]MyyۗMRRmnwka̘1:uꔚ5k&M~3}ڸߡC۔Ք7%J?+**ҷ~kul7|Sf)**2cMv-|Z~K+V$ZJWpL ~ tyZ4k "A˗/ͫ*I
TϞ=+ ?,???矯F -IꫯW_M@@z-I7orrrJmO?iÆ
CZ$}':vX.\вel ̘1+Veƚb#dM>̶o駟JԬYʵ[-۵k;IGk^ułV2
joKvԶm[IҬY,,W_i߾}J?G s='I2ez衁$驧҉'* (nfI GMֹsg
}sNl߾}ٳ͑$iĉ[ =zhJNN6֧i˖->|nZɓ%IsO<~XaEiF-Z$5J?WDDRSS˵s7
Ԋ~z=C.BZm={G 1OO?$ig5g Iҷ~ z(*,,ԿoK.4h.??_CUNNkƺzK7|222#Ȯ.ǤtM7i
utH Ag jPM} :&s
&sM
eoj䤠 jϞ=
Szz hL"___C&MҼyJ?""B|O2c={^~e SJKK3* @tIlRt [oxxrJ1b]RR |Ac:$)++K傝M6Y[v IG+((0Fnjnݪ6mڔ'66Vz2F ˗#4Y{\gϞ?X\\\lQ֪U++It钞}Y-YX#WWW-::Z#GԵ^'OZcI/D^^^.y{{>3FH۵k"""111ѣ$iӦMzUXX()??_ԢE ϻ'Ol6QFRaaSBܹSM4h 3f{%i
8qq(''8n}jڴ }]_bÇoWnnM_걯짊fdd_S}^3ofUQ鎎 *xOMO7F ?W\@" RuB j+j: `:uڵk %''+>>^F!vܩ+V(""BRHH ]vY-_|Q?ƌVZI
Ժuk
:T7n700P111裏Էo_);;[^^^رN7xTH >|Xcǎu]|}1c~WԩS' :tHO<5k" W\\UGgjڴ٣իWk
5
տ-ZHw.Y/֞={裏E*,,lV۶m5zhcĊ_ݻwW@@#G Խ[ 8P{!٬-[jĉ:x`?7nnݪ>}INNNk/U|UQ?bcccyɑN-\P(/J{RWtM7魷ޒ$ڵkWb #`@yGʌ$L#`^ WJ. W. Z~_H> @-X ? Ge2d2Ծ}{G` eTqΝ; 긪 QS>k PQ t)+E߭Ç5c *\p
@VUyyP3j*Gx7
N` -ɤoF4c L&ɓ'/]?Psf͚ .sL&EEEPo:u꤀ LRm 9qƩm۶n
2D6lPqqc?^>Zl)kȐ!:zh}Q\\uiРAj֬ ;jʔ)_Oun6k( ;v7o.wwwiȑ:~xJS =(777hڲeU۷O#GTV%???mVF֭[ \ Pyzz*88X)))*,,|||,qvv$kРAF|}}u9_^ׯפI4o}<>P[C h̅*.Ȼ-;;ۮ[nn " "W
@W\<y@CX@
ٷo6l IzwOBBBb
kÆ
zW4byxxOVV6oެ˚4i"IJMMՔ)S$IC l1:z^zٌ388X۷o$E={oՙ3gfYj;L&mܸQgӦM- o^zIjժmf+,,L;we˖iҤIFۥKԩSrss_m1B]wݥ ;v*҂ tQu]۶m_?j֬Y7n\\.d믚?$^М9s,﯁j6/].tj-5aю r @muʵj uW
! @M*Oϵ a,]V'x63gΔty:۷[榛n(땙)WWW-XԄ1qDJ}$X JWʖ4mݺU4ydJ :tЃ>(IZzź5kH ~a詧y+VH&L`Q|uAOIII VRqq4i3f04{l @@. #`5$66VԣGiF͚5SBBbccZu1+IرBCC;EAAAJHHPJJH$al6Kzis^ziݺu:t
ꪂ{{ٳK-OHHP||ˣ8;;GԇFߔs^RV^/&L`+8܊+ j
~d܂rc 9 5\ P[\v8kFn uݕy=y jVyr}{|\ y@FPǔ5Z=SV#YY.]2 9rD7xc y TΜ9Sv%ʖI2 7k{+OJֹq1K
l*d? ډ,89]f:u$ڵKV9zQ4tW8.]Hbccuܹ
:w,In3lΝ6۱c$[o5
Ԯ];IӖ?VSV$%}vU= ?($)--!CH<*˭nꫯJճg
OEEEzmUѣGK+}Wj{K͛ROڰa$iС"##%I| ;V l2 3f$iŊ:p@X=b9;;+99YӧO/- (7,r:wKƏ ŋEG3f>I̙3Q5w\Iڵk }8p222*[
nݺl6k7ogϞ
5evf͒ ?{$뫯R~TTT뮻NcǎhO+,,Lӧvi۷O={9Ҙ$M8QrԣG-^X4mٲEÇwmѶUV]QQQe2{lF<==f,KMMU@@$ɓjٲ$ĉ
pkʕ1bź$=o%I&I***RVV$iڴiEkj*((ty$cj͛k֭jӦMxbccիW/c1///999)++KZ|1RuY
>>ruu5O#Y!5nܽJ'! @ ~~D> jSn@> د6] P50PEZn]vi
Rrr "IGڹsVX*++K!!!H}UPP奎 ;jԩz7JÇ5vX]wuϗڷo3f_~Z|%I:uҡCOYf*** qũseܴiSٳGWր (<<\עE{RmxbٳG>ZhBfmVG6O #`@-6?[ u\ uMmPWs| Pզ܀| _m2
N` ( ;Q v ~[~~~ 8 - v D ؉, X `'
N` ( ;Q v fIRFF# n+y/-yo+ u5 *
HNN$5o P?dffa U U| T>@ 7$:u} k222Լys>}Z~~~^k}]sQ~6TӦM)A.аҰ7lk
*sQ[ Tr4ڍMMN5v$ g( ;Q O.wwwGR59u͡kFC?<
+q>8PYC ΟA?^<7MS{T-l6;: D ؉, X @-" o[oGW'h˖-5k |A]{2L2L*>Ο?'n7n[˗/lP$''kʕzԶm[y{{]aaa4h>ӫs4c
0@7x4i"WWW5iD]v믿2m7|x-1Leny]~jc5vXlR
RaÆ|$Usa9jV?**\
Ǐ C> {S,$,E>Pj`}.*Ԇx6>S~5{U qIsxxYYnݡCsJJìveٟoӧOjXs&M6>>>f{W$bY߾}VsW̸qJŲ@{ܶѣG }m uŬ\,dy۽{_~_???ȑ#5*sڍ a?}t$k
'NǂA> @YIPY55|z8u>~UU {qes\;Wgp\ #`:yBCC}vegg+''Gk֬ 8 {ѡ 5ҽޫɓ'k
)WtUrrnF߿_ŋꪭ[瞫PGsz+77WYYY:qF -IڲeƎ k-|tYT*##CZjA)==ݢ-5j(鮻*s[k5o\6ow}w6'N7k:vӕW_}UrJ͛7 8 w=n Gm}Ktҥ׆ 5| Uk @r`}.*9ΑOp_MuSN hF5?XcDXwZvז?X^~e$?(70K2;;; ;VU!Y\?vX=u:uV>#u[hYG 5SVy]q%Vc=fd 1ZO
UW#saڏ a_tzCC>"@YIPY55|8u>~U {qWOp_MuS> j*IR =V$2D-[$gggۖ}G.]ҿoS_ѣ%"I8ޝwi?s:m8qB/4i ^u{뚓
6H~i_$eddhӦM58 a| @u"'Ae ]VUr
U^u>S~5{MP 5,''G}$o߾V1Lӧ$i۶m5[Cr1:uJ<<<.]2sWoָu97feggk
*s[뚵g JjӦߜC } hX V$@! *:X ˪i{qU~)=(vKnfەKLLTJJJ/<=.&&Ƹ-9N~~N<ŋkذaVZ6sNSÇוsEuQ>>>_=c#Wy}*:pDEoÇu7K>>>4f 8pѡ
*S**ה|v`}.U
ٯ&~)=(vY~flnw+۠jTyPVVVUWiْ[7ps<<su=(iݕlPu5l0;wN ZxzBBB ,ooocY =h"hb[Ǝ tEEE/KiڴO~IyyyJII1?ٳ$iʕz-ڕ]Y}}לCz@kZnsرcSrruVuQfYz-G*5@NyMpN}^wY5o}k {MQ *g_|!IZdڵkꟓ'O*11QYYY:ϟszW ^G /T5aGSw}RvrvvV.]uV
8P{~sd@z `ͣ>ɓ'믗$Mw٣o]tG
P[ 7>>Q{{FG 0___~NN\weT
1i$?P.\QFچZ\s&NZ&I3g4>JwE?^= ťmKn+ׅ BYʺ Pڀxxx7ސ$eeeiΝ5FN{F^?\oSe]VM*ڱ͑gpݔ X PÚ6mjOHHݕlQO>>> W] /Cϟ?_=8GΝխ[7IXι]~SNUrr|Ixʲyuhժ%IRSSk}Iׅ غ P>뮻6} 9 jyeSe_wYխ~o=jT\;9 jX6mt_~]ɺ5nܸFbkHnf~ymV{Luɓ5o
u j9AN@^s@T]Vux:Nຩ9`<㒤]vi߾}'F2|0LF߮YF'O,͒%K%ggg=5 a4i$aa|\tIfmvܩ$EDD9/&&Ffmƶ%-Zd,㼮fxMqrr_Wc ,j?g I?4^C
Weo7lW{mK/${[a2)P]IPY5C>PTu~U {qW'vls\7` ԸB-bdn֬yǎ fltyݺuf???$s߾} iݐbxqkYy333-ڥCBB̒m۶5ƚf9??{̒O?# V3yd$$ *ԖsbN8a[˖-3bcݩS̳g6{{{%7nl>wE{1}t㜷bN8aKח.]2{m?6fl2Ϙ1l2̒s̩UC
We
=LL{54>}X^PP`ޱco7^ F> @yk @r`}.*9'vSݿgpݔX 'N0o^^^f :SRR fpV /666ܤIc___}g˫U}d.6oR/'NXnnn@CGɭe˖游8ܮ}`19+絻900n|ȑB/^^^ƶfgggW~!Pp5LUq=#Gmywe h;99MjA> {S~Uj
V~ک=ꘂ $<<\ ҫoY&Iرϟ ~A5rt^ǎ ua=jݺ
nݺ>Ж-[0 ˼eee|5mT|ƍN:)00P*..V-Կ-_\ V sfp^_pp}]=#j۶&WWWx5j٣?P...Vѯ_? :tHcƌQxxԨQ#Kׯׇ ~(TÏr8 P>0r-? 믿^JKKnV=3:x^uG*F> {S^5G>P;Uu>~Uj
V~ک \7d6_eO U v D ؉, X `'
N` ( ;Q v D ؉, PEEEd2v%k띜nM/N:e/^믿nݺ)00P
M7ݤkΜ9ڻwou?| :r @> fA T(͘1C \涉Y$]tIIII6>>>Zvg;woRRRervvVFFŶb T? 6r @> 1 h0˼Y3i$cŋe˖WYYYO:A)%%E,+--M[nոqԨQz G> @F. T
*Wcǎ
%IYYY6ֿʒ 9Rzw}Zxj8z P h ? >.R7 |17< 4l | @u0p cQQQ1c$UhutmqFu#Ghu:ub}ݧ>L T
P $999G_tjԨQjҤͶmڴQ6mgiيm/k8 h a, PUZO*j.\F))) F 4l | @u a8p$)55U/^tpD а ~ J*$wwwF | \ X UoUNNNdeeiƍ-[* " 5| \
X UV-4~xرCƺ [N]tQ||$iĉ
T 6r @> 4<. >quuUrr/^ŋK|}}e6elɓ'kܸq
T 6r @> 4<` T_z'}v ;/ҥKjԨZjnݺiĈj Հ| \
l6 EN *
N` ( ;Q v D ؉, X `'
N` ( ;Q v D ؉, X `'
N` ( ;Q v D ؉, X `'
N` ( ;Q v D ؉, d2)&&Bc #ɤ*jwE=쳺;nfΝ-[]7|֮]ks /}UXX yǗٳg5x`qz'Q%izGtud2g]AA&Oy{{W^:vX6oެ;C
7 [bzyxx[o_|Qccc5rHiFNNN_ZjG =zT<5j$ooouA۷op,9rd2hѢ*, z$OOOiF-IOOѣոqcꡇ ҹsJk뮻kVs̑lb#Fo6dRRR$ɓ2LZ~}~5JZZn_|Q٥-..
u7]!!!zGm{ӦM2L[hL&Sk~ǎ 2d奶mjy*,,,?\z<<?J|@D>?J|@D>?J|câ6Kg(
XbQX@Qc&b5&1Ԁ=Ʈ1`GޤuwqAw]{9s)gϨv""b+66ӦM'O`˖-(Uիy0qDc߾}DϞ=o>ΥKsNt {
q
4h Rtxyyشi/_#G/z >|]6l~7̚5 ;wDjj*Zh8NHHQZ5ڵ /ƵkкukHlق BF퍿KϜ9SNN:([:=ssM|'022
{nyeݺuX~=/^sΡG kn={Fӧ_~ <*|}}+V`ƍuڶm Ν;w߿ #Fɓ`>}8w\P~w}bĈ_ѡCܹs1tPaŰnJJ
F ;;]iPT)i+WDBB~ 0 `>|@71 x?>|@p@i)SB! ą CrrN߿/ m۶Lǭ*ŨQT?s&*u =z$d2?Ti&!DXXTv@?>u!X} IDAT;;;!Cy+WʢBsΕ (ʗ//?S 'O*W,z쩲Fm+nѾ}{:=sӴiSѽ{\ѣSŖ{7Z={V G L&RW_}%EjjT6~xaee%^znV իk&/^P+۸q .^( ;vLׯIg}u)J#T?sL!TnZ4nX^Ϟ=EժUM$`>|a>P|1 |a>P|1 |EDTH[ ;v nnn055 ݻ+VThժJ* 4l Vi[9T+Wжm[(
`ݺujq̘1033C.] :B̟? +W1*TE鴞/^@.]P(YfLWEQF011e 'ODƍajj%K" ,_< [nddcLzzoݽ{ hݺJ_ ```RxTv!U*kժlllpA Cꤥ M6EVV!++ ݺulllкukiy@1!ɤ2KKK w n߾ݻ,G 8~گ"%ܞ@pՃP^= ӧ1lذ\Kr`sr899aѢE1bʕ+7n
ڵt|)d2̙3} lmmann$$$
4y}Ha$ QT~ӹsgTPAe[lL&Ŷj˫]6 ǿ<==Qf7{],X K,ۊXt)r0סd*C+0V[˗/q1cQ:OٿDٽ{7̙1%Kծ]Bi(Tˀ}.44TK9sE1 `>4c>| 0 `>| 0 `>Ptv""*D=z4&N7ݻի|}}QfMرuEgZ߿ :tc4iڵ ^zuؽ{7j
JgL4 } ;PB_ÇcÁcbŊo\ǯ
+VΝ;ѻwoL8Qm4|ݻ7 :֭[ҥKhժͱm6̝;C۶m ܹ 0k,;wΝC^>^z 066V)WϹ _wmxUVPJz2 UTAXX ^z0a;&MݻwvZ\^RԆ&Z< _~qqqw &LڵkI&R[ Z*p}ʏnϻwk^:vڅ@t 111:-W9\rbb"ԩ-[WK,v킯/t;wDJ7n BԾ}{XXX`ؽ{7+ /_FZZ
CCCcر* ʗk@uIJJ£GַJ*dR=!:uꄫWbʕXlv) J =g}(lذ+V˗eiҥ
ڵk1g ر iń [oҥKKW$&&bĈ }4hڴiӧOP=/TR#FLMMѦMܾ}[mÇo߾pww:BlY?@2es[=zۑ) !___| 0 `>|@`[DDdʖ.]* cJe111B___,^XL.Zn2ի l2,11Qr1}t!tҢO>*mGeH;wL24B ;VKjRXSeʔQزeJ=oooQlY&)iݻw` 6p\\d*
!D@@ 6mڤ,%J.-J*o߾hժ>==]ԭ[Wԯ__===m!KV>oann. VZ"""Ba@<{Le . ̙3ZFې¯{m6@9>!={ lllČ3ğ)Ǝ +d2XbEbBÇ ggg:ժURYFF(_ѣYag_yf@"?sFFFbܸqR-[
/// "\\\B<~X 7oVP(̙3B 8p@mXꄄaii
F )
BCCL&K, ˫ j*!DhhNaxIеɧ~2ߤI\.RΝ; {nѦMQR%utҢSN*L4h@߿_ܹSTZU+WNH+X3pE
NL0A*t>ٵk ߿/ڵk
8qBm%KJ*ICOr߾}[(
1|p8wJH@lܸQ5(b>|@o|0 0 `>| 0 `>PEDTJ.իK+W hٲTfeeRJѣG C2e``` CCCis8\Bggg<~XjӧVk* ; AFFjٲ%"""'O8ڷoԩSԩ
UJ.YXXwޘ;w. :[NzSXn
Ek``Kìi_}[KnΞ=>}`?m6dee}HIIyglO777/}Caĉh̙֬/3gS[o8GΝqy @Ŋaaac֭Shٲ%,Xf͚aرohѢw9_pVVVO2333hBީSмysHeUTqޡC!v
!]9+99YN߾}닣GG ˗LBFFvÇi& ٿ1bMq^]
0 7n 8x ^x|@4 a>|@@60 Pb>|@Q!Ryh*бcG>}?qmVM+)[T):vvv*_| !J, CCC|Pik_| 033Sr~sߵEnݺh lll0zhL> 4o ~ʕ+Gkkk }LLJ2 dYvmbȐ!]
6 ͛7]|2֯_/@=м_Pv{V\G\\ akk ;=/\z7o\Ex||
lgg' #8z(ѧO S%zΝ;:z4$''K={毟r>LLLT%As |HMME:uШQ#m ! T;oX[[z*ŋCOO={Dll,bcc,UV N>CŊu?={bΝHKKCFFoߎ={J'O 777|Rʛq甖ooo*_Nz+jdc>|@r o|.|@ 7c>Pdc>|@r hEDTܹsW\
ѿxxx^zꑮxB*mll p\pA kk?gb!666j*W J( *Uj//jժR%Kݻwޖ*Tޤg6mpIDGGcÆ
tS 9mIjjjb&==]sz1ҠA :tطo ^xΝ; ժUu&K*Up-!Tg
W9_~L8^||<^zBsVRR www\xW֘1cuVڵK2}LΝ;5yf_bÆ
x1 8L711Q d?=66 *,, 'Oƙ3gpX[[KӫW/\t ʲ*VCCC jǘ.߇ ]1 |@^1 `>P 0 `>+٘0 +t| y|@^Q1Ls9s&m9::vR)߾}{/ P^=y~ҥ|M6ݻUz> =zhڴ)mM)HKF5```˗WeCBBгgO 0 &MF۶mq5RvIewŘ1cwa߾}x)Ə [n
===رC*APPqey Ώ ?ߺu ׯ_s}!33Sz}vd2ԯ_ /4r9j֬) =1118t #FRjΜ9Xh֬Y,? 7oDDDTǏ nݺ q!88Xr!88 ;vy^zA.
!v|SR%ԯ_ 7o͛QV-/6/V | ѠA!Co>ٳGe+Yfjצ@TZU:O)S0 `>+0 b> ]1 `>P 0 `>+0 y&Pի
*bȐ!իWQ}͛7ʕ+rʩzj@DFF.? ƌ# 9"F - ,o1c8z8xXxԩTgʔ)BPK2enj# e˖iOŋHxyyUV [[[ѠA!"33SXYYvډӧO .7o|!ĶmĶmDn݄><<\ao`qFg '''Sο+,--E5ę3gĹsם;wziiiFf͚b߾}"00P899Ku233E&M W2LRp)^[[[ѦM}NVVV?G L2"66Vxb@ 6L =zTlٲEԨQCى/_J6m$d2|
3gbĉ*غuػwYpwwRO(
# ?./^,ļy:YYYN:L2bݺubiӦQ/_^Caee%ի'v%6o,*V(ʕ+'z Dҥ. <(-[&Dnts_?O_Z8|5k000~4ߡC e!)))J*Yf"++K!ƍ ѻwo}ܹsŋR[\rnݺbb֭UHDŽ.ݺu զ-_\ SNAAAbĈY WpBP( ߸\cb̙oնS\\TԩSB___ | u E)1 `>|%/\%QŒ%KD
\)!R(⯿uFFFUٳG&|||066gϞi=l"W.5k@)66V+++aff&t"j87~ZzJ >\ !K=zPi
@,ZHZ.!wСP(\tU| u O&k$"""bԩ?> ;w.--
ժUç~d2̛7cƌyG}<+Wƍ J
AfpԫW!"a>Pp=t|0 "*z@a>P ҈믿"++ -[vh 7nիزe ~>DDT-00 (Z;EDD ,deeiL #8dff"A5
V!@ffzzz{OQ^̙3 www 8pڠMС"##燀 Q`>P8(_G( :u4/z0 xc>P8(_G(@#kj OOG\rxE-Xf
u)S0uP>)Qk֬yOOO8qB\r/ DQG>,G>0 ( |||x DQG>,a,""˗Z=FqqRSSNX~(ܿ_ҥKt1IHH[N/YEB֭[HHH:
FFF1 DQG>,G>0 ( |||x DQG>,a,"""""""""""""""""|*+%"""""""""""""""""*#}
d2!Ϟ=FΘ;wʳ!ɰ}mԩ8{l3332===!^aaa*п ]vųgիW!ޅXdy{dd$
8a߾}pww *WիWƭ[7ߠZj(_< v2n86ccc1l0.]&&&X",XAEDEOA NdR2 Gtt4amm
Lݻwj .Dpp0֭[RJյC"""RQsYz1̛7-BÆ
Qt&M{aDʕ ;$""*? =cǎE PdIܹsZ[{n\x_9sNBÆ
ܹs1e̙3nnnXlZnWB
txyy 6mڄd3_|ߟk MqxbHDDG)))055U+OMM!VVV~!yPD lٲFFFhѢ"##1sL :THqfeeO>Dsȑ#8rZn
puuEժUsNt}Z丹 G"khذ!VX h֬ɓ'V\Ϟ=1dd2M4Rnݺ(Y̙\~
@k J\DD[a,rUXXбcGuHKKr?""*>=-(G 6l[3zWyJXX>Sʚdff"++ |""ܺu B 0@_#_.sCh +̞=Gȑ# ~)*WKmm߾ 7oDhhԁ^^^ѠAq7ǥ>ΝC֭aaasss4lG GGG# %K)7n'O? ۷o+мysܽ{W^jj*{TPpttZ;9izTL&Ü9s0vXۣTR GGo?X0777L28q"2336
r
ڶm B[Nm 8p M4\.5<==q 'NT7d_^or!tFFFRY =sΩMJJBaii 5
gϞ! *T)\\\0aJRS~YYYXp!V
cccۣ[nSƍhڴ)r9jԨl^ :tVVVhժTZj*ug̘{{{K.xZ{B̟? +W1*TE
E.]`cc\wwwlY+5Jz\Zk.v~7+Wr -Z;w4]Mt`tMG
㗦Mggg|ZMJ(v,Ԯ]@py# @Ɉ#`mmǏ KegΜ>V\YhqQP'Oo߾)T~I3f\r066F*UTN{A*U`bb
c:p 6lSSSbHJJz:tΒ @]aaa!]֭[MFo@~S#F߿?K.yf@.C333T^?dرcN:%^ 5jpww1ۇ$|7puu\.Gr0h |L'/^NNN077?RSSqU4i
47T%Gz1w;;;|Mr"(ѼXZjGR^ CBBPvmZv%X#l{zzbС }}} 7n܀
,--ѵkW<|PIZSxx8d2.]C&I+skƸvNwϝ;'NDR`eeB ?~ jՂZhG ̛ &ƨZ*6mڤR͛h J(\WWWZוt PbEGjvڵkqM{ltw}Yǫ]tQ!^ժU+ب '~&p,""zΜ9͛O>+++\xQm{;w.dj
gϞEݺu^yaΜ9ĨQлwoj>>>?1a|'Ν;O?O>VRTcয়~>
.\#GĂ *3gJz0j(oGaxѳgOt 6mΜ9'OvǺu0}tDtt4٣r4?#TREJ*d SȄ кuklݺ/_ɓadd$˗/acc oԩS3^@vFaС/ ժUK+Wbȑhժp$&&J^KOOG^0l0L4 s΅ UoժU_?I&a̘1hٲ% =-o1qD4lgϞرcajjA QFprr%K`ooGl^pa̜9UTuݻwKٿ?+G tZG( ywݻHOOW۟zXX# dʔ)ptt1sL|gׯ uV N>[NիW˗/ﾃ>(Y$:v#G@P- `8r뇠 $''^^^8p`EDE_AϣШQ# ̙3QB* ڿ[OѸqc߿
Bzz: 9ڶm o___ӦMóg0n8`˖-o]hW_wصko߾X"Ұyf|g~cyt?Ə _~ӦMC:uyf7N%W\˗/1|aذa0` ީd_FxݩS`aa///]tF~QUʯ$)qqrѣ@o۶===UF9{,ѭ[7dddH-["%%[P(,í[chRJ !^ZzkHNNB`ŨVLMMahh^z!##˵?B#I大-[J˕+SS7rڴi~
___tҘ>}zxmӱc d~!
\?SNگ}}}q$%%!33.]zcLywߧs߰a"""xbܾ}M6 ۧAXn1ɓ' [?QFaŊxGs @ʕvZh
e~0a[X|9
=."*
|~q4o\ϟGzzkLdd$n߾-СJ~zLoƃн{w<<>>* ,YRz +g ʲ 飫%JaÆjׯGڵaffCCC4m Զ#*W ===4o\kHu||?8b>PJ.իK_ SNy氱ʪTwww뾶{R敛TrUV*+WҥK쬌Q
`cc͛+Z+W%JǏڵk9Q1uc;7ч 蝊AVVJ.k 3sCttJ{e+ ;88hA vvv:|@vbjhh(\\\ Ҵ 9@ {˗v\k-])c{iiiHNNVI@Sne'ŋcԩك˖-M(h?r255UnS])
o .]ʬն*r]m /_B%K'$**njgagg!bcc btw jSx4=!dIɭ}ظq#mۆ5j~ݑW:6kL^- @m={"--
KF.]L\DT4\kٳgbZL*1rdff5&9&gBBZn `
8u.\ wwwMAA9_ϏJSk.
4֭[_a])1#E8q"\\\PJ|?j8a>]AyYR~111IZW;b|6|%|g UbȐ!prrBzp|'ׯ5qFjju,&&2L߽>L }جOjccc/^?}oaӦMׯ_ݜk۶-كt,00VVVhܸJ]/0o\.uNJIIQK7nܨLCCC!͛7L&{((IIIؿ?ׯ/m1118~Tvm\rڵO988h9)G Ҹޏoߎq*m۶o۶
k׆B>ԩƘ
r͚5S[G^ PEƍ/ؾ}; ?~yj>|8fϞ}N[lC###nZe Gҏ? .`֭={6f͚ȍ>l' >|qy
45f֭(U
`߾}*1URwƘ@r] JrY繭5kT۱c̅]5GRC1c dddDc>աRYhh[ʕM*nݺׯ6: hٲ%"##addq_y}نq 2Q ^qưPyӱsN?k\v
TvqDEE{[MTq,""z̙͛e˖amm˗/dɒ@ѠAś .]gϞa yZV˖-Ѯ];ݻhذ!}v vUVaСܹ3Ξ=v0?Ǐ }}}ܻw{ ; ujK&aٳ'|||зo_ܹs_>>sxb_ :$Ǿ}`nnjժZj o7nDϞ=_ƍ7o fΜݻׯ z˗/c9r42SVO?FʕazժUg |P(puuEʕ1h |F-`ԩ*ͫSNayFrS,X *_5j///`011ĉ&=>N__ ƍaggVZ!((*ˬ\2 >}oEÆ
۷o#88w L2GӦMw/wI 5jRRR
6ٳسg'SNs%_^%wοL} 4i<==_{ƦMc2,X .` ۷oFG;qмysxxx9::J7⽼мysC{믿b᰷2e
7n^z Ə ^zbŊ ׯcʔ) PfMԨQ{A߾}qeJ\DT|ȑXn >3L4 *T{pm̝;%KСC1oi&,] qơ~ܹ3kܻwWyL&
_ )) ۷Bp̚5Km
:g'|333 2ƍÓ'O0eʔ| 6664h̙SSSԩS7oݻwUd.j
C ѨQ# L?
ahh(텯B1eP(RmrJkN?^nllVVVLtEP" >z666DTRE,YD)~)rqq+VPiΝrXԭ[W_AAAC(
P(DѣElll\P9&jҡCDՅpss TCu?RSSСC~~~bT/R[ SNJ^Z 1zhakk+E]_% m۶I4ɚ29rW^/R
SSSacc#Zn-\דC|`!xj֬) ;vF1&&Fmۿ~]h'kVZIȺt"U垔ZM4ݏҔs MkچT1m4"h֬XnBϟ{
*cccQT)#n߾zk.ﻚD =D% kXNjz/++K̚5K8::
cccѰaCqY ?~,t"̄F~&Ȅ%"""*V l۶
>>> G (ܺu ?nݺ|0WQ۷M6av""""""""""""""*$ (8Q>Q>Q>v DD-++ O>9d2YaCDDTl !ҥKCO愹 Q( |0 ""QS899vDDD G ѱs""Ur QAc>@DDD+ `,"" @¢!""*$][ DDD ""|w>XDDEr(a ~"""* m~DDD ""|W>P zLDDDDDDDDDDDDDDDDDTQ>Q>Q>Q>Q>Q>Q>Q>v DDj7vс ;""\8""ST܂(o>|01!"w#`;`;`;`;`;`;`Cxx8d2d2YLrH{!!!Һ%UV. 2 2?<&/^zzzJ,DDDDDDDDDDDDDDDւYpxzzi[[[ @ɒ% 9"
HJJ*䈈}(֬YiӦ!$$B(,--ꊊ+H{r puu-
"""cǎhpuuCK H1c
; zOW^-P
qDDDT[
GW*Ϟ=ԩSٳB>T|a1퍰0 ?~@kРV 'gFXXfϞ]ءQ1<>'N͛7 ;""k $DDDſJCTgϞaڴiEDD wZf
d2N8 6md2+<<\?͛7Gɒ%all2eʠ[n>2 SNEzz:,Xz
2LmP 2ժU9 =z`ǎ Һϟc(_|(C&_7nD&M`nnKKK4l+:?૯ r9憉'˗{] <...055ԩ ~:-_y=VNػw/+ԫW022BR͛7q>y~Ν;:gZZ~4k %KѩS' :tH|8$$/^QPre\"""""""""""""""8HOOBJ }}} @\\ :w,u ׇ9={۷c3f ͛uy^'Ξ= u;w.&L u211\.ǝ;wpm"&&VVVjyxr9 ŋ ġCpIK%Kʕ+y&W1 ;v 99fffҥK]: >ÇqI899u'N@Ν P( /~wݻM6F^o7oƼyul ([Nm !п^@v' +++\x7allk?#Ə /Wr qnܸիW]ڼ[nE߾}=,9p\r; 9U>X:9s:u[XX
BPPvڅ-[@OO_˗ѲeK >a͚5عs'~\h߾4̽L&?{b4h/_;wG x9LLL`hh2GEDD7n 3f """T^NNN #$$FFFXd O" 0|XB-[ׯcՈGtt4"## X|9ƍ,tW\AJJ
jdwrqq D =zСCUWժU ^֥KSn1tP!лwoܹs^B\\ qE|(URzZj8}4'O111ԩPΟ?/uYt) *dooUV 0i$zJc;شiajj
(Qlllc zݻwVZҼr Z-[`aa};;; =z e˖Xr% ԩSjӧiz'O @NCP(zjTXQP(Pn]hNY kkk ?~ M4}8v,,, ٳgkm'?Hhٲ% `ݺu 055E]ujիW6mjҥ(Y$ SLرcQ^1c oߎ & @(nuő#GPn]<~'NDzz:*U ԬY :000ݻws|W +oN
4+вeKcÆ
êU K,Qkx!pajJQF8vX\p!ၠ xxxHYZZbȑ3c dddhlGOOǎ Cͥ+W~V]jj*U^DDD`.@DDDK9Lyfff033>DDDov
8::/Xg /_j zСi۷oGBB
p\;qh3zhJNm۶ ƍ*z===<~Xhc#5kS #!***kN[
$u
lٲEk[FRvfSvzr$Ν;\6
r^5Qv5jvx%.]N>}ɃٳgRz)G#""s"""b>@DDD> Eŋ ͚5UL2*_sԦם={ Pn]888+Ά
j,700- Hy @ceʎZX"Tt4lsիW모r)MZj 56ҕ7U 5s*U%֭qڙ3g <|2 tׯ_Gzz떛Y'̛7 (U T:
---Mysi)kAzlߴ}k#ѣ_] 1 xn3x!118}taDDD (@/^ JGGGD*do^M d?;v/"u@z* ex{{7n ƍ\ƍK.8&9gnA^x˫6҅\.֬Y_yfdff^갤{<}@ =3deiegdd ::vvvo\Λv:۷ѢE Ur VVVR'ϟ :#u5
dcWm䗱8D"""0 """DDD|22Ї
h|jQA(#`}}}\DڵK*W~UN TV-Om#,, ;vW_}5j %%ǎ _*UQEZHHʯ/u?(hQ|}}!x+<<ŖܶۮS~c+W۶mCTT DDDɓ'R]`!p:Ş_Ns """""""""""""""vR6s M3:= {
233i& y*'###t+Wč7+V =Nܞ}9t '''deeaƍ /_m\46mnwy9y6#q7oF]bVF:SnǛ> ps"""""""""""""""Z,cǴvS^= @pp04 :^ԯ_?14n pE<{,-eǠcǎ!""B _|E,D8p Ν r
8_N2Ǐ Zرcr4=~}dݻ7ufSww< =zwj K.i֤I K_թSG:6t0442m)dkXG>322s6rI.ܷoNEu jޣG #L< PdIl21t
ȑ# ghٲ%J.LlܸQ4ԦM橭\|Ond2|}} +W8BӧOrJ @Ϟ=nSvfqEi$$֪U+X[[ OΏ?Ӻu+++cԨQWYYY*ǀ Crr<]Î ; 6dii)ڵkj0c )m۶֭[j_x+Vh UVʕ+Z@tttӉ^Wh jԨ 8xLJ5h >>> C:DDD`ضmN2&&&y?# 00z4=99 @N DOOOjժUؽ{7 O>ynk˖-hҤ V\{I噙8rƍ hԨԹM&L +++DGGe˖c ̙3hٲ%bccacc#_XT"\5x`> ѣvZ!%%ÇǡCp!3GzzTW&[N _ }ӧ իW '''۷fffZ:aٲe͛7Q^=r ٝ
٣u; ڵ ZBxx8Zl \===$&& 4RpaСC1l0XZZ"++KseJ
bSG%sE""b8"""*H-}05Wj:u>x6 ѱcG"**
< {ǏcժU9aoo cyo q5 0@ꀑ;w Xj֬ZjIu떯Ѽ:vu֡_~pww%`nn
`y&Tv=<<ѣGjժʂUVŘ1cZdF
ٳ'
y ֭[uO>)222PN X6mzc
[0f033Cz0tP =zTc}}}qM 8+VDjj* PV-L6
V˯'O}BFFJ,~ҥKhѢE˭W _/eʔAFF,--˗/A_ti>}7oFǎ d\rС/^'O6"""""""""""""""L("*B֬Y~ QL:ӦMBBBt/>> &9:R6%"*rԸwCwQqU\sOuE>>; (;v"dd*j$*j/^,ӦM+p=1( 4155JYɒ% )7S(jM!ECDDDDDDDDDDDDDDD ;`Q __ҥ Ξ=yv܉ƍ>>
; """""""""""""""z ϟiw
ǃ > !!! G k¢ ""B\""E """""""""""""""""*((((((
; ""R05Ri!+h}O.@D;|>1s|>DS8;;VVVO0w\$&&j/<<\vz]r ɰfcbbߢbŊ011uK.imS ~@:u`aaSSS`w^чà HOOBJ }}} y:w,uׇ9={۷c3f ͛uy^'Ξ= C&ԙ;w.&L u211\.ǝ;wpm"&&VVVjyxr9 ŋ ġCpIרQ#r r 1118<Ο?u!88JzFͣpxzz ###$''cػw/mۖo6m !ܹ;w`ظq#|||
d]XD/"""иqc 1crrr !!!022%K <} cŊZl2\~ WF||< 777 1n8deecǎr
RRRzzO } .\$$&&ѣppp@||< q : 00Ϟ=CRR;w/
Mff&u [nERRpM4l~~~j;!! :tǏQL 8p IIIիW' 55zµk
x͈T$%%!%%C!w$-#)*7x^ #`cǎ K⫯cժUÎ ;0i$DDݻ :tJ( 1|cǎ 6mR K.UVhժ8pQ вeK\ ;vĩSc8::̷w^LMM
bŊؽ{7 >|e˾q{jǎ x" `۶mhѢ4Zj8|0ܿ 8|0jԨ!MswwGPP 'bZJMMEjj>>>>Q\jXd6(Їjjexq/ _j3}t ˗/qQuW*۷#!!Xpڣ u1zhUNm۶ ƍyjL2pwwgϞsLٲe I&*r9|̺vJ\jСCٳaii)ǁ 1 """DDD v"*DQ5kUVE2eT꿮I&ZTn]888+Ά
j,700- ::ZmzVV6mڄ ;lٲ055L&^7 Kvj:M&--
ׯ_ lRk=hbYYY|zǏG\\zQc"""⋹ 1 "ȎӧOv( }Fv\U^o|@D AHT^x R+m J*uވ s> IHzzJyrr2>sKeFFF!N[HJJwl]_\.}f `ll ccVFP(a>@ *#c
@DG" kG3g"88Xh sݻWccƌCBB|||`jj
+++TP7o̙3i2vXTV
2d`mmebݺuXf
lmm5k``m۶ ڵ+
PZ5;wkzc(S ?~vAPnnn8{,akyd[ !!!U{Y }<,%h0,XDDDD,%о ,5uTD y+v"z\\\ ;QQQx >>8w(@m*TWJ !`bb]ٳرcFy&NZj X"
7oknBvDD텩"OY EEDDT(qqq]{\]w*Nk. 0 "˻1MnID8Q>Q>Q>Q>Q>Q>v DD (00 """DDD|xXDDDDDDDDDDDDDDDDDDXDDDDDDDDDDDDDDDDDDXDDDDDDDDDDDDDDDDDDXDDDDDDDDDDDDDDDDDDXDDDDDDDDDDDDDDDDDDXDDDDDDDDDDDDDDDDDDXDDDDDDDDDDDDDDDDDDdP[uL_ ""z窆vEs"0 "*.xݦw } >|#`;`;`;`;`;`;`pd2d2v8ך5kxZjvXy LSv(T /^OzzzvXDDDDDDDDDDDDDDDv Ydʕ lmm %K,hcP(`gg HLLDRRR!GDDDDDDDDDDDDDDDDC k͚56mBBB
;*✜ ;vá}q̘1 Q<==q/#6ٳg:u*={VMO|(9UEDDDDD ;8qnYء>=%*JkiӦ?'M> HuZf
d2N8 6md2+<<\?͛7Gɒ%all2eʠ[n>2 SNEzz:,Xz
2LmP 2ժU9 =z`ǎ Һϟc(_4/h֬J, ###ۣSN8t?Uн{wԬY6660113 _Z:u*d2<== [nlllP(Pn]8m&+РAXXXM6ŦM.?Ν;FFF]v} XXX@&a d2xyyA5V"""""""""""""""8v 9HOO:䤯 CΝ077dzgϰ}vl߾ cƌy.իWٳga`` sssd2:sń NV&&&sn߾@J7o" /^\. x8tN< www^| """""""""""""""x|}}ƍ ƌ ,Yxӧ ̟? +Vкe˖Xz5 H /_q!++ ;vĕ+W($$$ ((Ӽ\pIIIHLLѣGx :Tm t'O1Ν/^ && x V\jժwehҤ _x$''cpppGСC$&&̛ oooĠlٲ
BRRpY899aZcRRڴi7o!!!HIIAll,bccpBaŊ駟0a* nBbb"ݻwM6ZG =lقDEE!%%7oԩSakk?~< 9L3f &&1111w\\zUj²epm˗HMMѦM)))ytҘ2e
.^dDGG#%%ݓF~5j\W 7|ϟK!p ?^w֭[1}t ::ϟ?7| Xz5.]u7 &L@`` d2&M(DGG˗0a `͘4iJ}A>} d֭֕Æ
ɰvZ嶙ػ*u㿕{HTH3[QAAR*RD [JPfA#(lQQl)PTid?x3EY)+u]<3Y
ʺe OW233kLNUF^֬Y#IzwSOõxbj͚5z5h yxxOFF֭[ =z˂%]pO?K;wVΝm-[SN:i
ٳvڥ?CfKobڵkս{w}֭[4iԨQ#m qI=zPdd믿j7nv:uܴqFNmݺHו̙cǎCڼy9$STT z!M:UF?YfIy͘1jիze%8q o,.մiSM<ٮ}rYx襗^ҤIuu-[>z"\\\ J[Çf=cϏbQÆ
[o)??_{=-Z} *K/ M:Uƍu_K/d.
ջﾫ}'2ezRgΜ1 ^x6zו9shΜ9zganӉ'k4j(Iҳ>kwArrr$]sP P > ;}B;f F> ͟vլxYzN+'|m^{5I۲eK4k̪JWVzz\]]5g ;vY|unݺM?Z%Iݻw/R|eKJJ9+jJ=$)Y|$o߾NO #F<ŋ%] 5+[~~~JHH?`._t
lNW^233 /`U|UV^|yzzZ])66;;;@w{?Nw}+. ?~<==oJ^sSߤ$5k(??_ z楗^zju>>>Z|ܴm6k0`222ԪU+ ӧO( fX$uM4Qz]we{$ijt_Zrsʻ$sy~~+I6K~!Iԩ
G:|$If=cugΜQ\\B0>,^ܹsbG:*}4<+Im۶_tMӖ]viРA[#b|MI aWƍ6m( @ JwըQbM6l{%ƍsm #00Pm۶y6mhڴi.۷OfaVyyjN>]n U
MV#i> WFFFɩ.\kDFF̙3WSͶn(%___
+,D W9),jWRRn{gϚ?'$$\WYYYqm)wEߖ~>3aJL O X,Urrr{+\o]niI13F˗/7 f͚n}u P}x89A,̂ *.w3UGrls=S#YV.]2>z økРAf||JҖ-[֏? %%%)>>^z$ z%&&jРA=ܣ ;v(++K:jժrO孤{ݻW 2߹sg T?ղ pkMVrl d=]^E
k{+sqqQPPyb[1<6Wy.Kl#egg&55ףY˗Kt{O͛7/Rpt^~z)00P_~:tPz}ֽW%)!--M P^^nVY,-[LK,)u< fXN2[mV}vͱc"o1kNo>;whIҗ_~yZnmm۶n֭n,rssS-$]>_*vyTT9[ib.TxlRb=ڶmkNW%<t W(v_~fP1[jUz0l^k)7,//b)\k?k?Hsq͉'sc}I+/]fU>}쬋/j֬Ynꫯ<$YMw ~+1k.Ok_ZgϞ/E})vԩS%] s˕#eggkƌn3m4յHK.է~*ŢKN:1cZj
0@ LU y.!7UYttY41zh͛7,:װaôj*ID<<ܪȑ# uU۶m3{uHc4vXzVǎ 5o<%&&SRRa
8P۷jۨQ# ?^o'a(--1ֵkWI҆
k)33S8m4 =Z6ɓ'eXdXk裏WnnV\O6]_K.35m4lԨQ[W=3Zt$_ޭW9boɓ'Ŝ)))z5sLIҘ1ca=q℞~iIs=g6iٲe?':. `:tf͚9: T=..`Yɓ'[?3 @mWRTSU'xB :q4hpEEE)** uhСrss5zh+((Hu5G7nFaw Ç״i/BZ x@֭+8\\\g}?:u(00PWƌSd֭[?vޭ-Z_uYկ___~|||>S@@N>>ڸqj
U``ݻ?.v䠩SK/M7$___) @/qzĒDFFjӦMWN>~W!!!ҭުYfYIҽޫ \h"(00P3f =z<3µ'*00PAAA
{ｧ &e˖ŶPW^1Epp&M]jȑvƍokUdd~G5ӲeK=zwTN )00P&Maԩx
{}Z:u 8p #\=#2 CsX8րFӀVZiV̙͛7 nЎ ;lD F< zB]\tHje! @y*)*NMɩdVƍ}vSJLLT\\┟/:۶mŋ#___edd(<<\}Qn_ԡC4l05jHт7nhڵ攉%$$D;v'|nݺ)44TR6m /hڴiEOG u7*''G...jٲL~IM4)m۶ÇO^zϗx ߿_%\n][˖-SϞ=,***J=z[o;wiyizGՠA0 5mTC5OG֭uQ;t*&&Fs",Y~[-[
QU
6dڷo+WX͛7kɺ馛*0 kݺurvvQ"x
Ptt<==Fiĉ:pQ6c -_ M-[o7^se˖WV/oooy?KM+Vիխ[7+==]֭֮]O?Tf'j߾}ֲeVdC 1 . # ,F\s@5d
9: | P0 ؉, X `'
N` ( ;Q vrqt kw|z= '~tt ,<a p,ߨ
gPu1 ؉, X `'
N` ( ;Q v 5ɓ'eXdXtIGSnZdOe˖TbbbdX[}gA1 @CLK,QllvPP999),,Laaa
qt8UyN<<< UŸ
*/%Ocǎرcu9G lK+/%Ot9 TI
`ZdLB~W||npW_}<'st8 T)9gr7ȑ# J9shΜ98 UX~JOK)Sx~ *$[xVUX `'
N`AK,b7|#I2e,ɓ.]҇ ~{G!!!rwwWzԷo/bQll4{lmVX,E =zTFRӦM+ |߿֬Y:y5lP
SuرEAAV\{^zrwwWhhڴi &觟~*ݯ#GqZnW_}Uiii% ̙3>|ׯ/wwwEFFj:qD
Wǎ "777W^ڰa5 \ SaaaJJJR^^Ytr%gggIRjjzmK9;;WΝիzj7N3gδylhϞ=rqq,63fĉ"+yyyĉ矵b
%''+ 9!C
$]pA+VІ
sNvmE%$$O>ڹs, @ڿ߯Ǐ?jrJ
8P999$___:p-ZM6I&E~uIɒ._T-YDk |`>U~}IСCcwQZZuY
2D4k,-X{= >|X }Ҕ?S-Z$͟?_/
ԳgO 8p@/^TbbӵyfONN| q5nX{Uff222eEDD(--MG.&??_{Ν;3f
JNNVzzΜ9 iӦV߯ {L999뮻ta)++K֭SDDN> =z(##êmzz |A%''Aڼy233={~>|]ȑ#ю ;tE(%%Es̑,X~H'WC Ǖ$%''?W]m $OV.]tiկ__˗/WzzuE 9rD
kڴi,Nd%''+>>^#GԌ3tA[l{O?.^W]~[B4 = T-3?.* >~LJgP15kH}]=Sp-^XZf^~e
4H Eu֩G `IRrr9JQ駟ZΝ;s6 Ӗ-[)IrqqQNpBSv HҥKbhڵ>֭k~ M4IyyyjԨ6ol =z(22R_`7l;|:uJnnnڸqYwynZJsѱcԡCmY:=sC=SjԨQrq|Ϛ5k$3^zW^6_'*!!Ao>(]M6ɓڷ-gϞջ+Iz饗4i$s]:u+%%E˖-z2EGG뫯R֭uaYF=Xc>}LR \ jFu[b$)22RO>dۼk.Oe˖bi֬UՕV^tjΜ9E& cǎ5ԭ[7I~Gu ~${ElIII1?~Y|uVZ顇 "?˗/$۷ 5b_x$i̘1VWWݻ ~\tR(88ܓL /X_UիW+??_VnW{`w{?Wzj
k ᤆji 0p= t+##C<*p'Iر4iz̙3ڷo_VwucٳGԦMEDD_b(44TgΜQRR>^t
7ks]|Bh^^J:'S +))IaaaJJJR~~uٳg͟+ެ,8϶oKi?ř0a|Mwggg#edd(33آ, P;1!*]I53`Y9eu%G0k
d\ ϧ$mٲ,oQRR {N̑ (u+ Zӯ/i+[%ɜ^2յǽtN
(((m۶$m߾]ns1o/u ڵ$۷OΝ+u{{(::Z_^w֭[l۶m6ۺu$n3 ԢE Iϧ-] eNW -[XdmۚS6@Iӧ]_PP}h 9;;+11Q'O.miyyy7xf1Uqn6Ig}V_t>}Y/^ԬYW_ylIYq,X~ @빫Cj֬C Rh̘13f""" %EBpyM<7 ]LgPPS%].B*nhG4zh͛7,:װaôj*Ik&R7|Sb
=:x>++K_zꥴRߖ \w} P>}4sL%$$Ϟ=sj VN*WWW8qB]tя?(HMׯWݕoQÇj;rHEFF*''G]vնm̢^:u9Ҙ$ ;VzձcG͛7OmذAT6jHǏ$z'/o_]!!!JLL]wݥ+Wŋ.$O?i㏭ =zTO=~4w\1jӨQ$]M>]钤?SO?>j]v$mذA233%]>ӦMѣ l S;vP˖-
5UgٳOq 0 W=&׀˳XDDD(667 ]LgPPO>>Vmg) @'OTN#___q :y.\h3f mܸQwqRSS5zh*00P
T +77HSK/M7$___) @/qzĒDFFjӦMWN>~W!!!ҭުYfYIҽޫ \h"(00P3f =z<3µ'*00PAAA
{ｧ &BW^y<4iv#Gu> @EL7ճgO*11QqqqS~~#Tm۶M/VLL|}}pG۷o̙3 ˋ/CiذajԨ$)77W7րvZs ;vO>QnLyyyM6z4mڴ"#Gh#lRSLO?&M{̶mz'U^=_O<߯c[vޭe˖gϞPVVrss =z護Ν;uvvּy{n=jРd6mC'ڣu:zx
q UzzBCC9so[vK,o-[S(ZNІ
nnn2 C۷ʕ+ol͛7kɺ馛*0 kݺurvv| by װd
qYu$ *x~W j*Q| N`(8Y,Y,lT >yN.]p @%qqt OOOY- qP4U_sh @e uׯ0W_}U ) N U̿o9: | P0 ؉, X `'
N` ( ;Q vrqt L-UUl# T4r @mmJF> x/bW v D ؉, X `'
N`B>>V8;;KRSSջwoXY:wV^իWkܸq9segg+&&F{쑋|}}bB3fĉ"+yyyĉ矵b
%''+ 9!C
$]pA+VІ
sNvmE%$$O>ڹs, @ڿ߯Ǐ?jrJ
8,Unn 8hѢEڴi4iRWN,HMMՒ%KvZ}6ϣ$7 X,n:[N#Ftƌwy\- ;vLG ڵkm:oU^?S7o͛girr]_G Qnnd~W͚5KVҎ ;Bze?z嗵vZm۶M6ۏ ;Vs̑bQ@@@XΝcǚdw}N }6nܨI-[6mh߾}8qb{eѢEZr +$$f| `*_~Wv$IƍS||ի~Cjǎ rss;Ｃ4%''ٳ2d$i֬YZ`{:|>#)))IZh!I?^xgϞ:p.^Dkׯ \7{ mٲEJKKѣWsNkƌp႒3gh
jڴU SNN. >|XiiiҺuӧOG Ȱj |PjР6oެLkϞ=_ n ~a:uJJKKhС6?hΜ90aΟ?$effjJcƌaի~7(##C?Ç [nѻ+I={6ol;v옞yI?{N8 (bL]xѡ 2g(3xB\}X55kH}]=Sp-^XZf^~e
4H Eu֩G `IRrr&L I߿>SѱԹsgufaaaڲe<==%I...ԩ.\={j]?iYt[Y,]VݻwgݺuoI&)//O5͛ԣGEFF*::Z,XqƙmϟSNM7n!;֭[|]iΜ9:v:t͛7\{NQQQz衇4uT5J..oşYf͒tgƌVW^F*hEGGY [wyGm',,Lׯ<ԹsgݻWkU߾}R FԈ#'|R}ݫoH 3Fo]7pMڷo5k֘Ĺ W@@z={>?lgȐ!ڲe/_Wh<+''j:kM jr @muGYG>ް TwwLnbXOWPAXb$)22RO>dۼk.Oe˖bi֬UՕV^tSرc͂+uMnnn j݇ ~(I{+[RRRi&I+jJ=$iٲeV/_.I۷okĈ6xbIҘ1cԻwo)!!A?|ҥ*((PppLR۬$_U||Ə _ԩ9b[y,K/THi6 lwrr2va =z}Wv v | `߾} ;ڜI&WW뮻l cϞ=6m(""®8E$sy~~+I6 Ê~!rA-u >|Xyyy\{ֺ3g(..Ni!Ë}EDDS n/yܹst͜9S :tP:u&"bU [3o߾uNNN)}FTNbqqqQ lXx;-[&mܸќq_bsxͩ SSSur? ^ݻ Ñ ݃w êQE<0a
pI2 lԙ3gfE9R4op
B(IJLL4/q|%0%%%)??^ٳ oV
Y $?^*///{ϟ$effOIpYHHn_細͵Qxl y͞=[ԯ_?=6g/ww 5 tۻѵk @uj5]a?]ۻ] X07][!{ ,+G .]d|Qr-j_y?3g{QPPҥKf1\(bUMIUV߿_aNh `)kJ2% dKxx$*ZPP\]]K}+_IpYd9sf[[
Sic}EӧOS .[L?U#p][ݾ$mcsk Æ
өST^=_~O?]x @GV5P85ڶm+Iھ}
رcfa^ڵk'IڷoΝ;WpqqQtt$/v[6ٶmlnuVImfzE.O[<**ʜB41*<[lQvvvۗ9[jU6ZJ:7
?W]]gۗM~~>@W>c-^XtR-[1 jOiъӿIGZhQۼ+uԩ1W~~~s=Wi
:T~z_K.3g*++6 Қ5k$I Zׯ_?IҪUt"m/\ >^Æ
ӪU$I<<cijѢ_gϞU_Ǫ>3ɓԩ|||;C'O
m㣍7;PjjF P_>ci?uT5JzM7WAAA
PK'''{rqqё#GԶm[y{{[ڵǵbŊҥKk.r-
$5h@WKmCCC5w\\RWPP4|peggn3<9sb?_G=t뭷jժUrvv6ۍ ?^ >S=<<|ryzzwN\.ڱcZlP "|%X n.lO%P(7n۷gϞ
Ubbgm۶iŊ222 >}h9sfcyu!
6L5$q0`֮]kNX^BBBc}'֭BCC)///iF/MV]~t
>\7xrrr⢖-[jʔ)駟ԤIbٶm[ >|XO>ի|'9:Sݺu{n-[L={TDD(CovY͛ݻwG U
'0ԴiS
:Ԝ> xܹS
=c_;Cs{Un]sJ$_"l֬fϞ-Iz7* 5([
@d
@MD ؉,,Y"ŢoF4eY,ɓ'/]?PsBBBzo߾%N_#ŢXij۶dX=zFMW>>>տYF6uy=3jذ<<<رc%\R{VzPiF&LO?Tl_U#GTƍ)???nZJ<3g4|pկ__uĉ ;*$$Dnnn
W^aÆk5x`5jH^^^SӦM5dmڴb ;OOO)))Iyyy6ΒT,rvvΝ;իWk7nfΜixيў={"___Y,mf̘'EV ҉'?kŊJNNV@@@ 9rDC х %IpVX
6hΝۊKHHP>}sNsY@@~߿_Ǐ nnʕ8prrr$IՁt-ZH6mR&M sԩ%]Zd֮]>y8nE~~~:֭[uiĈ?~.]Ҙ1c;˼cǎѣZvRRRJ ToQ~ vIƍxW%ICՎ ;wyGiiiJNNٳg5dIҬY`{ta}GJKKSRROhB4| *((PϞ=u]xQJOO͛կ_?99?jܸݫLeddh˖-PZZF ]M~~z;w]3fЅ t9sF .TӦM߿_=rrrt]wJKKSVV֭[>}Z=zPFFUt=JNNV
yfeff*==]{Q5|p133S]vՑ#G ;vŋJIIQJJ̙# -X@ov'N4 Ǐ+##CIIIJNN]<> 5ŋuEG UAn̴zj+{^! j"F*5kH}]=Sp-^XZf^~e
4H Eu֩G `IRrr&L I߿>SѱԹsgufaaaڲe<==%I...ԩ.\={j]?iYt[Y,]VݻwgݺuoI&)//O5͛ԣGEFF*::Z,XqƙmϟSNM7n!;֭[|]iΜ9:v:t͛7\{NQQQz衇4uT5J..̟Yf͒$=1cվիW/%)'' K5[ 5 BA> >s H-@>> X@%Yb$)22RO>dۼk.Oe˖bi֬UՕV^tjΜ9E& cǎ5ԭ[7I~Gu ~${ElIIIѦM$IǏ7ԪU+=Ce˖Y[|$o߾NO #F<ŋ%Icƌ*R秄K@2eJIoDӧO* )
| P3PT}I:vhs&M^zV_y={Hڴi_ZrJڻw$, +e$SN6+ ˓$E`sͶ֝9sFqqq.O ^+""p{sΝ zJMM5_O{_ ! Twݻ F@> ?|eddXx֗ TW WC DLAT .HY`eKddΜ9cn:ul /IRp
B(IJLL4/q|%©0%%%)??^ٳ oVVsygIrww{ /r @MU( j=C>> ! j"
j:{ ,+G .]d|Qr-j_ 3 LAT+\_HWK.յǽtN
(((}h̙JHH0ן={VsՄ M:U:q℺t QTPP{7ިÇ[ 9r"##]j۶mfN: iLƎ [oUر͛Ds}JJ6lؠ}Vm5jK|M=_iiiʋk
С5kP X>Ak"""ɓ'+""a R(|; zv>C Dj_~E-ZPvvjQ[JMMU^7H<}RRRqi̙E oF'OVlllL>]/Y)OOO˒ I:y6l(IU~>H
Z zHv$Y,+??_^z?jb
8P.$kNW~}mڴIM4)Ͼ}Թsgs1///999)##CZh9RVqٳӧ[\PP`U֨Q#+It钞y{2 hIKK?RN^7wt P->SSSSQJr @MT\@" T]~ Pq[ |rҸqcm߾]={Thh8K}}bJ_ԡC4l05jHƍkZvmw0!!!ڱc>uM̔ڴi^xAӦM+Ү_~:r䈆 oQ999rqqQ˖-5eO_IR۶mua=䓪W'xBWttt1[WwֲeԳgOEDD(++KR =[oiΝE:;;kyڽ{ }Q5h@yyy2 CM6СC @X PEW- Ou+Wr @MT\@" T]~ Pq[ b, &( ;Q v D ؉, XiJ9: | P0 ؉, X `'
N` ( ;Q v D ؉, \f$)-- P>K 壺 | Tv>@ T;8 jt;:F. @nD> @y# P UDPP$ԩS ---MӧpPN50 C[C)r~fl\ꥺ G?{U j||Zjg
prr$^mu5fAI.P{l\ߚ[}T\@" @O,^UC>w|Zjg2J; 0` (*]'OCA9L\ך
G3Xqk6oEe{AYT +J%k3R,a #F ;Q v D 8Pzzbccu뭷G5{l::#=cjڴH[}CD9z7dX4͘1CڵShhyvQJIIqtf*~y ;V7|<==kѢE2 JR766buĉr~7RVV6lؠSꡇ
7`r9T Fq`/r
<"GZf*}ÇaÆPhht5k֔ݡ{ <㫶kʔ)ٳnU뮻+))}ЇT" 'O4 I$pww7oժ0Q
...dxxxV˺ufdff::Tѱc k_f^G777# 8pa}`s_>>>VϞ.]999p'O6$FXX^o
۷[W&O\s;VE^_a\| "׀c`/r jge탿kprr2 |#EYhȐ!֝wpP~~?^=ݫl%''+33SI&ѡ{jj x%&&[n{L͛7Oڴi}cT9[v oU~o
Խޫkٲe
/rW
u} qC" 5PV19}Ag}V>#<,u]:~RSSW^yEG i̙VQJe^}!::Z3gw}d]xQiiiJOOҥK[VmC ` P-ZȬ ."O?5oݺ J\?|p:uBy{뭷 Iƣ>j% )Ui/1$Gvt8!*K/ OOO~+~ڴi$8~U-|tС=, "/0D> {k,c`/r jge {1C n$''YSO#0c{ ZG R `ҥ ;;NհaCI?J
رc ZAW4i5w\G2o)<<\oA
QǕF - ]tI?K:[\a v`/r
y ET5 rԚ5k$I#GT@@@/$)--Mq~e^};0?чT>
eeeo$u֭m,v*Iڼysņa|%F{
6L3gBCC ʠ }7{U:uTQKo\ 70P{^{3 {.^Xb(5iҤl}T־g|ͱk.o>1(JvQH7onsuJJJPva|뭷:.>жmԩS'
8 rrr̿jӦN:zJחԣG}IE>"K/_eoG Q% |6l 8`_ka^{P {J}pi 9rĮ8Ql}T־g|'Ojyz%I5R =mC , dgϞ5W\weTO)))>}$} J̙3?~<==pBG2:yrss%I7o>@.\.\J<
&0 1|viiiȸ}TVGS999矵h"iF/R{ E偾d|_ [ 5zJ^}CE jذF du]ڶmmC , d^^^6rݕmP sy9:$ÕX/qt8(dSUVRFF}J-Zs:*TT#|'wp7֛oǏ+;;[ԦMԦM_]g϶
k>a" 5PU,{l)g^Urao{8Fxxm.رz-5hj[Ǡ
3諯${jѢ#Bi|'ղeK3 \ŋUԠA-_\v$iڴiwH jG }TǏM7dEnnn{n~풤X:2T POp +8'OT||222ty͚5K Ttt^yGQ 9++vW
qƙ0w\
24Ο?g}VPS7n{Iƍ$%&& ~PMU y
<<<4m4IRFFmV.E ݸk؋\U}*
}>-el\R+:{| WN ;V7nbkfI!B Tu?9svW
B{֬Yzg J^Pbbz)r-Ȱz Uӕr-6kڴs\\\Ƅ"ݷ|||k߸>N~= ڃ| "@UE߃DsONNŋٞUp+_sDGG$>1(J֤I99]~' WPPPĆ3~x͜9Sojر%Iϗom =
)((Ȫ0̟-KE"͛7/Ҿ}_Y1( x'$I۷o_dU̇+56q㬆z۱c ð+k oކ$}?ladddSL1, { <㫾ݸی QPP`;u1}t
2Ν;g՞>m! 8nDEEaުU+#))a:ř +5sLGr@Vay
ݍ@wIɓ &{0 c߾}Fppj~}k۷ow<==kdL8ev
7X]['xª pPח{! 5P19R5a
///s{}VE |X7o."WWWiFfҿo::L\ϟ?_+##Á.}v}/y{{+##CS*66a{6mt=sjܸ[ |6l wwr~GRE\[oUfR>}tM7S)))mݦ~Z 믿^A
{f Fq`/r
TU=(-R>{:|
(egg+00P;wի bT;B^}nݺZjFm*$$Diii*((P
ԣG-ZHG QV}H5& N` ( ;Q v D ؉, X `'
N` ( ;Q jXY,z]NNNW֭ԩS6_w߭*$$D͚5S =4c ٳ> P,a jʔ)7.G|||$I.]RBBVXݻ[g۶mzGd.Ҭ% P @ _8ƍ3JKKӂ 뫌 Oϟ7?uz$EEE?Trr222mڴIFR```eu P , R5w\IRFF,Yb_p222oFV@@_w͛3gTr < P W >.R{5 j(խ[W4g [nzW_
@ P T\w.]>}Oۄh>|l٢LmܸQ7n4iӦ~i
8М T j7r @> "X & 5XllL"Igb\sP]Vw}w =uӁt)w yxx\w\ X %$I>>>_{WC QppͶM4Q&M_h銋͛K/i֬Y> ka, PZ&OrjѢ.\@%$$00 | ڍ\ hܽ ,,,Lz$%''?tpD P f
*$wwwF | ڍ\ 5X h]*q ]VaCTBd P jРF [*--\+W]v$ ;Q
B> @F. $JLLԼy4oc=SڲeWǏ ןK.)00P5w߭AE T j7r @> >0 G Ց , X `'
N` ( ;Q v D ؉, X `'
N` ( ;Q v D ؉, X `'
N` ( ;Q v D ؉, X `'
N` ( ;Q v jꫯR֭kɺtR+aÆrwwW
4~xm o
Swy: ;vbyf ~A).77WǏWxxչsg ?~IKKSll0C?cmSSS5tPW?Ν;Wd͝;Wr G }c~M<"##uo߾矋l{1uY
?\m*00PEyZUedd(22RE3W \ tF>@>@> # 6T
T;}d<ƍٳgرctq}52>#cǎ ҥK'Zm7}t͘3guV 4|}}_Tqm߾ݐd|Gw}g~b/((0#,,.~ᆿxbcƍFz)))Ϗ?h&M26md|F
///?ն]t1"##+V_|Ѽysn3{'0cyƎ ;e˖F3tm 2 j 0on|ƭjԭ[?풒뿌7/62i$/HJJ*X 3${5W \\\6r m j P( ҥѺukef2\]]xs٢E ٳ6uExe999
7` 9TqD$/65jdEQuiXp,116f̘q]0233AAAO?m.۳g!شiرcb1VXa.ۺub >|_^~gC\6m4HLL4-\pvv6Μ9c.ӧѾ}Jב =jx{{ ,(Y, 8 @ &# P|
R͵cjJފ?`n3{l~W:uȑ#奛oYoSϞ=Un]y{{e˖Kt}Y-ҥi&s |}*""٣4=#2777=CZ~$)??_m۶ wa5to!www >|TKRJJ^x͝;WnnnEoY۷,((Hwx{{j5jgϚ6lؠ u\v7e˖V[o~j֬<== [{ ,IVCoذA:uRPPG QAA6o,IX,Zfve,g}Æ
j<<<ԦM"CI[Nm۶Զm:yyy
פIׯ%IK,bQBB6-[ԠAm\\\tym榅 Z- =zFoȱ x@?t7S111@B.@.@.@. D> v J, ?~V\l=˓$ z_hѢE*((Pv픔dG JNNŋ_kܸq4鮻ҢE_O>:t.]zqfggjYG t9ݿn
8P#{ڶ[!1f$%%i ɓQV-mV1hѢшK>q TXQm_nnn҇]QB > 033C˖->X={C֭qlڴ ͚5CtttDZZBCC?:uRNNNR̗.]_~5jҥKt6,, trx{{˗ K.\2 ;v૯BTT[c766 6mڄLiydd$ߏ~?0 `>@b>?0 `>@TCpB&;wHT;wN}zzHLLз^ ārL& 4hXkժ%ZjlӦM0`B0@֭[cǎ-[bŊkB.k_ PvҥX\pATZU4nXdddH_.Ǝ + :$N:%&M$rhӦZT !6mư~pssӈkѢE8JCRO[[[Sٳ eˊw節722خrʢ :t jm|}} q)iYttǏB+Od2qiي+B~z@mz!V~ԩhذZ&M]W E[N!z{jm`` V^~pb. s } c>|:D0 `> )V*W,^R% Yfмys
ҷPdIL07n.**
CEɒ%all ccc9qQ,_ 8<&MCCC_m ٳ-Z@Ϟ=qF\t gѨQ#4m >ƍa`tFX`ڴiMbXd > k|+}[~=~WZ
y>338p ڴiN:ax1n
Y&"##'N 11QXg͚v킓c
M6U /_ _5 <<_\9xzz~_ ]teѣ.]$ XX/Cj3{l888O>] R
\]]˕+jժI瑈>>r\@n1`.@1c>3r 1 `>[,! IDAT|NXEDDƓchXf
.\+WhѢHNN5QbE <...]6Ξ=+m۶a8~8\}J}䆟 ѣG͚5àA`kk+nmm
L
A CCC lll(d2a}e2wj
eʔyk] رc0tPDGG#::ZwњGEEe[G ŀ0elllPre888H,ϦMb 7ѻwostҨS|||p1ddd`
yY"Eh,spp@XX |8tbbbЩS')R۷x5n 7۫mJ۷o ܹر#,Y3f &&ш K]~\ PhQv#}| s1 |0 c>@`>?0 P|taQ. Ǟ={ХK4lիWH˗/DEE̙3h 㑜Ca2d6mڵk} dŋ߿?^z
J*Շ7K>88%J8q"jԨ]v͘`DEETR
,X`R\/^ИG\|oK.̙35WPwޅ"}eӛ(op|+Wľ}0f̘<ŜB@Ŋs ,,,WիW^x!}(-[ٳĖ-[pڵ<}ksΰ-[n:-[ zSSS @jjvofffرjh J% ÇHMME6mk]v &MK ˗ۿy \ s 0 mXEDDKIIId066ܹZǏGll,?dffJߖ8 8p@kkk\KVK ۶m .%'222PV- @Æ
aii MZZك֭[_~ŹsСC9ƨJԩ 9OV)N> Zh[+**
Ǐ +'4իiժp)iٽ{m"00Ҳ`<}T:ۣ_~h9r bccq-oj
'ODtt hѢ[QP 'O^zm---W_{y:\^z_֭[ѧO ik P
'O4ѣp1jC
W^] W?, ܹa7$$7oz \ s 0 !+WV[% [n ѵkWqI|rQD amm- ,͛K?{uJB!ԩ#J(!{EzDҥRu/_ .Ǐ &&&ԩSj ?~,E-Çņ
hܸ̔͛7Orl2q)#,,,6044.\B QH+ٳ6mؿ8v7n011 ;vxM (o& ;&<<|gΜ6l5kݻRH$<<<ıcov-tAxxx-TRbƍbA\ !Xz۶mgΜׯE|MC%00P ӧjRSSZ8tغupwwjו(V)))9r労,$*T v)]..."999O版
0Pa.@b>| D/0 Pa>@y,""m bӦML2Tԯ__?dɒR" 7߈2e\.-*|||Ľ{>?ѴiSP(Xhz sssann.5k&.^055o߾"**JMff;wpvvr\ԫWOT*նۻw +bܹr\ˋӧ5ubԨQhѢLxyy\,Zϛ 4E߾}077;wϞ=]vBT
ѥKiEfD"E\.e˖ӦMSKs~!\.eʔ~~~j pU_ѬY3aff&-*F qu}\8t萨X0115jԐ>0 !ŋE6m011%JÆ
:ʗ//ZjuիWE:uZ8yVƇ*!>d{>}CiFoTRB./Ro?)DT(1`.\
s0 |1 `>|
oLXIDDDD4???\zwy>\]] 8tPABDD's""1 c| 0 >=x" %KC }@ 1 ωQA@DDD222 FF}\'`hhyLdff\ohhL#ҏ:۷/\]]e^DDy\ DD c>P0r@`>@#`2͚5ΟЂP
<5k֬C0s cqcÆ
4ƍs<3g{.5j gΜD^0 c.? | w@`>@Da>? DNeDDDݽ{qqq:FT8֭[:[XXFvϟ?u/]4>`DÇu/V+#""0 | w DD a>Ppc8!XEDDIMMرc_BTB&Z^x
4J, dصkW6}t\xQڧ'd2OppZ@.]oܸL3g
L
6s_^°aP^= _Zj055E~w{.~TT
Kw}~ذakd}FGGcС(VLMMQlY,Y䓈
{1}w'?sf߾}d
-8"##ѩS'@&a߾} ) L&ŋՖ ;NNN000?h/_p%h@zp i}dd$ {{{aÆ8{Z h۶-v777iӦZL<eʔ\.34NTq2 Ǹq舢E ~
QdI!22@ִ`055E1i$dddH}
@VT*M6iÇQFP('GGGL4Icnݺ쫎m = ;DZֽ{wDGGҥKjmЯ_?XYY#GDzz>,, }E2e`ffWWWL8)))13FJMZdffbҥX"r9 ѵkWĨqm4n
UT`6 =
kkk4o\Zիȑ#jmgϞ
GGGsxFB,^˗\.G2e?j
BΝakk Bjժa۶md 9RVzػwF?+J*Bf͚!$$Dq&7SRRpitUmyK3f@ɒ%1yd6vvv5j N:֮]}JEN > 666x
044Ě5k
,."8gϞwppp*T˗K3331{l*U
r *TNe~TP[.\5Ç^z033C"Ew!!!ǜ9Kv) >.]RWmڴ 7?j۫@~[ÇVVVׯ 5K.m6BP]v£G
sssT\ d2Νr/ C*UpTV
r DBB~APTR4hF>ʓ-[XXX)))q5jRuj&Gz)+888K }ho>zF洆իWW{?9s5jԐe]ӈU5¶' 044T{qmx{{CT
]tǏ5陔>BCC!p5l2L
\slܸnnnyfb_`&MEcǎNBannf͚ɓ'jۦ`ĉ(Y$r9*V[ D֭aggB777,\Pчv ʖ-+PkP}.~~~ظq#ϱoͳ/"66VK&&&ܹڳG]m͛VgoG""
hڴ)ׯkW^ .eddUVx,X X͛7ŋQV-7n`ѢE?>2220rH|7j|||bĉ_>^z={Q~}[NhpuuahhRKbر1b,Y !ٳ'#G_ ԩ#
ûc :t֭[abb .ٳgQi&̚5K =0GBB}pq$&&8p`ED_~=@
sA2e3˗/ɓѰaC :t
BZZ~ Y9ZjKÇꫯ4n]֭3f ,, ǏGTToN%96 7|{ݻ7ʖ-Tl۶
_~%nݺ6-On &ƌ3PfMl۶
ǏW!/p@@ ^~ŋ#&&CE#F¼yйsg<~XK.aܸqѫs :'OF%PD $&&"##sA"E̙3 ;vӧն߿?T5k9r$LLLp% 97n vWes#ϟ?NJ+Ǐ ='"1 .?СC1~xXYYa ԩ߿ud+-[&=TC=kذTHOO~?eDs2T*˗/eׯ ē'OB ?~\ [nhӦڲ @>}ZZ@TTIdff-YҲXann.&L_~fffjZJj /
5kBL,u 'd28r䈴lŊL.;U,^R[S@l۶McR)̙#Ç/L2E(
uiiiFFF"!!AZ@,ZHݻwL&sy ӦM Çe6oެs;!:uXn8{ؾ}]066/^xyyi}- ,\]]BbŊ^zիu"d2XfZq GGG!믿EZyS WV[ޠAQfMzi}} !rʹiӄRX~y@\tImW m222ĉ'DϞ=Bo߾y+))I5j1c cǎ 1x`ͥK ann.z-N<)~aee%w﮶G˗[[[
{>qD!ÇիWX ?^my =D"E|[ntjݺu 1m4!DVNSdIѣG =*d2sN]smN>- A &mnaff&L_~) HA^saԨQ_eo ľ}r< :es6--M߽{WZ^dI,RRRe>>>8zq"9R+V16"OtYZ5+*8w?x`F!J Lv陔>Z]E߾}uU\9xzz~_^BCԩ]tAbb4˖-CJ`ffcccxA}Bh I*;xyyI*U
fffo3зo_|֭Μ9bŊa֬Y9nOZEv'O5JۛIxx4Sr|-ϝ; v֭ \v1Ǔ۾*}˖-Dzep=4nX/
bӦMjyoooL:hѢ~' 9WFXX9 |ظq#5kAaؾ}u^N L8cƌ_~SEDW~=?u6ms˗/#--M=իWwԮ]vjݛ1ݻw =W_}
ŷ'۴i,((:u
allwJǮHJJx|||~k.loo/+{ eoN[vvvW͛7F07717n Cm
M6cns5kb_c|+V +W~yTn;wM6B
V}_3)}yeyyӼys˗/bŊ쬊Qu ?[[[4mTZi9;;;,Y&Lƍ9G7>M,""***
(VXmTs|gHejd Y;99i}@\-5XquuKqd? 93 ߿? 8ׯ_͛ȱh+T9xjj* xTFUl25
:t?`ժU ###Gvfffj 6 s[Jmڴkפe666ȺnUBu΅ׯ!uz0N"""166x- @tt4^z\+ STT h Cv6CvSo&tˏ| /Q~^뙔>ڼҶ<ׯ,oEzz: pqTX jguDDTpxhǢ ɤ竹7goOQA@DD6kkk:˗_xaaaB,255Ejj278٣
A (][cW tC -[-[ ޏ.J...s߽{B|^x Ҩ>h߾=͛'s ˷a/*Tɓ'5`TZUPL&5 d 3iii>'`mm
SSS*& ]mʖ- ccc졺rJ}ظq# ;;;; ;vPVmNV\sbÆ
jEٲe!u833~-*TG a֬Y1cFEDW"Eyn1@=xSwrr҈)66VoO?[o+6,ڼ]t O>šCPZ5iyLL _@۹T\8?hQzuYFZ_>s#999~ڵkq5̞=ݺuݻwQL|a>ۻB>#9UTח + NnڢH"8rB˗?pEL8ڵóg`nno^ƫw=F%`ff&S"ݻw5F}-~&4q,""zJ%4hM6!##Ckƍ#66Ǐ cDܹSggggAJ}AP(x)j*
7778;;c9ի~Wlݺ}ɷsZ&-۱cѰaCo`ڵ
B*NJJJHw}k4m2"$$$СCSUV©SeC@@ Zn
rrrz.S<: u ڵ qqqZT/ՖFP*044D͚5S~^r4iDc")]S^ ЈE䄯صk>}Ej۶m6l͛{oCCCiChB'N
qܹܹs5(^^^?c[.51;wDѢE[. <߽S
h][ΒIII \xyjժ033xvޭ{~Kn9}6GR100@:u0{lsJ"1 ș:((HZΣrq8uZݻwq֭Bs}^^^xLLL^+o ?~'
߿ wBU_2 /F ={C.ҥKSڶm+߿?-[CCCG"!!A*8x ,,,PR%TT 0fѣ{ܾ}-9s4> ܿ }Aqư7#FFfj9/_~ ˗ǖ-[&+V_@T
˗ǠA0ydDFFYfHLLÇ1}toչsh"t Jϱd}jРѷo_,Y4iݥ
1~x 6 h9 ?ӧO|]͛bzkW!^N ,Y 0exzzW_}ӧOc֭{2,Y+W`ĉݻ^_EӦMၿ[Z,=FӦM0l08::JL6
6DϞ=&L@Ϟ=QlY [0m4̜9UVE*U~ׯ_iED_~= 1b6mڄ/SLA2eܻw ,= EǑ#GuV\ ǣN:ر#{ .&MB!,,,Dɓ'_~~~VrѠAq><< =zTT\Y
wwwq<4GJJ2dWlY > -uuAxxx-+Wh.F%)",,,D.]- {j|^۽"7cbbD݅ qN_m?sB.zꉋ/jSѹsgann.E߾} ?{}^dBd>*߇+S}vXEDDݻw1sLܽ{ 80чfPQ @DDD : `Q!JҒ 6L?s"" ""|>T>qMzLDDDDDDDDDDDDDDDDDT HO,"""""""""""""""""" ,""""""""""""""""""= HO,"""""""""""""""""" ,""""""""""""""""""=t DDn_WtDͨ
:"|L > S>Oi9:;DDEDDDDDDDDDDDDDDDDD'`XDDDDDDDDDDDDDDDDDDzbXEDDDDDDDDDDDDDDDDD'`Xz
L&L&Chh
PT|̙3ұ&Sz|݇ d2
d{?TYl5Y чcTP;ްaBCC" E p4D)J888 㑐PчP`#`mذ3f3g
*Pl|OP
nnn_~sqqAxx8q| nnnprr~ԿZ =ztACD$6)CXX0}ttXDDNƍHDDDy%E0 ""O*MJ2ѧS~d:u`:u*_[. /}L͛`̛7C!"B*69'O*1c? LKg.ADDD U"-e"O 6l L 0c d2P}FF~74m(^8vd>}:ҰdԮ]daT,,,`nn777t wFff}xÆ
Cҥajj
t]RÆ
L&C͚5s}̙3y%<< cƌAʕT*T*Qre ;/^кMhh뙗s[ժUL&ȑ#slB& ?A&O綿;5j XYY^z?!r;w0` BP 4i^~w}WWW5k̙3F^+ SZZ 8vprr -
oool۶ٳg8p \\\ >} $$$WǙM4=LLL :ѣ:S]gΜ˗/1rH/_
"
} >HAT\! && ;v
@
aaa0ڵ vѣh"KNN'.^###XXXhM,X 'NLMMP({aǎ F۷/^| B x%vGٳVԾW^Xbʕ+k{HLL9:wI6n܈>}H,r FFFx1 ?~g%ǂ7_ر# J ڵkq4nXg y=GիW/3۶mâEM6(Q"W} !Я_?_ @V5^O\.ϱ b uP(۷oX~= >5jhlsN))YCZXX 55ڵkq1TX1WǓy{L.\@ ---ajjW^8~8۷@.BTT#111ذaك_5c{ڴi@ Y%^x4h~}{xLMMall>G֭ѰaC ѣ ׯ Μ9Xѷo_ ŋzj[jnݺ#66x Ə Lo HJJBDDpqtMkQU+W !!8q!C]6*U j#[IsRSN1d!7 $$ɈA||<^1cƠhѢoKɓ'RUJpy#>> gϞСC<{Lg?y=GճgO"<< 'N&)) w ;}\R* KDFF"22ӧOǎ ;~ۯ[ƍB9s$&&իhڴ)о}{ǫm{u|7HIIAFp-"11 >> )S 99Yk?غu+`ff -0n8 @ݱo>T^]VPyؾ};,--'Nv ###xyya͚5 siLW/ Yiٳg8}4
ݹsqqqP*X~=ʖ-+S*U.\֭[? ;w.accSNQFҺ/'O%"##1o ]tUɘ1cjʕ XYYaڴi7n4*؛0zh ]0qD8::ŭVZ8vjժObڵjO4 iii(W ?U Ю]; >|FFF~Ňy{]nݺXz5K:t(֭[ XbF\ ?~ h42W
p \t)ǏC
#Fٳg#==]k?8y$6m*d/_-g5g)))U!©q9HDO!ׯ_A@DDQwvA@DDT>|SK|3Ph v pvvƷ~ͬY _9RʕѮ];vڅ8cҥ9q2j(X%VZ p={ O>
Sf9;;I&C5=bjj*"""xعs' `РARMv4h `: 喪MUh&Hb ;vE<~8"## SNf055պn݈F5z 8v옴<::Z}̘1Rav5jԐܶm[)7rz1F6m Gxx:յյkWS.:::Jס6⮑#G6cǎׯq5mzgg L ̛7VVVҏj?"""<0 """DDD|Ph ^
hҤ*VŋSQtE @ZWӺE G OCZ*ʍeˢB
HKKCz`ܸqzC)n(S4o jm9ʭN:BmA/^ 6ZrqqAr崶BZp ((:fΜ x葴! ~-r{]I%..--
d2d2B죡JyM6uϞ=bׯθi &HS2ɓ'" ϟ : "|
؝)k QOM`Ngo2g˗ XgϞIThQ۪F)YQ"QN{?ݻ?K(7n;w6ehh۷SNx!Ə CPaÆܹ3|}}Mk}D/_tm=GP(
6`m۶
ptt
r#/מ6ϟ?51KLLwzz:"##MNw9& w 5kV\P(`mm-x u-n kx3vΏr4 "nFr$"zWB.ODDT>|]tDDDs>e&FAT3gЎ_ sϔ
ػw\5U͚5QRSN W([WٙsXf
n߾W^aհœ'O뛫~Ϝ^A dff ~:Kqd/
Fwzލ`kkջ ӓ'O>mۆ.]hĬMRSN^ `DDDDDDDDDDDDDDD+,մcF]6 &88X*SNchذ! ի JUtIK522_/` @@@ """]ҥ"SNlwIi?ڦ d2 +fS ժUSkɓ'6vu5 \v-U͚5Fn{j`ll}])d5jFuªd2t
f#= k֬ #/;U1ۿWJ#a3Xacc 5k6 .DRRu]v50r L5 -Bbb67oݻ|.dee%͛75a:S]{/_b:߿? `ݺu "##s\ODDDDDDDDDDDDDDD+R
ȑ#Z[.||| C O?$U$cjjpB ;ЩS'ܸqCZÇCsoc`` tn:۷ ЫW<}v4jk֬8vƏ hРT\6'N5"##%M# .\akk+_P*T \wɓ'044k$1333L2 qF >\5,66fܹsammu{kkk,[ @ҦM\|Y-33AAAXd *WCm?{l#$$}ݑ#Gкuklٲ8p`O rL+VD% }U9ҥKDTT}wpvvFJJ
ZlSNI`/_Q `ԨQZ*ѤIOjEGGѣݻ7⋼ "YѼ+iӦɩ""*p+WF"""3U.`i\>9\T.-e"OQAK,AHHJ("EHETϟ3֭[ׯ_㯿!C0bXXX ::Z*=z4
w Ddd$&Oc033
Lއ{c
UҾ}<#ŋ")\sssDEEI+V [tvvƾ}СCQFP* )}IӼ{իz*Mb
l˗/ʕ+aeeXddd{ظq}}}aÆѣ8z"--Mj+ԶY&6oތ{pww%RSS pqqannCc200UЩS'vP( J%߯s;KKKݻ͛7Ghh(P(```xXXX`ڵHYo277 2d
+++dffW+W.?N}B,L]t?>}zDDTH>V^ իW/p#% `?g'&{DDOFQHDDDѣGxdPu)[@||< ӧOcѢE ˄ pM_*HMM+z={HS&淪UE]ͫ}شijժ
111@ݺu1k,B
yAAA5j*VL!PbE =AAAfԠ =zX]U i&lڴ ׇQfM^[n}k
ݻw1zhTV
r 077Gڵ1d8qB4ݺuC`` e"%%FFF^:f̘;wbŊz =mٳhӦ
{{{]Cfroڵq-|(^8aee___\~ uqbŊضmڷo'''$&&"55JBvl2={}^dB5Q!a%KDhhhAC'ӧOnj33gzXXYYav0VLHDoԎCooDDH>| >QaǙ+DD>0 Q@EDDDDDDDDDDDDDDDDDcj =L&L&SYltΘ1!""""""""""""""Ĩ j (S*mECDDDDDDDDDDDDDDD
P֭uVa|4:w/i={aÆ)Oѿ>0`}"##ŋ@DDq1( >V,"""""""""""""""""" ,""""""""""""""""""= HO,"""""""""""""""""" dTL ;iA@DDQTr""9|>FE"O&EDDDDDDDDDDDDDDDDD'`XDDDDDDDDDDDDDDDDDDzbXEDDDDDDDDDDDDDDDDD'`XDUhh(d2d2BCC :Bu|gΜQ[9 ;Sa 3E mذ3fa
7ccc
Gɫxzzƍ
QW7"&!CAXXO
0!"OE~;yox
ŋGpp0Qx
Q6!6p`͘1xSѧ"7~XEDDDDDDDDDDDDDDDDD'`6lL&_ 1cd2Ohh>##6m
{{{r/^ ]vqBOOOd2L> iiiXd j
kkkd2m0x`TT 077wݻw#33S^xaÆt055w`7n ,YF`k.44TyzSR ɰa룢0f-[prrB]q5}v1119s&j֬ KKK} Efffppp@dd$ҠT*ann@V!Oǎ b)CCCXXX ,, v®]0zh,ZH鉋/djm,X'JEVP( {cDEEZ@/_B |;vѣGqYTVMc
HP(P(˗/شiN>EQhh(<==# k. 8p z -[ӧ Υ1BBBO ^ =֭ѰaC ѣ ׯ Μ9Xѷo_ ŋzj[jnݺ#66x Ə Lo HJJBDDpqt
$ r
'N 2dڵk ;v ,, Dbb"7774hY t =
v܉ 00Ճ^}š]vx)/Ç#!!qׯ7o##"""""OQjz
RҒ'!!A>
DDDT%OaBDD{'&ET._ݻw V\H n:`ݘ2e
`jjO||< 8vI dM7n8 @ݱuVѱ
7o͛'N ^^^Xf
ڷosӧpvvV}SN[.ʖ-}(Q[WnW^ YfҺJ*?;? T"V ?wwwbҤI8tξRRR" x\T~<0<_[tC ""*0"OKng~8Qڱc ~6f ~'NڦrjWڵ qqq066ҥK5&̍QFIWٵj
&&& ۷oŋZjBŋy)'۷o 4jHJEP`رzzͺtV|baa!}Qky~T灹 1 """DDDETT45iD+VD3\zUkUFtCUTV-899gz.722B"E3DFFj۱}vܸq^BrrFO.ڴiSmrZKjj*nݺ N5Xff&_&Mhm7a 9R=66 >#HeDep/ \7n@ƍ9DDT%O'a|S{gNL?^,"*@/_5TN3-sp @ɒ%2kT']-OLLD۶mqii lmmall DZZMܜ7K̍Hdddo] r %J(S&ըZ NYu<={L9V*v5˾N> ,"z bFx]6 &88X*SNchذ! ի z FZׇ"$$D:~t=DGGk]\ Nɓ ^5ky?DDDDDDDDDDDDDDD`{dii : w k4kjm3uT = C]aiit1Mgee yǏRDٲe wfΜ9:֭ 8s$,ZH9Qfv;w4c
֭[K灈>M,"zT 8r)֭ !CO?!11@|? `֬Y055s VVVRAЎ ;ЩS'ܸqCZÇCsl 0{lٳ 믿Ν;FzS = ~g$%%oŎ ;P(n#<ݻw### VZիWz w}ҥK#--
ZѣGn߾
ooo<|r gk GR@ʕ :"""""BRaVzRaWС ӦMSABDDDSѧ"7~XE !!!(Q QT)*U
O> [ HMMŐ!C`ee[[[+VLk4hq 8s΅ߏ5j@Ph۶- 8sD}̞=`mm2e`۶m3g46ƍCJ 666(Q6mڄ
6H"Z522?\\\.]@T*U¥KqF @SnJVVVpwwŋ!˱eTVM}|
\Ù3gPzвRڡMm_X) G |<9}*{NjXD+N>ۣH"G#iD(+++:u
֭',,, GGGzO݄ pM ʕ =z={)Cɒ%qUŊ m۶8v&L8n)a>@DDYag~>t>,""""""""""""""""""= HO,"""""""""""""""""" ,""""""""""""""""""=t DDn2XZZtDDDT@ """b>@DDqXDDDDDDDDDDDDDDDDDDzbXEDDDDDDDDDDDDDDDDD'`XDDDDDDDDDDDDDDDDDDzbXEDDDDDDDDDDDDDDDDD'ݭU憆 GbpPA@DD>\u""ޯ! (lQAXDDDDDDDDDDDDDDDDDDzbXEDDDDDDDDDDDDDDDDD'`XDDDDDDDDDDDDDDDDDDzbXBCC! Zv\6lQT^OOOd2L>C|l2kӳ"""""""""""""""̨fÆ
' (G(R ϙR >> }r
6`ƌ8sLAB 'Ot8߿t- =,''7n(P NDDx/_`ԨQ +pQP( &$% :"""zG}<^ccT.]$""B ,""""""""""""""""""S*ڰad2/ 3 ~BCCдiSC.xڵkzzzB&aHKKÒ%KPvmX[[C&ilRJ9н{w: 6 K) н{wx.233sNtŋ\.G"EPV-7wѺwfffD͚51sLgϞapqq\.3郐 SIMM?&M &&&pttD pѷneʕB%*U}رcA,] Rl%K JL&Æ
tL???322rJԬYJĮ]Ç`l˗R9*UÇy9|5QQQXn +TZ055Eɒ%_ֹ! عs'<<<`kk RZj᧟~BFF3!V^uhܸ1nݪsy}ٳm۶LLLmb}2 Æ
Ӻ8B&B DZZTt! && ;v
.
aaa0ڵ vѣh"KNN'.^###XXX@&Y`&N(YBP $$Î ;kkkѷo_|
KرG ٳgQZ5^~
={VZfmmd\~ ׯ_ݻwo>v܉{#%% `aaT kűcPbE}^~ ^^^Z`Æ
سg~W =z6mHSd2XZZŋ8p 8A_~6###GĊ+eJFFFFPPك cر#}H&&&@TT
6СChܸqc[|9f̘ }cii x1 ?~۷cٲe:th7 .L&Ӹ> :C.ܾG رc `eeh\p.\ɓ'n:ݻ7v >^~Çѣ6nccc @rj*+V@-ЦM~ AѢEi&Pխ[7aÆ ѣG#<<\ Я_?9s&&&Xbbccϟo߾ ŋc:j*ܺu ׯGll,"##+ ~ ? h߾=֭ ^zW\ABBq 899!66C &== ;vٳg!˱`|QQQógϰfTTImׯoAJJ
5j[n!668p kjšSNB%pq$$$ ../^ <&$$e˖ 'Μ9$DGG#::K.9V^˗kl?qDo߾{.(۷-[Թ@DDi( ^
hҤ*VŋSFtŋ ZjӺH.22RZ+W m_! Kg;h]nBZZ 55U*kڴmu{ =z kZHGGG?NNNԇs͵t.^* g^g~ j
KKKmʗ//];w~~~P!ɤ >}W <ѣQV-X[[P[US...(WuU ｜4*>]6uѹԮ][>jժs*z*JT_&Lɓ'7~ 1 ""4|S| ZgϞIThQۆ J,gu)TB !f?Ή%== ppp@dd$s훞?.ׯsobbquQ%ק6ƍQlllx$$$ !!!ݻ=zPdHMMETTT}TurӧOzȑ#}v@k(_|}\.\.>\""OG9V~144ԹN)UA]eddH ![m1 N8!_}}6RRRpcĈ |[CJJ
6m3g 11111xǔrz+W͛gϞ}$""""""""""""""GYmӯ4Z. ԧ{lmmallf?Ήjlmm}YtƔ:ykon>Ħ,99YgUGN3۷o x{{cժURFQѽȑ#
< n =z --
UVL&öm۰aÆPn] s]͚5sv)N< VTebbwww YS?SRJIS%fy>qDN]4e_^c lgffJSi/ukm,RFZ !t o
Bku~߿u]\\ ]M9>'Ͽ"hÃ'Obذa !CsLDDDDDDDDDDDDDDD*X $;Qr֮]ԩS s ]v%1bĈ>>Xh^~-9~G7Nmٳg!!!۷dtn([, w}ggge˖8uTttexyy i FU"99M4O?i}tt4 ={/P۶\r3f `
oFο>̙{{{DDDQFعs' d$u 37oVNUH
q|S1x` Yy x~lٲE*zS˖- G ŬY <Ν;C c
L&L&ֵhDϞ=Wjj*v܉-ZBg*VVVظq#
&]qqq;wT\6x`+VLoׯ_cذaظq# `ʔ)zw/.X5|L6M*挎Ɣ)Sh" ȑ#$m ~ 0b۶mRĵk0qʣb! HUQ,X R(zK#$!}~9&'Au]{gfϙfVճ>+04a*UlôziWRRz쩘կ__&M:}g7}tmذ!!orqQpp (< (< 4tM6/8 @ P5kj֭ر ӧOJNNts͛7k
bbbTlYuU[n5Wɍ1c?UF
I7W YzիWL+
_vOrssSÆ
5zhȑ#4hW988^z?~ >ڵkgxFzUB%''[}QHH3|ڱc,Y ;\rSbbV :h̙ھ}{5kvء^zrJJJaS `>>1'4hG>PƍׯO>}{t.\>HՓRSSUnu֙d5k˗g
6hܸq{(0ٳgkڵYU7n |/RruuURRjԨcj6Wɓ'kҥmNNNW>#Ϸ]jɒ%T~}%'']>K-ZHvv9xrrrҲe˴rJkNJUT)kNWŋh;v+www-YDNNNۿ3^Е+Wr # (,FڳBd
ׯT@8q~m5mTK;((HǏW }բEԧO -\0Oc+HrN5j#$;ǎw (mLvZc; y5 ` d'' [@\ @mٲE2\%
y,jObbQz;|]vm۶j֬Y~S̜9ӼǏ ! rUeʔO κqF~Q,'K,O ,J
0 5p e<@0
e.\q x! 0+// `> @
X C$` @ 9D X C$` @ 9 5^X Fpϡ ((
V " r, ! H " r, !P$bb(4443u-\Ќ'mW^~-X,
ڦo߾yz
P
ʔ)2e7)6qqqpo( @ P @.0 P0v )*I\ց/wH b,.\Ǔ۪T.]K.iӦM N0aM;b!|mۦ#Gw( (\ @A ]Ց#G4~x , ! H.\(Ţm۶IƏ /bSRR_e˖*Tݻg Y,)))IӦMSF#ŒѣGꫯN:>CVRjjk]|YU&)SF=zбc2mT-_\;wV
,???5lPFÇ3wI+Y\]]
h ϟנATR%9;;bŊׯLILLg}-ZWNNN*[:uuݶo~FrssԩkYXM>]͛77cX7oiӦVV*Ţ Q
%K* @+WmNɓնm[{rwwԩ7xCgΜ ŀC~2e(""BIIIfɭ%I]SΝd){{{yzzŋZrV\Çkʔ)6 ܹSb*3yd ;Lrqqo8qB˖-Sdd|||ҝȑ#߿\"777Iҕ+Wl2[N۷oC=^XXv۷||| o-_\/$IJLL~͛7OׯWڵ]3$$DZRdd}q5-\PWܹsm$>}ZO= 9"IX,˗vZ]V/fϞnJJ ?. ;vLG իi s:{$N>>>
}v73N:cvvvrrr۵m65*3H =zTG
?iӦy3 r&9 +`At钚4i"I>|.]dUTI4` I .SjΜ96駟Z`W $͞=[GVjj:vƍ
a.W͚5g *&&F7nTr Ν;krvvɓuEFF:>sթSǪ^HHz=c:x𠢣kת\r:{:t蠘ׯ_W.]ʕ+kÆ
sNUTI
َj۶ 9 ƍRTTO.͙3G }Qcǎ5Ǐ+&&FԷ~mڼ~fΞ=6mٳT.]ׯ+<<\7nБ#G$?? ߖ1chX,8q"##K.W^ɓuիO?T'NЍ7oj۶]@ݸq#OMHHPtt `>;
XȆ~MV$}'z饗ce˖u5ZJo+t牉ڵkաCs_R$I*E=zŋVrssS֭պukq)SF7n$AZ ;_~ѹsTbEE믿bhj߾9˗/ozӼJJJR5as-;;;uA+VN<9shfٳg̙3rrr?lB֣>M6Ku15o\6ly[C QժU3hĉzWp- M:U4rHM<ԩ:ud;vTT)fts:uhܸq9:-.\'|"Iz뭷oJ.>LQQQZds̜93> ~P
tAZJ{u̓&Ms} P81 VB-[LTbEyw%|ƍ3,Sn][\Rׯ_OфY1l03Vڵ$СCVǾ IR%_e>nĈfխׯgyF%,]TԽ{ OXlY6?|IСCnչsgyyy),,L3/ZH*UTOccc{eVWwʕ+,WWWD[`;vyn5f]v `>;iߗ X Yw^IR-l>vڪPΟ?{fhcټΝ;%I
6Trr#<~0'''kϞ=d31,#!!!2 CԪU+Zn˗JJJ$-[ڬ۲eKM4)Ӓn.jooo i><}6iܺu W)ˍ{*))IR3/+I5We^_~{n;wNʜ;w.Obvvvs > | wh; @AAʕ+
*dZbŊ:YJ.mK$IUTaciKKp\חY=099Y*S"""v簰,glKڹmQF?4W%̕bbbaR (x!Vk#s+?[)))G akֹ0$mܸLC)!!At.]!CH XȲn㙭teKٲe%|PdI9::f$혃J,i^3--:NٍD;&mvZ#ҥK%Imڴѧ~?]୫{ H$`vv7o[4jHuVfXرcf?4i"Iڻw.^9 Iz
40l6mڴIC^NNNz%lO[lْUNiyƍ&;DFGe7%JHΞ=TݻuNLLLe/ri_~
ð " &///IRTTT{!
B˰;#IUV Cd
2=m~I?SM6)S(...]?CV$X``$iŊ:~xW\ќ9sl^ka{߾}eoop7.Ӻf+ |drqqʕ+lٲZVj:4|5o\zJ.\PJê֬Y# UV𐧧7nP}6c?ƍڵkdI7Wڼyϟ yzz*&&Fe˖U]uVM2%ױ3F 5jHUfMSW6 W|}} Zڵbcc榆
jz ԑ#G4h U^] rppPz4~x >|Xk5/B
JNN飐gsc-YD ;vTrDUZU :t̙3}tu5k,رCzRʕ$0TN
0||bN4h@G |ƍSׯ_4}t=s-\P }ի'WWWh}Ǚ^E֭3;990 5kL˗/ |`6lؠq{ ÐfϞk> &3 _~RBCC;o߾Zh飅 f^ttU{vmٻ Pژzڵ P X C$` @ 9D X C$` @9w TQrw 䍠k A\ PT09 q?1 d
X C$` @ 9D X C$` @;*44TEE N)hk
f ULIRbb"##9" pV!pB ?^
J*ҥKt6mڔu5} T] ]x.E *y| n P9BQ \14~[.xQǏ/P PeuCVRj%.__]ժUʔ) =zرcEjj/_Ν;B
rvv6lQF;y^yլYSR
4aEGGgzkРATUbEOw$&&>S-+'''-[V:uҺun[~S~TF
KuQ~,oIIIZv^z%5jHʕJ.6mhɒ%23VNݫnݺ\rrqqQ54bEEEeX?~Z$iƍjꪺujĉϰ~PP,$IVғO>ҥKNAAAV߯^xAUTJ(&Mh̙JHH*b VRRR1b|
˰ (^ ; 2e(""BIIIrwwU{{{IҵkԹsg3Y^xV\+Wj2e+ @;w<==d4'Oرc$+։'l2EFF' 9ʕ+rss$]rE˖-Ӻu}v=Cꅅkھ}G
QHH ?o֪ /6JLL~͛7OׯWڵ]3$$DZRdd}q5-\PWܹsm$>}ZO= 9"IX,˗vZ]V/fϞnJJ ?. ;vLG իm&:e_UN߽⢫WjÆ
ڰa֬YKv~w}g}Va:yN+V(88Lg}^{MaG?okڼyJ(aa4}tY,uƌ6lLXڵKv҂ ?\r$ggg-YD
6{5vtyi岳W_}%___
@]tIM4$
>\.]*U$I0`䤏?Xъԅ ԿIԩS5g OuA-X@ъիWJfϞѣG+55U ;vu
ڰam&JWرfUuq(""Boնm[ό
7ڵkv횢 >H^^^Zbf͚y&M?Եke˖D:}}YdXիz7ԭ[79sFٳk6o>M>]F˗XO?hС2 C:u?(/ K/d +[ϟkiժUzշo_;OLL֮] :J*%IԨQ$I=zŋVrssS֭պukq)SF7n$AZ ;_~ѹsTbEE믿bhj߾9˗/ozӼJJJR5as-;;;uA+VN<9shfٳg̙3rrr?lB֣>M6Ku15o\6ly[C QժU3hĉzWpx M:UDɓ'[[:uZ*;nɒ%W˗W )SF?U_>*YZn={h{qqqj*[z_|Qk֬ў={ C>09;;J*n$5kLV2Wsrr?/ uQ;wԚ5kԭ[7<ƍtR :x𠼽գG_&Lm;gUBBoXL
n0#s @Ag9<
ls 0+(&V*-[&IX^| ˼n>oƍ[UխV\ׯ|[v
6LعUv$I:tձ/BԾ}tWDEEi#FW_yIҒ%K-]TԽ{ OXlY6?|IСCnչsgyyy),,L3/ZH*UƏ ˼#z)Iɓ'u%Fa_j\- 3[oeRZ~<[d;x =j^#-V :t04g UVM/_V>}4|p Ԓ%KliNL4I斶 ( E+`{$hj
*ڻwoV=kܹSaC+W.Gq>#wppΟ?srr#I62"0$LuZ| <$9::*11Lkٲͺ-[ԤI?N>-c!3J JӧOmέ[ppu͙3G? =(%%%+w9-[6sܮmviޗf͚exNm֯QJ.ᱴ:j[n=kx{{kɒ%jڴ~gsٳu= (Lr
r$B
XΟ?o7[I-Ր 6=Dp\חY)SFJNNrpsXXX⍋%3s =:wM>>>f˗%I6ϓYۤ u93Oig<^uM6M^z<_N9;;g' EN=Y0 txS60 Xٞfܷ:7s + I_A1$`jM9y`n5s+%%ѣUVܯ_?;wNUVՔ)SԲeK,Y{%K[nVWu;Yi[YXXs\?3iu` ԙ3gTB*UJ^{l >
GZQF[*5552ǎ 3[ ~lФII{ub֟$}YנA6olܦM$I= |Pe˖-ZĜ&7nܨlٳg͟ׯa6&X}o_2-#=ztsyeXwޑ$UVَ{Rrr r O7` IO?~)Ku|||ԦMIҔ)S UVIziu,00Pb
?~<]+WhΜ96?p@ILco߾Wxxƍi6? ~&NsM:5[_협{/dEbշ T:u$I'Nz$d~I}QQFEԩH^~e:u*q Bt3q$Gkk֬YfѥK4p@XB%1x{{?$-[L]tсqqq թS'EGGg>^-ZЬY n Һu /YfVukԨ#FH>C믿fgWڵUreIRo>خ],ŋzꩧd\Rݺutsgy&únnnڱc{9QW^yEԩS's>>PUVMt)UZ5VZUOւ Էo_caaazgY,y{{+99Y111n&|V-[^xAn$hTR%_^kN{պuks1777)&&F7oBSF
ڵvmsjjUZ5$)%%E>Ss Zm
ׯT X~uL rsst3]}B֭[`
V-$I~wﮤ$y{{+>>^ ʕ++88b 1b ,0TD ())IGBnYJ ǫy
%iƌ6ly(..μ x*_Y>yzz{y=úqƎ {/kNx 5A%15s5 ɇyTx @!O- (@ `Bf͚ںu:v(???:}H͛7k
bbbTlYuU[n0*ƌ?CT5$IYzիW`}j+7775lPG^`` 9AzJHHի&_IRFtA⋪P>}($$䶫3/_^;vВ%KԱcG+WL Z:t蠙3gjk֬Yڱczʕ++))IaN:0`z駵}v=SQrr|}}կ_?۷OODgKujڴi>@[nU h`,
XcA
X +`es @SLk^b> (X+0 9 X@riY,Y,իW/)6 8`{W xp 䞫ʔ)c7)~ ӵ? ( H@wVݺuuҥ A Kd Q}0 m` , ! A ͘sW~G s | 0 Pa, ! H " r, ! H r ^vn YS B\ p70~l a / sa, ! H " r, ! HPY,Y,w8y. Pp8w @QpB* @ V
5~xw( wTrLv|/^TPP.^O P|s r&/uc `K^9od, ْk.QJ XǏ. [q~b. @x lɫ9 9DK .bѶm$IǏbBCC)))/ԲeKY*TP3}|a@@,iӦQFbIWѣzWUN yzzCwzUV)55.__]ժUʔ) =zرcEjj/_Ν;B
rvv6lQF;y^yլYSR
4aEGGgzkРATUbEOw$&&>S-+'''-[V:uҺun[~S~TF
KuQ~,
/ (\]]ULEDD())I*coo/Iv:wl&KS/^ʕ+rJ
>\SLyxhΝrpp,Uɓ'kرfĉZl"##G QuI\e˖iݺuھ} ztԵkWm߾x($$DǏ ~kUoz ITbb߯kyZ~j!!!jժ"##%ki
ZzΝk %zꩧtIb._kjڵz5{tuSRR4tP}>www988رc:zV^Lc +`K.I&ÇҥKV[J$I Ppp +::Zp/I:u̙cz~ < (::Zz |AIٳ5zhcǎڿnܸp]~]6lP``2~?Y٣Xhƍ*W5xtuչsgm߾]ΚJWرfUuq(""Boնm[ (0Xƍ Nx-mˍ q Fn 72
X]oiժUO>D/ylٲ?]UVV߾}<111Zv:t`+U$)22RF$C/Z M[V֭mYLmܸQ$jJ:v_~EΝSŊ:-ү*ŢիW}V,_Mo*))I5jІ
ԡCUXQ:y̙ÇugϞ3gI?
Y>6mڔ.VӧOױcԼysmذA1ooo
2DUV3<'W_͏'Nhԩ#GjVVNԩS'חs/I}" #%I 39B /]00 *:x| p13X K-[&IX^| ˼n>oƍ[UխV\ׯQӧOOh¬6l|uvIt!c_|$}钯l%I#F0nU~}=3%KX [t${>lٲzm^CZ%_ݪsRXXg_hRSSUT) ?>I&VJ s | 0 h`,.ٻw$E6 WvmUPAϟ{3Lzl^cΝ
\r9G ptyEDDg I!Ijժr[uA%%%QfX˖-mmٲ&MnuiI7 iooo i><}6iܺu W)˪1chСa (T<ܣ|Դi|`. iqvy@1 *zf8q| pv|#c$`wɕ+W$I*Tȴ\Ŋuy.]fK.IT(%OOO d 7uo}}Iڣ2e(""BYo.\0 RqqqyΒ,gg\ dqt%)U4as @^H+mK VY 3 -s# (d2[)' ̭fn? =zTjV s 乴Ν;i㙭teKٲe%|PdI9::f$혃J,i^3-6:Nٍf; , |;FInݪ ;vLzᇳ C&M$I{ŋ]?' /I\AfmfM6I z!3I>i˖-[2_jUхى9MZ;oܸQٮ <%Ix =$\iyyw$IjժUc!CLk $O駟TGmڴ$M2Eqqq Zj$gϞV%I+Vսr̙c$͟?_4ֈ+{{{kܸq E X@ %LBwڵ$i5ktt%
8P+V$rqqv %I˖-S.]tx\\~GuI>-?6m*0ԵkWM2Eaaa .hƌ5jU'QڴiCIRSSO?}JNNV5h *VmV7o6~7jJc4l0=W-4k,ǣn: j֬U5jhĈ?P//xtt E %XO{_\97Nʕ˧ (>l 9:1 % 72f19@_ T||g&QرC+VԵkԩS'm۶Myzz***L >|LڶmƍLc4iz-3U2EFFGjժIN:UfxުUZ`ku,,,L<~Ib#Iԩ[z˖- /DI7WKLL4 WR%_^kN{պuks1777)&&F7oRVF
ڵvmsjjUZ5$)%%E>Ss "ZFJo,[ B?x*C6^v\0`.
\@b> Ț< ^as +`yf͚ںu:v(???:}%\j͚?驘-[V]v֭[3Lʮ1c?UF
IRbbj֬={jy`}j+7775lPG^`` 9AzJHHի&_IRFtA⋪P>}($$D\|yرCK,Qǎ U\9)11QUVU 4sLm߾=]]{{{͚5K;vP^Tre%%%0 թSG 0 LW:sc ___U\9@¿j V_\ p7;/ֹ| 5Ey + Ƚ>[Μ9ڵk+..Fnnn:z(IX Ř0믿/:zz0 1 Xij
܍X P " r, ! `60 @>a. .uiΝ9 GeʔQ t!r{QUF
f͚3fbcc%
8̙3Oj;w]ƍѣfe˖鯿ȑ#O?7?u!^ LZj:y}-[T*UgO>$5J~~~f (QBzҾ}aû ; Px]v'<==#hƍPY,-ZH J,C*99b$1BEEY~=VW5j
[ԯ__ 3r믿* @ 7oVZN:̙3VƎ T-_\#F'|z+OU]5}t_?LXڵkխ[7թSG~F 9sw9ޥKW#F .]())I4IڵkgZnǎ Zje 'G 9rjԨ-[^O+Wzu-X@ԦMݸqCӦMӨQTD\6nܨ Q/VϞ= ƍkŒmM
ҡCdڶm֭+U-ZoMqƒʕ+?7,;F )Ţ_~f0SNYf {V\\vޭ>}6K.VwMqqq:tPv/k͛ =zdx<&&F PnݬJ|f$թSGtܹ֝?cb;w>SUX12IIIf̞=;[ 3^+22R*_m˖.]2eH.^ r|_TT 0[Ί5}||~wrr$g\n:KzէO /ry vt
ۖr/_$3AEIIIVexJ*/0 |||dXr5%$$dɒf,hU.222˱ݻw[nӧ&L`õ|rOz衇b @GG _~LˮY+WM<$bŊJLLɓ'2[lyoժU+i PzrJi6mj"IG 5 =zTgϞr,rttՊXz)lRs̱Y>Ќ3pB=9
Xͤ-[UV%J($$D߿Yɓׯz衐M4IC Q%$Iڵ QFܹs裏ըUVj߾ﯓ'OG QDDV\e˖IԹsg[{5vXuL{GTR%
2D&MRtt>*U*'MڵkSfO0`-Zdu\6mUC {c^^^S$i3fzjժifի/G aH9
4ml-, 7
\@b> @^a> |a߾}$c߾} _-ZnnnѸqccӦMaƩS IƂ >}V矍u...FƍƸql7ѣG+W6 +*rJ 4eoq
2{5 ~axM6=ѧO L>}uZՋ4_o_~hРjH2njϤ
0[2ܚ7onrƂw'Ofccccccv [nm.`zc>v8+$$D
6Ծ}ԠAjժiŊ֭[An -QQQ*QΜ9#oo؋VJtYyyyw8 Q a~ʗ//;l?u=0'r6=0h+mXX8':o 7(lBf?x@ A {i f+ (X菂V%s{"hܣ
s6̽چq. 1 ({8o.b>&d)nݜ,0b]Z Ptܑ쥪U0\?~p۶mrtt?" ++`5lP{y.F wƍ'gg?
`?
/{aцG mX8oE}[tѷE}½p 3 ǝg1 @HH6l}Aw;.@ = 㣠;r`Y,y{{+**ط~+ŢP]rE8qs5lP?$)((H V-9;;|j۶ϟ ڻwo\zUԍ7:秖-[_~IW?T>}Tre9;;[M4ԩSut1{ѣGs~z衇{W ,HW
<==ձcG:u궯-::ZAAAʔ) :СCV5rH=rwwbQXXXܹs }QJ*cEGG,ӫW/? [x ս{LSre5nXM6ճ>/B٣Cf)ŋvڪP,Ց$;;;5nX7Vn+99Ys̑$]tI{VfoijڷoSjܸqy7n5jX w#hΜ9jѢ}]CYfŊ:t֬Y@uY?.^sfU'Ojĉz'ձcGOrqqg}fQDD6lؐ*]iLRJiҥz'4d
6L@N]~]AAAz!ooo=Ú6m;"%<<\ ,PUN Y+VTΝf͚۞>3C_9њ|e
GGGcƌV; 9rQlt6lhh"K3dxl$իUTx $c֭v +BCCU Il^~}#""",2 ̶dVڵkgfX ;fX-ŝeL2f[:99>>>Vt蓼j 4Ȫ]]]
WWW}C ɰ>Q_63ʔ)cs۾}{~ohoo[k*Uʬa5jӦp_H>M7ΐd8::fz\T1 Ͻܴ!wZt1f]jkܹ3oq'QtgWs7/(
XyiժUԩΜ9 8p@]vU͚5zjG>lUN: :諯ҪUc}
*gXԷo,]t4hf̘awz駵tR}ĉ}`f|I%&&*$$$r/$=sz甔+V
W%IP֭o^XF7nN<$ժU˪Nڵ%Iǎ $9ݹ #UQQQ:|yٳg[V-Y,3^ %''C
UrqF*..NK.߯{wEFrrgɓqbbbt)
0@n:
4(úם+>>^>he;_SO=˗/gў={ Hכo)ooozI[p>sIRnt )..Nǎ SN$I3fH/
[*UҥKln=8(Qx 1BK,QٲeTڵkz ZjiϞ=~bcc5k,9::jz7 ( rچi4i=Zj;x|iѿ[/)]lϟ7ɓ'~,øB$cǎ aÆ F\\/+`φ$cҥUݺuƍתUh֬Yx$ԩS3w/?șlݻwO>nu+:t`unݺ+w9G*T a8qª\rrϟ/_^ дiavکDZd$0 -[L]tko_Q+g :r䈞{9s_Ϟ=cǎ,6+GGG9::Zjںuf͚6mژe}0 z!X5yd=Sjٲ&NiӦ .\S?s_TLLUL斒,XsO?UŊ_-Z$IjѢE=zPj$I_~]jѢE$ݻu:uJoJ*3fܶ<}qg|WTlY}٪K佋/J z!988;zIR\
ݒ[[
rӆ( /ע1jܸs笎ŝgGі
]n?r& XJMM5 c/^Tҥyyy=55U ;vԎ ;4amݺU{Qv F(1/䤮]jŊJNNuܹ\=~P{fR7S6m(o^a aquu՞={w^*,,Ly|ў={g =ӷƳ>+cnKJ}\rPdիuK/V>}ndobbx_U$όX,mVaÆ[qvsk'ug
8P>}2-K_9iݻ[n>3 I7 OJJҁ$I52 ?uzxx,>En0 =[mxb|-_~#xNwj\ᳫߗp
Ge;Gvvvp2ih?- 'Mrt}VU߯ӧkj5j7nd7 U\Y}є)Sl&tKW\͛d.]ZZuׯ_/ggg5l0ahҥJMM}ݧ%JDz%)KYvvvjԨ6l*UκtY =zFQF*UT_WcǎYOVZ6m̙3??h۶mtLʗ/o&ٳG9wޭnݺO>0aBc$wwwUT)]Ǐ aVyѣJMM$6˥ t"""JlYpp<`L9s͛ժU+ -O_ 濬jذΜ9^zI*Uʔ) : LW>3^yI7={m ;~}Y?^ b ?
+]zU
6\]]u=wVc
8P
g
oge]<1~YEKBBBCC5k,=5jCfW*\% _~~wsNrwwף>/濦K{ ܭ=aaa
*<+W*hd;}ƌWjܹ UB-ZH+WT```矵zjW_tYi֭VۨQt! :t(1H7h===5tP%%% K.$31Y-ZH˖-SڵUjU]UcǎiӦM8p} ֨Q#Iǟi>
qΜ99=MvwYe&M@FnM`Pr ,I&I5k{,X'1pqqbU+22R=6o,ggg,[I|vowr\ᳫȏK?
#粝%I |N8VZiŊڴi>C}n&`=#?~V\oV :tyF 3gΨsZn>SV˧ժUK+Vѣ?hҥz'3hfUjԫW/mٲ%vvvѣ.]l=~̙3ڽ{vء+VhС yM:foF6lÇQ/kUٲeW_i֭jܸΝ۷kÆ
6m6o,OOO|kjÆ
=z Ν;[=֮]G7n/^[]ԨQZfl٢3f /-[f+WԦMjȐ!ڻwvޭݻwˮ\nV\if~ZrU#Fʕ+ٳl٢>HSLћoiׯ?,w[ oŋYfY Aڵk
2 v;ŝq
'NVXEFF$iyVϞO7xCW6 }
&$&&*&&&#
;˗q (>>^2i+/Xj8d
iިY>C ?~\
Wll֯_
0 {6mZ~/PⳲ.1h)[ʔ)c[h3grVe[I_ם W
n?r.G XM6Upp,gyFk֬Q*U2|jԨ}jzt$~ZbN8.]hZtտ,pvvիݻwћo͛+%%\3vLWի P#d| }QlRuy}ھ}<<<2+WK.SO=ŋ[e2DNo>խ[W&LPVԽ{wZJYn]\RzO?5khرZl96mիWkǎ jӦ/^y iΝ;W^yEڵ7|3KY:wΟ?'xB>?*+{O%IWݵ|rL5a;wN۷ԩS5~x
6, y?H>S= Q_ Tz4tK[6[nrtttqK.C=$IzYS``~iU\Y6lիWuUmذAuW_}% =ߒjҤ֯_N:I>3_*^ziĈ{IO>va( .8`>{YiC>+ P]tI111|N_N~`\%n
'VlҤ$z#].]K.V~[٧YIF
Ȱܽޛ
Y~L lc+L˯Y&cKݺu_ܘ1c4f̘,cǎرce,laFBCCTI&ڽ{w֕lحuw n3f_yz7doos!Sθj֬Ν;+cggÇWxxۧG y>CF˗/b:l֭մiSիWO'NЫ_~E;4uT}wJMMʆߧ>Vܧ9_[VLL6oެgy&úk^Vv?kĘU.]ZÆ
Sf裏wߕ~iIoqgQ<ݍqϮ/?/(
X +_mحu7F i>eԩz72,GѣG+<<\/jժ-y֒/[ \V-ԩc|iIɝpu$VWi\]]kIvء+WH?
Q|}}%IO>GSd>3l{G͟=|<چYge._ Ƭ_M6$os-,>;5U% _~~ws$`]Vvm=|riʖ-%Kޕ؊#Fhʔ)?0G_yԩSٳg36i$lھ#GJ/%KZMmuI">N8aZzuj֬i?
odxe
QpqO cVHos [I|v/ws\ᳫK?
#HM=$ΰaZ~$'kÇԩS%gĈ
IkG,?WVM}r'Q$߿V˗͟Ӗ?
9y${Uʕ%پOcccGrݻ͟=|<ۆYge._ Ƭ)m%[ w W*ڸ?# gH2gPl@n͛7ϐdX,c/[̐dH26mڔMÆ
3uԩYG=ƍ32#ŝ}v֬YxJJ
*)))1$o$AFRRR2F&M IF%dQ_z]t1$vvvƱcRd_*U IƸq2-[o 777ԩSO<ِdǏ 3PYiݣ#{9iC>+ 6`*zo;&nڴɰX,$cȑVxNⳣ˯qϮN_Qw7 9ShTbٺu!سg]*璓ɓ'wjTV>|qte =jt1܌z6lnܸaFժU
'''RJm1c!xꩧڊ@d<怘6˗/7 IFv9ҢcĈDb٪K=sufH2J*e\L9}ϚpBzI; W*% C~}w3E:ڵkƮ]U ?ppp0&Mdlٲ
瞳*waxg ~ذa ~mb}ݻwQT)hOwEe9iôVggggx
,]1=YEC'''LJ۪Ufdx(
¸gWu7/(+ X}ݗwypvv_M65}\]kyq
,^xh9 X_A6y 7
OOOaÆԩS`fkByY0;%%%Ř;w%K4
*=z0~L'y+55XbѩS'bŊlTTڵ?dZ(\诜3>3ԩcQV-Ǝ ;;!7_a\t2dQfM16mj̝;jy,'mfL:ڵq%K4 ///㡇 2^{5㹗6䳲p(waZ0f] Ɗ+W_}hԨQ|ypuu5*Wlt7oyxNⳣ)(
]n?b aÆڷo4h`uo߾ڻwfΜC-Z$///_f Q3g xծ][ǏW۶mqy3FׯWttTW^yE$jժzUvm}ᇊR-4w\IբE ٳG5$Y,MDVZl+W傂dXjZ5رc5rH ?~\oɓZhnv/J[lQLL~w}'z Iڽ{$)&&F
4*WS5IJJRHHT^xAԳ>K.YMIIk7|Sʕ5 !#""m6խ[Wt
P
6ĉf͛7e˖f-͛77$N:JJJZ*#[N::w
ԢE ,YRJNN7o={y %''J*^A_} :6mڨm۶z饗̚5K }f̘m۶>ӺuVJMM$GoZh>@={Lc9<==zj=ի-Z]vi˖-+Ww䔥s MW=-I'n&tQ]ӄ TF
wљ3g:s CS/!!=,,LwcZ kVk.V5|pM2Lz{_ޫ%JHJO5k(I2Ѣa> (joUN7nܰ*WT)]pnfdɒj۶}tnMZʎ9s׃>aÆEruu$ɓ'5|p 9R.\1cƘ abUo߾}d&L=*U6mdUnƍ\O?_j-[(%%E
6$͜9S[n z!5nX[n (
XjRŊ5zh(&&FƍS
2D_~ qۺ{?ĉ5hnh? {XǧT)ee_*@lZSrl*9mv] c.cwsfdacKߘk," ~>_ ĺ߮}]ޯ~麞^/ dS AwҥL&lmmob4 ;`ۗx5mڴ 7nL&Skocc#٬VZ^Svvv[o yxx^ {~[?CU> 5 Po&==> OO
qvvV:u$I:yvء ;v?ŋ n5gݺuz'tiu.^={hϞ=o$Ia7 E> @F. Tv (jٲeu @:vXWqFaٳgG E۷ ?n =zURRrss3g(//Ok֬Ѱav^ ? j6r @> *P P...2d}IR~~Νk={UV-*66V~٬N:>Çor F f# WXӦMӷ~[W?|5Dt
Qnݪ5tKV\\ 9RܔOM6d*w߾}zբE 999qzgur7g w}rvvVF4h 8q @Y/+ԵG ܹSۻ5 ,> P ;iӦ[nECEQf͒mƐ{Gj JOOWzz JYV^+%%Eiii2d~eee7І
sRן7o x=믿={h VڵnjzG a2L },HgƵhB~~~U M:u3fo߾
рKwVFFFs IDATNL%%%l6;&**J;wK/EGGk
_7ߔ`M2Eĉ?~g s9ճmVt!1BΝ @# f# TVLLzjvjݺnj3|py{{VsW^-ɤ={WYYY9sL&L&չ4sLl6G ɱR>}d6Ν;kVc gXZ%;wtqӧOVZ={m˖-
fgunyA͛վ}{fe˖O-s;pƍ
<<<4` ݥ>oz
hĈtR,R]z}%c/^Xae> U╙Y>C/?}ذa;%I<==եK9 jF> @F. TJuQ=sג%KΝ;[8p <yIIIzբE }WR>}emٲE]v]bVX3gjJMM /`? ;v裏>u)o߾ȎGeU={NK.j/$I_l6kJLLTzzw^dQF0 -\PSNٳgյkWխ[W .5x`WFF⢥Kʕ+եKZ =zllld
:T>cKBqƕ@^{5jJAAA[Q6m$5oqqqqJNNֲe (33So""" %ILL4-u8w\b+~> quNiWسg1yd{w:u2Ͽ l | @UllVhh}ǎ ?Z
ӪW,X {{{=zFpp ,-Zŋu yyy)--MV;#իWOaaa]ʻVYԷo_egg[G Ujj͛'I:6oެwyj[5Rzz-%IM6Uݺu_c=SZZdoooiԩ\]]i&EDDX_EZ~ǧBGΙ3Gyyyڿ~muQ7oVݺu˽Fy :T ?Ph}>|Ǝ +;;[cǎO<+W9 8;;N: [N 5iD :tЀ,GyVyʱc_kҤIw}1cwީ G> @F. GW 驣GZ###K>G...z}Z$I.\$ӳ8N>]ҭ[79;;kѢE%Kvѣ%B+ x}nnnJII~iyyy)$$Dw.1}顇 R_U^8cƌт tR:0
:T
رc ~Z ,ЪU @y1Bǎ ӱct?QrrKUq4d㏺;$]9 P etVNNNǏA{GGG_|o߮;w*..,^z:qDqԫW=G.=z`-ZHrvvtd2iJOO/r]=vcj۶V^<\R'Nxfi<==հaC߿9c M8QsQΝ =V>$~a Wyzz{TW 7 j6r @> T ̙3VG؝9sFkC=d5nРAڹs^yu]zGNLAAAڽ{giڵkT5K ;vh͚5ںu,}jݫ"op=:p@'((H˗/::0%%EyyyU%9x𠲲ԤI
K/iҤIzg5~rrrۭ322$B VŒI 5 5 >*]U^=iyZbt"0/[զMmܸQ (N͵~zh۶m:uTcvEiԩq#] tb?uTZJ}W_}
6hֆ
J]{ժUի>3jŚ1c yծ]9Gֱcԭ[7ZJIII߿ڶm
[VV4aR
:Tcƌ룏>Rhh<==v70a씕ee˖iٲe:~7UVhhuVu!˸yN(6x`͜9S Z~>=S{Z @eח_~)Ijܸq | \ 5O 4h>@'OVdd.\5kӳ؞={VE&N
wjӦV\Y\\\aRן @F. Jw2L2z5a
vQv׳:5܆ R%w;wnUdԡC Iٳ wh .(W!
Tw %Y|f͚UadҲeJ nݺĨ LX˗/msUw(6k,Vw -[wТE @
)SHV\)ɤ_jNnn -
FLL>Seffd2d2)&&Ʋ֖-[*gggfO'N_= >СC;C$iѢESݺuիW/U3#Ǎ:uh
fy&IU۴id2,/t!K͛ekkٳgW VjΜ9ZjFsICIIIVs>sIR~\cر
W&MemٲEcǎt*$$DfY/Vbbս{"q =ZZt"##5|p5JӧOה)S4sL_^<___TӔŋ߿bbbW_;Puԩ
3i$yyy)66VaܹsVΝ5dȐ*
AxI 8p@ӧOWDD$ *66VIII?)[[[IRRRz%WW2hڴׯ,=V 9r_Zvjٲ*<<2]vz%I/4c eee]?kΜ9JLLG}hɖ8Ըqc^ZO=TqttԼy3f(33R XѨQ駟Z wU\\ =dIҮ]}vŕ{?^7oVdd
u%]tIw}5jtaaa_ڪI&z,Wtw+//Okݺu(`cc ;Z}}}Xop6m_W|| A72\ pdw}͛kذajԨqF___YvcJJJRƍ-\g檰Pl 6mڴi#???͟?_ k.%$$h jٲ_gUvJ? @T K^^^4i >,٬G }Tϗ؞={jԩ0`@O?^xANRttΝ@mڴI UAA6l :YfSϥKTXXXn :hڵ~رcu ?^6662d^z%
>\TPP~ZmڴQ||$d2iܹjٲF iӦU dQ\պukeddUVZ7ЬYta988@[ɍ ;`Ǿ}o>͘1CÆ
R`
2D[n?QFU# UIֆ
bY Tw p *, $
wޛp[ R)X rrrSO=u3nNNN0 PLa%ufggɓ73 !oo ը, @l; ]Q D TX PIQ& IENDB` torchcodec-0.14.0/benchmarks/decoders/benchmark_readme_data.json 0000664 0000000 0000000 00000112035 15207546147 0024766 0 ustar 00root root 0000000 0000000 {
"experiments": [
{
"decoder": "torchcodec",
"description": "concurrency 10batch 50decode 10 -> 256x256",
"fps_median": 2.3072522931318598,
"fps_p25": 2.3072522931318598,
"fps_p75": 2.3072522931318598,
"frame_count": 500,
"iqr": 0.0,
"median": 216.7079870235175,
"type": "concurrency 10batch 50decode 10 -> 256x256",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec",
"description": "decode 10 uniform frames",
"fps_median": 1.477463454793549,
"fps_p25": 1.4981802700867375,
"fps_p75": 1.461590980076232,
"frame_count": 10,
"iqr": 0.1670952308923006,
"median": 6.76835692115128,
"type": "decode 10 uniform frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec",
"description": "decode 10 random frames",
"fps_median": 1.8105466778443378,
"fps_p25": 1.8194976229493673,
"fps_p75": 1.8088107961513857,
"frame_count": 10,
"iqr": 0.03247163351625204,
"median": 5.523193697445095,
"type": "decode 10 random frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec",
"description": "first 100 frames",
"fps_median": 123.45201452123612,
"fps_p25": 136.23413453401804,
"fps_p75": 120.55812232567067,
"frame_count": 100,
"iqr": 0.09544499963521957,
"median": 0.810031333938241,
"type": "first 100 frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[approx]",
"description": "concurrency 10batch 50decode 10 -> 256x256",
"fps_median": 2.3318413850180746,
"fps_p25": 2.3318413850180746,
"fps_p75": 2.3318413850180746,
"frame_count": 500,
"iqr": 0.0,
"median": 214.42281761206686,
"type": "concurrency 10batch 50decode 10 -> 256x256",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[approx]",
"description": "decode 10 uniform frames",
"fps_median": 1.5105623374510002,
"fps_p25": 1.5179390832868598,
"fps_p75": 1.4960534032246207,
"frame_count": 10,
"iqr": 0.09637371636927128,
"median": 6.6200511902570724,
"type": "decode 10 uniform frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[approx]",
"description": "decode 10 random frames",
"fps_median": 1.847505387951745,
"fps_p25": 1.8571342508460336,
"fps_p75": 1.8386277266726003,
"frame_count": 10,
"iqr": 0.054198560770601034,
"median": 5.4127041064202785,
"type": "decode 10 random frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[approx]",
"description": "first 100 frames",
"fps_median": 158.33267258277337,
"fps_p25": 162.8805442209975,
"fps_p75": 153.6270203896357,
"frame_count": 100,
"iqr": 0.03698029136285186,
"median": 0.6315815830603242,
"type": "first 100 frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda]",
"description": "concurrency 10batch 50decode 10 -> 256x256",
"fps_median": 7.4842138849368816,
"fps_p25": 7.4842138849368816,
"fps_p75": 7.4842138849368816,
"frame_count": 500,
"iqr": 0.0,
"median": 66.80728366225958,
"type": "concurrency 10batch 50decode 10 -> 256x256",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda]",
"description": "decode 10 uniform frames",
"fps_median": 1.239050280436723,
"fps_p25": 1.2411198876222282,
"fps_p75": 1.2277642069865577,
"frame_count": 10,
"iqr": 0.08764705387875438,
"median": 8.07069749943912,
"type": "decode 10 uniform frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda]",
"description": "decode 10 random frames",
"fps_median": 1.5565471547160319,
"fps_p25": 1.5597249511895048,
"fps_p75": 1.5557212587471598,
"frame_count": 10,
"iqr": 0.01649988442659378,
"median": 6.42447610385716,
"type": "decode 10 random frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda]",
"description": "first 100 frames",
"fps_median": 193.9253867102129,
"fps_p25": 196.31956600910664,
"fps_p75": 189.00639334945907,
"frame_count": 100,
"iqr": 0.01970905251801014,
"median": 0.5156622435897589,
"type": "first 100 frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda,approx]",
"description": "concurrency 10batch 50decode 10 -> 256x256",
"fps_median": 7.569243135860152,
"fps_p25": 7.569243135860152,
"fps_p75": 7.569243135860152,
"frame_count": 500,
"iqr": 0.0,
"median": 66.05680264532566,
"type": "concurrency 10batch 50decode 10 -> 256x256",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda,approx]",
"description": "decode 10 uniform frames",
"fps_median": 1.2739614041007767,
"fps_p25": 1.2759208005527078,
"fps_p75": 1.2538538385953066,
"frame_count": 10,
"iqr": 0.13793418556451797,
"median": 7.849531365558505,
"type": "decode 10 uniform frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda,approx]",
"description": "decode 10 random frames",
"fps_median": 1.6082342098712124,
"fps_p25": 1.6093958499100993,
"fps_p75": 1.6067390481860326,
"frame_count": 10,
"iqr": 0.010274268686771393,
"median": 6.217999802902341,
"type": "decode 10 random frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda,approx]",
"description": "first 100 frames",
"fps_median": 326.44677676721756,
"fps_p25": 327.9623921538813,
"fps_p75": 319.3140318092132,
"frame_count": 100,
"iqr": 0.008258320391178131,
"median": 0.30632864870131016,
"type": "first 100 frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchaudio",
"description": "concurrency 10batch 50decode 10 -> 256x256",
"fps_median": 0.24019753938959174,
"fps_p25": 0.24019753938959174,
"fps_p75": 0.24019753938959174,
"frame_count": 500,
"iqr": 0.0,
"median": 2081.6199919059873,
"type": "concurrency 10batch 50decode 10 -> 256x256",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchaudio",
"description": "decode 10 uniform frames",
"fps_median": 1.5540320730092783,
"fps_p25": 1.5587828651400377,
"fps_p75": 1.5462203380681485,
"frame_count": 10,
"iqr": 0.05212187208235264,
"median": 6.434873625636101,
"type": "decode 10 uniform frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchaudio",
"description": "decode 10 random frames",
"fps_median": 1.2288967434597657,
"fps_p25": 1.231097530553542,
"fps_p75": 1.2232565356917986,
"frame_count": 10,
"iqr": 0.0520668332464993,
"median": 8.137380177155137,
"type": "decode 10 random frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchaudio",
"description": "first 100 frames",
"fps_median": 134.49116467672755,
"fps_p25": 141.51284275959787,
"fps_p75": 130.22875071645763,
"frame_count": 100,
"iqr": 0.061229947954416275,
"median": 0.743543267250061,
"type": "first 100 frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec",
"description": "concurrency 10batch 50decode 10 -> 256x256",
"fps_median": 17.73409061829548,
"fps_p25": 17.799733186252308,
"fps_p75": 17.668930429974797,
"frame_count": 500,
"iqr": 0.20795202162116766,
"median": 28.19428471196443,
"type": "concurrency 10batch 50decode 10 -> 256x256",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec",
"description": "decode 10 uniform frames",
"fps_median": 14.969685302516243,
"fps_p25": 15.436814650753497,
"fps_p75": 14.827793275442701,
"frame_count": 10,
"iqr": 0.02660714788362384,
"median": 0.6680167149752378,
"type": "decode 10 uniform frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec",
"description": "decode 10 random frames",
"fps_median": 16.79105912633641,
"fps_p25": 17.36567128058909,
"fps_p75": 16.357802132242757,
"frame_count": 10,
"iqr": 0.035480327904224396,
"median": 0.5955550465732813,
"type": "decode 10 random frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec",
"description": "first 100 frames",
"fps_median": 154.55037049176087,
"fps_p25": 240.70872577355553,
"fps_p75": 145.64352122578927,
"frame_count": 100,
"iqr": 0.27116809040308,
"median": 0.6470382418483496,
"type": "first 100 frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[approx]",
"description": "concurrency 10batch 50decode 10 -> 256x256",
"fps_median": 17.827029984970764,
"fps_p25": 17.908065632446302,
"fps_p75": 17.746724421345395,
"frame_count": 500,
"iqr": 0.25383320543915033,
"median": 28.04729674104601,
"type": "concurrency 10batch 50decode 10 -> 256x256",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[approx]",
"description": "decode 10 uniform frames",
"fps_median": 14.62193988979937,
"fps_p25": 15.094490808610594,
"fps_p75": 14.16848480225286,
"frame_count": 10,
"iqr": 0.043298407923430204,
"median": 0.6839037826284766,
"type": "decode 10 uniform frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[approx]",
"description": "decode 10 random frames",
"fps_median": 17.064476084158482,
"fps_p25": 17.445494506660747,
"fps_p75": 16.47666937334179,
"frame_count": 10,
"iqr": 0.033704872243106365,
"median": 0.5860127173364162,
"type": "decode 10 random frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[approx]",
"description": "first 100 frames",
"fps_median": 427.21032217067926,
"fps_p25": 432.7140372206906,
"fps_p75": 405.96235006741597,
"frame_count": 100,
"iqr": 0.015228756237775087,
"median": 0.23407674115151167,
"type": "first 100 frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda]",
"description": "concurrency 10batch 50decode 10 -> 256x256",
"fps_median": 48.495390567883085,
"fps_p25": 49.089439887782476,
"fps_p75": 48.48358729869717,
"frame_count": 500,
"iqr": 0.12727823667228222,
"median": 10.310258235782385,
"type": "concurrency 10batch 50decode 10 -> 256x256",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda]",
"description": "decode 10 uniform frames",
"fps_median": 9.434494293855785,
"fps_p25": 9.449536373041624,
"fps_p75": 9.304849424220611,
"frame_count": 10,
"iqr": 0.016455440316349268,
"median": 1.0599402245134115,
"type": "decode 10 uniform frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda]",
"description": "decode 10 random frames",
"fps_median": 9.437257105564711,
"fps_p25": 9.44721404995691,
"fps_p75": 9.421506679130895,
"frame_count": 10,
"iqr": 0.002888241782784462,
"median": 1.0596299208700657,
"type": "decode 10 random frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda]",
"description": "first 100 frames",
"fps_median": 320.0237978512167,
"fps_p25": 323.16233493916013,
"fps_p75": 306.3065473827573,
"frame_count": 100,
"iqr": 0.017028328962624073,
"median": 0.3124767616391182,
"type": "first 100 frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda,approx]",
"description": "concurrency 10batch 50decode 10 -> 256x256",
"fps_median": 48.503194993180855,
"fps_p25": 49.07521500878542,
"fps_p75": 48.48521408903198,
"frame_count": 500,
"iqr": 0.12397986650466919,
"median": 10.308599259704351,
"type": "concurrency 10batch 50decode 10 -> 256x256",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda,approx]",
"description": "decode 10 uniform frames",
"fps_median": 9.473295226319413,
"fps_p25": 9.486161561653448,
"fps_p75": 9.396198138371846,
"frame_count": 10,
"iqr": 0.010093070100992918,
"median": 1.0555988978594542,
"type": "decode 10 uniform frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda,approx]",
"description": "decode 10 random frames",
"fps_median": 9.462767252922706,
"fps_p25": 9.477660546626687,
"fps_p75": 9.435246293419956,
"frame_count": 10,
"iqr": 0.004743047058582306,
"median": 1.0567733235657215,
"type": "decode 10 random frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda,approx]",
"description": "first 100 frames",
"fps_median": 326.44417845649633,
"fps_p25": 328.54173274045644,
"fps_p75": 311.36793665483043,
"frame_count": 100,
"iqr": 0.016788112465292215,
"median": 0.3063310869038105,
"type": "first 100 frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchaudio",
"description": "concurrency 10batch 50decode 10 -> 256x256",
"fps_median": 0.43204749593665775,
"fps_p25": 0.43204749593665775,
"fps_p75": 0.43204749593665775,
"frame_count": 500,
"iqr": 0.0,
"median": 1157.2801710516214,
"type": "concurrency 10batch 50decode 10 -> 256x256",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchaudio",
"description": "decode 10 uniform frames",
"fps_median": 3.333683383879292,
"fps_p25": 3.3845434447798723,
"fps_p75": 3.264147107712041,
"frame_count": 10,
"iqr": 0.10897916043177247,
"median": 2.999684987589717,
"type": "decode 10 uniform frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchaudio",
"description": "decode 10 random frames",
"fps_median": 4.0320476398428164,
"fps_p25": 4.08048815838108,
"fps_p75": 3.988325712721163,
"frame_count": 10,
"iqr": 0.056630611419677734,
"median": 2.480129426345229,
"type": "decode 10 random frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchaudio",
"description": "first 100 frames",
"fps_median": 147.47687944606517,
"fps_p25": 164.43478442136222,
"fps_p75": 139.82087368831037,
"frame_count": 100,
"iqr": 0.10705696232616901,
"median": 0.6780723892152309,
"type": "first 100 frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchcodec",
"description": "concurrency 10batch 50decode 10 -> 256x256",
"fps_median": 86.39723372823053,
"fps_p25": 87.03377280102,
"fps_p75": 85.7202156789611,
"frame_count": 500,
"iqr": 0.088033486623317,
"median": 5.787222326733172,
"type": "concurrency 10batch 50decode 10 -> 256x256",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchcodec",
"description": "decode 10 uniform frames",
"fps_median": 33.189626149822594,
"fps_p25": 33.82967232841853,
"fps_p75": 32.29903259547232,
"frame_count": 10,
"iqr": 0.014008307829499245,
"median": 0.30129896476864815,
"type": "decode 10 uniform frames",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchcodec",
"description": "decode 10 random frames",
"fps_median": 25.449082527813868,
"fps_p25": 26.79222726554132,
"fps_p75": 24.524487863611032,
"frame_count": 10,
"iqr": 0.03451313264667988,
"median": 0.3929414739832282,
"type": "decode 10 random frames",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchcodec",
"description": "first 100 frames",
"fps_median": 617.6635956687804,
"fps_p25": 628.2239831467241,
"fps_p75": 600.5979590421235,
"frame_count": 100,
"iqr": 0.007321836426854134,
"median": 0.16190042719244957,
"type": "first 100 frames",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchcodec[approx]",
"description": "concurrency 10batch 50decode 10 -> 256x256",
"fps_median": 87.75689657095042,
"fps_p25": 89.24487106466817,
"fps_p75": 87.25458127727161,
"frame_count": 500,
"iqr": 0.12779527995735407,
"median": 5.697557907551527,
"type": "concurrency 10batch 50decode 10 -> 256x256",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchcodec[approx]",
"description": "decode 10 uniform frames",
"fps_median": 36.322339809887275,
"fps_p25": 37.29793868919022,
"fps_p75": 35.170676938309825,
"frame_count": 10,
"iqr": 0.01621643453836441,
"median": 0.27531266026198864,
"type": "decode 10 uniform frames",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchcodec[approx]",
"description": "decode 10 random frames",
"fps_median": 26.979835656258963,
"fps_p25": 27.515479490594046,
"fps_p75": 25.92843393481319,
"frame_count": 10,
"iqr": 0.02224518498405814,
"median": 0.3706471798941493,
"type": "decode 10 random frames",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchcodec[approx]",
"description": "first 100 frames",
"fps_median": 643.5947586298682,
"fps_p25": 656.47484325684,
"fps_p75": 623.5290946988358,
"frame_count": 100,
"iqr": 0.008048678282648325,
"median": 0.15537727531045675,
"type": "first 100 frames",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda]",
"description": "concurrency 10batch 50decode 10 -> 256x256",
"fps_median": 15.345577075570482,
"fps_p25": 15.345577075570482,
"fps_p75": 15.345577075570482,
"frame_count": 500,
"iqr": 0.0,
"median": 32.58267822302878,
"type": "concurrency 10batch 50decode 10 -> 256x256",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda]",
"description": "decode 10 uniform frames",
"fps_median": 3.407761670388029,
"fps_p25": 3.4202746688103467,
"fps_p75": 3.3969136141529384,
"frame_count": 10,
"iqr": 0.020106985233724117,
"median": 2.9344775155186653,
"type": "decode 10 uniform frames",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda]",
"description": "decode 10 random frames",
"fps_median": 3.7973972922812367,
"fps_p25": 3.8124376376208704,
"fps_p75": 3.790402797717343,
"frame_count": 10,
"iqr": 0.01524831261485815,
"median": 2.6333826119080186,
"type": "decode 10 random frames",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda]",
"description": "first 100 frames",
"fps_median": 110.2995578623837,
"fps_p25": 111.15407175801678,
"fps_p75": 109.40982617706682,
"frame_count": 100,
"iqr": 0.0143425352871418,
"median": 0.9066219478845596,
"type": "first 100 frames",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda,approx]",
"description": "concurrency 10batch 50decode 10 -> 256x256",
"fps_median": 15.23259609859067,
"fps_p25": 15.23259609859067,
"fps_p75": 15.23259609859067,
"frame_count": 500,
"iqr": 0.0,
"median": 32.824345683678985,
"type": "concurrency 10batch 50decode 10 -> 256x256",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda,approx]",
"description": "decode 10 uniform frames",
"fps_median": 3.4718027074233855,
"fps_p25": 3.47798844584686,
"fps_p75": 3.4609039071190297,
"frame_count": 10,
"iqr": 0.014193371869623661,
"median": 2.88034800440073,
"type": "decode 10 uniform frames",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda,approx]",
"description": "decode 10 random frames",
"fps_median": 3.858107203058677,
"fps_p25": 3.8664557511567494,
"fps_p75": 3.8345751484790545,
"frame_count": 10,
"iqr": 0.021502859890460968,
"median": 2.5919445659965277,
"type": "decode 10 random frames",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchcodec[cuda,approx]",
"description": "first 100 frames",
"fps_median": 114.8213885831863,
"fps_p25": 115.2712004761549,
"fps_p75": 113.44311305477977,
"frame_count": 100,
"iqr": 0.013979705050587654,
"median": 0.8709178771823645,
"type": "first 100 frames",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchaudio",
"description": "concurrency 10batch 50decode 10 -> 256x256",
"fps_median": 10.21334669225676,
"fps_p25": 10.21334669225676,
"fps_p75": 10.21334669225676,
"frame_count": 500,
"iqr": 0.0,
"median": 48.955549543723464,
"type": "concurrency 10batch 50decode 10 -> 256x256",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchaudio",
"description": "decode 10 uniform frames",
"fps_median": 36.388329147307815,
"fps_p25": 37.55219240164325,
"fps_p75": 35.35600100896176,
"frame_count": 10,
"iqr": 0.016541380435228348,
"median": 0.27481338754296303,
"type": "decode 10 uniform frames",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchaudio",
"description": "decode 10 random frames",
"fps_median": 24.726120386350882,
"fps_p25": 25.75113744599497,
"fps_p75": 23.5940473221088,
"frame_count": 10,
"iqr": 0.03550335764884949,
"median": 0.4044306119903922,
"type": "decode 10 random frames",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchaudio",
"description": "first 100 frames",
"fps_median": 555.8913919266115,
"fps_p25": 567.8774810250327,
"fps_p75": 541.1648658024393,
"frame_count": 100,
"iqr": 0.008692249190062284,
"median": 0.17989125475287437,
"type": "first 100 frames",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchvision[video_reader]",
"description": "concurrency 10batch 50decode 10 -> 256x256",
"fps_median": 1.8541771100028126,
"fps_p25": 1.8541771100028126,
"fps_p75": 1.8541771100028126,
"frame_count": 500,
"iqr": 0.0,
"median": 269.6614025179297,
"type": "concurrency 10batch 50decode 10 -> 256x256",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchvision[video_reader]",
"description": "decode 10 uniform frames",
"fps_median": 0.18817676322761367,
"fps_p25": 0.18817676322761367,
"fps_p75": 0.18817676322761367,
"frame_count": 10,
"iqr": 0.0,
"median": 53.141524109989405,
"type": "decode 10 uniform frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchvision[video_reader]",
"description": "decode 10 random frames",
"fps_median": 0.14352962734215088,
"fps_p25": 0.14352962734215088,
"fps_p75": 0.14352962734215088,
"frame_count": 10,
"iqr": 0.0,
"median": 69.6720265019685,
"type": "decode 10 random frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchvision[video_reader]",
"description": "first 100 frames",
"fps_median": 62.11008734268979,
"fps_p25": 64.75018630104657,
"fps_p75": 60.40429858139817,
"frame_count": 100,
"iqr": 0.11111421789973974,
"median": 1.610044427216053,
"type": "first 100 frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_120s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchvision[video_reader]",
"description": "concurrency 10batch 50decode 10 -> 256x256",
"fps_median": 2.6822789710641577,
"fps_p25": 2.6822789710641577,
"fps_p75": 2.6822789710641577,
"frame_count": 500,
"iqr": 0.0,
"median": 186.4086492843926,
"type": "concurrency 10batch 50decode 10 -> 256x256",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchvision[video_reader]",
"description": "decode 10 uniform frames",
"fps_median": 0.28704203185592897,
"fps_p25": 0.28704203185592897,
"fps_p75": 0.28704203185592897,
"frame_count": 10,
"iqr": 0.0,
"median": 34.838103448972106,
"type": "decode 10 uniform frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchvision[video_reader]",
"description": "decode 10 random frames",
"fps_median": 0.3587892813609353,
"fps_p25": 0.3602738830879335,
"fps_p75": 0.3573168647927119,
"frame_count": 10,
"iqr": 0.22970355488359928,
"median": 27.871512666344643,
"type": "decode 10 random frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchvision[video_reader]",
"description": "first 100 frames",
"fps_median": 54.001491664575575,
"fps_p25": 56.83671582466259,
"fps_p75": 53.51161286891831,
"frame_count": 100,
"iqr": 0.10932716354727745,
"median": 1.851800698786974,
"type": "first 100 frames",
"video": "/tmp/torchcodec_benchmarking_videos/mandelbrot_1920x1080_10s_60fps_600gop_libx264_yuv420p.mp4"
},
{
"decoder": "torchvision[video_reader]",
"description": "concurrency 10batch 50decode 10 -> 256x256",
"fps_median": 46.16884016490128,
"fps_p25": 46.689432103577914,
"fps_p75": 46.065416577914036,
"frame_count": 500,
"iqr": 0.14506806060671806,
"median": 10.829815048724413,
"type": "concurrency 10batch 50decode 10 -> 256x256",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchvision[video_reader]",
"description": "decode 10 uniform frames",
"fps_median": 5.010414284300553,
"fps_p25": 5.045901248432863,
"fps_p75": 4.8791391096793335,
"frame_count": 10,
"iqr": 0.06773537024855614,
"median": 1.995842944830656,
"type": "decode 10 uniform frames",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchvision[video_reader]",
"description": "decode 10 random frames",
"fps_median": 2.696112087777776,
"fps_p25": 2.7403023596517895,
"fps_p75": 2.6260657275855532,
"frame_count": 10,
"iqr": 0.15874546021223068,
"median": 3.7090446073561907,
"type": "decode 10 random frames",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
},
{
"decoder": "torchvision[video_reader]",
"description": "first 100 frames",
"fps_median": 170.39584114627687,
"fps_p25": 173.51038231092855,
"fps_p75": 166.20399157926386,
"frame_count": 100,
"iqr": 0.02533587906509638,
"median": 0.5868687834590673,
"type": "first 100 frames",
"video": "/tmp/torchcodec_benchmarking_videos/nasa_960x540_206s_30fps_yuv420p.mp4"
}
],
"system_metadata": {
"cpu_count": 22,
"cuda": "NVIDIA PG509-210",
"machine": "x86_64",
"python_version": "3.12.8",
"system": "Linux"
}
}
torchcodec-0.14.0/benchmarks/decoders/benchmark_transforms.py 0000664 0000000 0000000 00000012722 15207546147 0024417 0 ustar 00root root 0000000 0000000 import math
from argparse import ArgumentParser
from pathlib import Path
from time import perf_counter_ns
import torch
from torch import Tensor
from torchcodec.decoders import VideoDecoder
from torchvision.transforms import v2
def bench(f, *args, num_exp, warmup=1) -> Tensor:
for _ in range(warmup):
f(*args)
times = []
for _ in range(num_exp):
start = perf_counter_ns()
f(*args)
end = perf_counter_ns()
times.append(end - start)
return torch.tensor(times).float()
def report_stats(times: Tensor, unit: str = "ms", prefix: str = "") -> float:
mul = {
"ns": 1,
"µs": 1e-3,
"ms": 1e-6,
"s": 1e-9,
}[unit]
times = times * mul
std = times.std().item()
med = times.median().item()
mean = times.mean().item()
min = times.min().item()
max = times.max().item()
print(
f"{prefix:<45} {med = :.2f}, {mean = :.2f} +- {std:.2f}, {min = :.2f}, {max = :.2f} - in {unit}"
)
def torchvision_resize(
path: Path, pts_seconds: list[float], dims: tuple[int, int], num_threads: int
) -> Tensor:
decoder = VideoDecoder(
path, seek_mode="approximate", num_ffmpeg_threads=num_threads
)
raw_frames = decoder.get_frames_played_at(pts_seconds)
transformed_frames = v2.Resize(size=dims)(raw_frames.data)
assert len(transformed_frames) == len(pts_seconds)
return transformed_frames
def torchvision_crop(
path: Path, pts_seconds: list[float], dims: tuple[int, int], num_threads: int
) -> Tensor:
decoder = VideoDecoder(
path, seek_mode="approximate", num_ffmpeg_threads=num_threads
)
raw_frames = decoder.get_frames_played_at(pts_seconds)
transformed_frames = v2.CenterCrop(size=dims)(raw_frames.data)
assert len(transformed_frames) == len(pts_seconds)
return transformed_frames
def decoder_resize(
path: Path, pts_seconds: list[float], dims: tuple[int, int], num_threads: int
) -> Tensor:
decoder = VideoDecoder(
path,
transforms=[v2.Resize(size=dims)],
seek_mode="approximate",
num_ffmpeg_threads=num_threads,
)
transformed_frames = decoder.get_frames_played_at(pts_seconds).data
assert len(transformed_frames) == len(pts_seconds)
return transformed_frames.data
def decoder_crop(
path: Path, pts_seconds: list[float], dims: tuple[int, int], num_threads: int
) -> Tensor:
decoder = VideoDecoder(
path,
transforms=[v2.CenterCrop(size=dims)],
seek_mode="approximate",
num_ffmpeg_threads=num_threads,
)
transformed_frames = decoder.get_frames_played_at(pts_seconds).data
assert len(transformed_frames) == len(pts_seconds)
return transformed_frames
def main():
parser = ArgumentParser()
parser.add_argument("--path", type=str, help="path to file", required=True)
parser.add_argument(
"--num-exp",
type=int,
default=5,
help="number of runs to average over",
)
parser.add_argument(
"--num-threads",
type=int,
default=1,
help="number of threads to use; 0 means FFmpeg decides",
)
parser.add_argument(
"--total-frame-fractions",
nargs="+",
type=float,
default=[0.005, 0.01, 0.05, 0.1],
)
parser.add_argument(
"--input-dimension-fractions",
nargs="+",
type=float,
default=[0.5, 0.25, 0.125],
)
args = parser.parse_args()
path = Path(args.path)
metadata = VideoDecoder(path).metadata
duration = metadata.duration_seconds
print(
f"Benchmarking {path.name}, duration: {duration}, codec: {metadata.codec}, averaging over {args.num_exp} runs:"
)
input_height = metadata.height
input_width = metadata.width
for num_fraction in args.total_frame_fractions:
num_frames_to_sample = math.ceil(metadata.num_frames * num_fraction)
print(
f"Sampling {num_fraction * 100}%, {num_frames_to_sample}, of {metadata.num_frames} frames"
)
uniform_timestamps = [
i * duration / num_frames_to_sample for i in range(num_frames_to_sample)
]
for dims_fraction in args.input_dimension_fractions:
dims = (int(input_height * dims_fraction), int(input_width * dims_fraction))
times = bench(
torchvision_resize,
path,
uniform_timestamps,
dims,
args.num_threads,
num_exp=args.num_exp,
)
report_stats(times, prefix=f"torchvision_resize({dims})")
times = bench(
decoder_resize,
path,
uniform_timestamps,
dims,
args.num_threads,
num_exp=args.num_exp,
)
report_stats(times, prefix=f"decoder_resize({dims})")
times = bench(
torchvision_crop,
path,
uniform_timestamps,
dims,
args.num_threads,
num_exp=args.num_exp,
)
report_stats(times, prefix=f"torchvision_crop({dims})")
times = bench(
decoder_crop,
path,
uniform_timestamps,
dims,
args.num_threads,
num_exp=args.num_exp,
)
report_stats(times, prefix=f"decoder_crop({dims})")
print()
if __name__ == "__main__":
main()
torchcodec-0.14.0/benchmarks/decoders/benchmark_wav_decoder.py 0000664 0000000 0000000 00000015416 15207546147 0024506 0 ustar 00root root 0000000 0000000 import io
import os
import subprocess
from time import perf_counter_ns
import soundfile as sf
import torch
from torchcodec.decoders import AudioDecoder, WavDecoder
def bench(f, *args, num_exp=100, warmup=0, **kwargs):
for _ in range(warmup):
f(*args, **kwargs)
times = []
for _ in range(num_exp):
start = perf_counter_ns()
f(*args, **kwargs)
end = perf_counter_ns()
times.append(end - start)
return torch.tensor(times).float()
def report_stats(times, unit="ms"):
mul = {
"ns": 1,
"µs": 1e-3,
"ms": 1e-6,
"s": 1e-9,
}[unit]
times = times * mul
std = times.std().item()
med = times.median().item()
print(f"{med = :.2f}{unit} +- {std:.2f}")
return med
WAV_DIR = "/tmp/wav_files"
FORMATS = {
"u8": ("pcm_u8", "int16"),
"s16": ("pcm_s16le", "int16"),
"s24": ("pcm_s24le", "int32"),
"s32": ("pcm_s32le", "int32"),
"float32": ("pcm_f32le", "float32"),
"float64": ("pcm_f64le", "float64"),
}
DURATIONS = {
"10s": 10,
"5min": 300,
}
def generate_wav_files():
os.makedirs(WAV_DIR, exist_ok=True)
for fmt_name, (codec, _) in FORMATS.items():
for dur_name, dur_seconds in DURATIONS.items():
path = os.path.join(WAV_DIR, f"{fmt_name}_{dur_name}.wav")
if os.path.exists(path):
continue
print(f"Generating {path} ...")
subprocess.run(
[
"ffmpeg",
"-y",
"-f",
"lavfi",
"-i",
f"sine=frequency=440:duration={dur_seconds}",
"-c:a",
codec,
path,
],
check=True,
capture_output=True,
)
def decode_torchcodec(raw_bytes):
decoder = WavDecoder(raw_bytes)
return decoder.get_all_samples()
def decode_audio_decoder(raw_bytes):
decoder = AudioDecoder(raw_bytes)
return decoder.get_all_samples().data
def decode_soundfile(raw_bytes):
data, sr = sf.read(io.BytesIO(raw_bytes), dtype="float32")
return data
def decode_soundfile_native(raw_bytes, native_dtype):
data, sr = sf.read(io.BytesIO(raw_bytes), dtype=native_dtype)
return data
def validate_results():
print("Validating WavDecoder vs AudioDecoder outputs...")
for fmt_name in FORMATS:
for dur_name in DURATIONS:
path = os.path.join(WAV_DIR, f"{fmt_name}_{dur_name}.wav")
with open(path, "rb") as f:
raw_bytes = f.read()
wav_out = decode_torchcodec(raw_bytes)
audio_out = decode_audio_decoder(raw_bytes)
torch.testing.assert_close(wav_out.data, audio_out.data, atol=0, rtol=0)
print(f" {fmt_name}/{dur_name}: OK")
print("All validations passed!\n")
def main():
generate_wav_files()
validate_results()
results = []
for fmt_name, (_, native_dtype) in FORMATS.items():
for dur_name in DURATIONS:
path = os.path.join(WAV_DIR, f"{fmt_name}_{dur_name}.wav")
with open(path, "rb") as f:
raw_bytes = f.read()
num_exp = 10 if dur_name == "5min" else 100
print(f"\n=== {fmt_name} / {dur_name} ({path}) ===")
print(" WavDecoder: ", end="")
tc_times = bench(decode_torchcodec, raw_bytes, num_exp=num_exp, warmup=2)
tc_med = report_stats(tc_times)
print(" AudioDecoder: ", end="")
ad_times = bench(decode_audio_decoder, raw_bytes, num_exp=num_exp, warmup=2)
ad_med = report_stats(ad_times)
print(" soundfile (dtype=float32): ", end="")
sf_times = bench(decode_soundfile, raw_bytes, num_exp=num_exp, warmup=2)
sf_med = report_stats(sf_times)
print(f" soundfile (dtype={native_dtype}): ", end="")
sfn_times = bench(
decode_soundfile_native,
raw_bytes,
native_dtype,
num_exp=num_exp,
warmup=2,
)
sfn_med = report_stats(sfn_times)
results.append((fmt_name, dur_name, tc_med, ad_med, sf_med, sfn_med))
print("\n" + "=" * 155)
print("SUMMARY")
print("=" * 155)
print(
f"{'format':<10} {'duration':<10} "
f"{'WavDec (ms)':>12} {'AudioDec (ms)':>14} {'sndfile f32 (ms)':>17} {'sndfile native (ms)':>20} "
f"{'AudioDec/WavDec':>16} {'sndfile f32/WavDec':>19} {'sndfile nat/WavDec':>19}"
)
print("-" * 155)
for fmt_name, dur_name, tc_med, ad_med, sf_med, sfn_med in results:
audio_over_wav = ad_med / tc_med if tc_med > 0 else float("inf")
sf_over_wav = sf_med / tc_med if tc_med > 0 else float("inf")
sfn_over_wav = sfn_med / tc_med if tc_med > 0 else float("inf")
print(
f"{fmt_name:<10} {dur_name:<10} "
f"{tc_med:>12.2f} {ad_med:>14.2f} {sf_med:>17.2f} {sfn_med:>20.2f} "
f"{audio_over_wav:>15.2f}x {sf_over_wav:>18.2f}x {sfn_over_wav:>18.2f}x"
)
def bench_input_types():
print("\n" + "=" * 100)
print("FILE vs FILE-LIKE vs BYTES")
print("=" * 100)
print(
f"{'format':<10} {'duration':<10} "
f"{'file (ms)':>12} {'file-like (ms)':>16} {'bytes (ms)':>12} "
f"{'flike/file':>12} {'bytes/file':>12}"
)
print("-" * 100)
for fmt_name in FORMATS:
for dur_name in DURATIONS:
path = os.path.join(WAV_DIR, f"{fmt_name}_{dur_name}.wav")
num_exp = 10 if dur_name == "5min" else 100
with open(path, "rb") as f:
raw_bytes = f.read()
def decode_file():
return WavDecoder(path).get_all_samples()
def decode_filelike():
with open(path, "rb") as f:
return WavDecoder(f).get_all_samples()
def decode_bytes():
return WavDecoder(raw_bytes).get_all_samples()
file_med = (
bench(decode_file, num_exp=num_exp, warmup=2).median().item() * 1e-6
)
flike_med = (
bench(decode_filelike, num_exp=num_exp, warmup=2).median().item() * 1e-6
)
bytes_med = (
bench(decode_bytes, num_exp=num_exp, warmup=2).median().item() * 1e-6
)
flike_ratio = flike_med / file_med if file_med > 0 else float("inf")
bytes_ratio = bytes_med / file_med if file_med > 0 else float("inf")
print(
f"{fmt_name:<10} {dur_name:<10} "
f"{file_med:>12.2f} {flike_med:>16.2f} {bytes_med:>12.2f} "
f"{flike_ratio:>11.2f}x {bytes_ratio:>11.2f}x"
)
if __name__ == "__main__":
main()
bench_input_types()
torchcodec-0.14.0/benchmarks/decoders/generate_readme_chart.py 0000664 0000000 0000000 00000001167 15207546147 0024500 0 ustar 00root root 0000000 0000000 # Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the BSD-style license found in the
# LICENSE file in the root directory of this source tree.
import json
from pathlib import Path
from benchmark_decoders_library import plot_data
def main() -> None:
data_json = Path(__file__).parent / "benchmark_readme_data.json"
with open(data_json, "r") as read_file:
data_from_file = json.load(read_file)
output_png = Path(__file__).parent / "benchmark_readme_chart.png"
plot_data(data_from_file, output_png)
if __name__ == "__main__":
main()
torchcodec-0.14.0/benchmarks/decoders/generate_readme_data.py 0000664 0000000 0000000 00000007057 15207546147 0024314 0 ustar 00root root 0000000 0000000 # Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the BSD-style license found in the
# LICENSE file in the root directory of this source tree.
import json
import os
import platform
import shutil
from pathlib import Path
import torch
from benchmark_decoders_library import (
BatchParameters,
DataLoaderInspiredWorkloadParameters,
generate_videos,
retrieve_videos,
run_benchmarks,
TorchAudioDecoder,
TorchCodecPublic,
TorchVision,
)
NASA_URL = "https://download.pytorch.org/torchaudio/tutorial-assets/stream-api/NASAs_Most_Scientifically_Complex_Space_Observatory_Requires_Precision-MP4_small.mp4"
def main() -> None:
"""Benchmarks the performance of a few video decoders on synthetic videos"""
videos_dir_path = "/tmp/torchcodec_benchmarking_videos"
if not os.path.exists(videos_dir_path):
shutil.rmtree(videos_dir_path, ignore_errors=True)
os.makedirs(videos_dir_path)
resolutions = ["1920x1080"]
encodings = ["libx264"]
patterns = ["mandelbrot"]
fpses = [60]
gop_sizes = [600]
durations = [10, 120]
pix_fmts = ["yuv420p"]
ffmpeg_path = "ffmpeg"
generate_videos(
resolutions,
encodings,
patterns,
fpses,
gop_sizes,
durations,
pix_fmts,
ffmpeg_path,
videos_dir_path,
)
urls_and_dest_paths = [
(NASA_URL, f"{videos_dir_path}/nasa_960x540_206s_30fps_yuv420p.mp4")
]
retrieve_videos(urls_and_dest_paths)
decoder_dict = {}
decoder_dict["torchcodec"] = TorchCodecPublic()
decoder_dict["torchcodec[approx]"] = TorchCodecPublic(seek_mode="approximate")
if torch.cuda.is_available():
decoder_dict["torchcodec[cuda]"] = TorchCodecPublic(device="cuda")
decoder_dict["torchcodec[cuda,approx]"] = TorchCodecPublic(
device="cuda", seek_mode="approximate"
)
decoder_dict["torchvision[video_reader]"] = TorchVision("video_reader")
decoder_dict["torchaudio"] = TorchAudioDecoder()
# These are the number of uniform seeks we do in the seek+decode benchmark.
num_samples = 10
video_files_paths = list(Path(videos_dir_path).glob("*.mp4"))
assert len(video_files_paths) == 3, "Expected exactly 3 videos"
results = run_benchmarks(
decoder_dict,
video_files_paths,
num_samples,
num_sequential_frames_from_start=[100],
min_runtime_seconds=30,
benchmark_video_creation=False,
dataloader_parameters=DataLoaderInspiredWorkloadParameters(
batch_parameters=BatchParameters(batch_size=50, num_threads=10),
resize_height=256,
resize_width=256,
resize_device="cuda" if torch.cuda.is_available() else "cpu",
),
)
data_for_writing = {
"experiments": results,
"system_metadata": {
"cpu_count": os.cpu_count(),
"system": platform.system(),
"machine": platform.machine(),
"python_version": str(platform.python_version()),
"cuda": (
torch.cuda.get_device_properties(0).name
if torch.cuda.is_available()
else "not available"
),
},
}
data_json = Path(__file__).parent / "benchmark_readme_data.json"
with open(data_json, "w") as write_file:
json.dump(data_for_writing, write_file, sort_keys=True, indent=4)
if __name__ == "__main__":
main()
torchcodec-0.14.0/benchmarks/decoders/gpu_benchmark.py 0000664 0000000 0000000 00000016321 15207546147 0023013 0 ustar 00root root 0000000 0000000 import argparse
import time
from concurrent.futures import ThreadPoolExecutor
from pathlib import Path
import torch
import torch.utils.benchmark as benchmark
import torchvision.transforms.v2.functional as F
from torchcodec._core import get_next_frame
from torchcodec._core.ops import _add_video_stream, create_from_file
RESIZED_WIDTH = 256
RESIZED_HEIGHT = 256
def transfer_and_resize_frame(frame, resize_device_string):
# This should be a no-op if the frame is already on the target device.
frame = frame.to(resize_device_string)
frame = F.resize(frame, (RESIZED_HEIGHT, RESIZED_WIDTH))
return frame
def decode_full_video(video_path, decode_device_string, resize_device_string):
# We use the core API instead of SimpleVideoDecoder because the core API
# allows us to natively resize as part of the decode step.
print(f"{decode_device_string=} {resize_device_string=}")
decoder = create_from_file(video_path)
num_threads = None
if "cuda" in decode_device_string:
num_threads = 1
resize_spec = ""
if "native" in resize_device_string:
resize_spec = f"resize, {RESIZED_HEIGHT}, {RESIZED_WIDTH}"
_add_video_stream(
decoder,
stream_index=-1,
device=decode_device_string,
num_threads=num_threads,
transform_specs=resize_spec,
)
start_time = time.time()
frame_count = 0
while True:
try:
frame, *_ = get_next_frame(decoder)
if resize_device_string != "none" and "native" not in resize_device_string:
frame = transfer_and_resize_frame(frame, resize_device_string)
frame_count += 1
except Exception as e:
print("EXCEPTION", e)
break
end_time = time.time()
elapsed = end_time - start_time
fps = frame_count / (end_time - start_time)
print(
f"****** DECODED full video {decode_device_string=} {frame_count=} {elapsed=} {fps=}"
)
return frame_count, end_time - start_time
def decode_videos_using_threads(
video_path,
decode_device_string,
resize_device_string,
num_videos,
num_threads,
use_multiple_gpus,
):
executor = ThreadPoolExecutor(max_workers=num_threads)
for i in range(num_videos):
actual_decode_device = decode_device_string
if "cuda" in decode_device_string and use_multiple_gpus:
actual_decode_device = f"cuda:{i % torch.cuda.device_count()}"
executor.submit(
decode_full_video, video_path, actual_decode_device, resize_device_string
)
executor.shutdown(wait=True)
def main():
parser = argparse.ArgumentParser()
parser.add_argument(
"--devices",
default="cuda:0,cpu",
type=str,
help="Comma-separated devices to test decoding on.",
)
parser.add_argument(
"--resize_devices",
default="cuda:0,cpu,native,none",
type=str,
help="Comma-separated devices to test preroc (resize) on. Use 'none' to specify no resize.",
)
parser.add_argument(
"--video",
type=str,
default=str(Path(__file__).parent / "../../test/resources/nasa_13013.mp4"),
)
parser.add_argument(
"--use_torch_benchmark",
action=argparse.BooleanOptionalAction,
default=True,
help=(
"Use pytorch benchmark to measure decode time with warmup and "
"autorange. Without this we just run one iteration without warmup "
"to measure the cold start time."
),
)
parser.add_argument(
"--num_threads",
type=int,
default=1,
help="Number of threads to use for decoding. Only used when --use_torch_benchmark is set.",
)
parser.add_argument(
"--num_videos",
type=int,
default=50,
help="Number of videos to decode in parallel. Only used when --num_threads is set.",
)
parser.add_argument(
"--use_multiple_gpus",
action=argparse.BooleanOptionalAction,
default=True,
help=("Use multiple GPUs to decode multiple videos in multi-threaded mode."),
)
args = parser.parse_args()
video_path = args.video
if not args.use_torch_benchmark:
for device in args.devices.split(","):
print("Testing on", device)
decode_full_video(video_path, device)
return
resize_devices = args.resize_devices.split(",")
resize_devices = [d for d in resize_devices if d != ""]
if len(resize_devices) == 0:
resize_devices.append("none")
label = "Decode+Resize Time"
results = []
for decode_device_string in args.devices.split(","):
for resize_device_string in resize_devices:
decode_label = decode_device_string
if "cuda" in decode_label:
# Shorten "cuda:0" to "cuda"
decode_label = "cuda"
resize_label = resize_device_string
if "cuda" in resize_device_string:
# Shorten "cuda:0" to "cuda"
resize_label = "cuda"
print("decode_device", decode_device_string)
print("resize_device", resize_device_string)
if args.num_threads > 1:
t = benchmark.Timer(
stmt="decode_videos_using_threads(video_path, decode_device_string, resize_device_string, num_videos, num_threads, use_multiple_gpus)",
globals={
"decode_device_string": decode_device_string,
"video_path": video_path,
"decode_full_video": decode_full_video,
"decode_videos_using_threads": decode_videos_using_threads,
"resize_device_string": resize_device_string,
"num_videos": args.num_videos,
"num_threads": args.num_threads,
"use_multiple_gpus": args.use_multiple_gpus,
},
label=label,
description=f"threads={args.num_threads} work={args.num_videos} video={Path(video_path).name}",
sub_label=f"D={decode_label} R={resize_label} T={args.num_threads} W={args.num_videos}",
).blocked_autorange()
results.append(t)
else:
t = benchmark.Timer(
stmt="decode_full_video(video_path, decode_device_string, resize_device_string)",
globals={
"decode_device_string": decode_device_string,
"video_path": video_path,
"decode_full_video": decode_full_video,
"resize_device_string": resize_device_string,
},
label=label,
description=f"video={Path(video_path).name}",
sub_label=f"D={decode_label} R={resize_label}",
).blocked_autorange()
results.append(t)
compare = benchmark.Compare(results)
compare.print()
print("Key: D=Decode, R=Resize T=threads W=work (number of videos to decode)")
print("Native resize is done as part of the decode step")
print("none resize means there is no resize step -- native or otherwise")
if __name__ == "__main__":
main()
torchcodec-0.14.0/benchmarks/decoders/memprofile_decoders.py 0000664 0000000 0000000 00000003064 15207546147 0024215 0 ustar 00root root 0000000 0000000 # Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the BSD-style license found in the
# LICENSE file in the root directory of this source tree.
import argparse
import importlib
import torch
from memory_profiler import profile
from torchcodec._core import get_next_frame
from torchcodec._core.ops import add_video_stream, create_from_file
torch._dynamo.config.cache_size_limit = 100
torch._dynamo.config.capture_dynamic_output_shape_ops = True
@profile
def torchcodec_create_next(video_file):
video_decoder = create_from_file(video_file)
add_video_stream(video_decoder)
get_next_frame(video_decoder)
return video_decoder
def get_video_path_str(filename: str) -> str:
resource = importlib.resources.files(__package__).joinpath(filename)
with importlib.resources.as_file(resource) as path:
return str(path)
def main() -> None:
"""Memory leak check and profiling for decoders."""
parser = argparse.ArgumentParser()
parser.add_argument(
"--iterations",
help="Number of times to invoke decoder operations.",
type=int,
default=10,
)
args = parser.parse_args()
large_video_path = get_video_path_str("853.mp4")
# We call the same function several times, and each call will produce memory stats on
# standard out. We rely on a human looking at the output to see if memory increases
# on each run.
for _ in range(args.iterations):
torchcodec_create_next(large_video_path)
if __name__ == "__main__":
main()
torchcodec-0.14.0/benchmarks/encoders/ 0000775 0000000 0000000 00000000000 15207546147 0017643 5 ustar 00root root 0000000 0000000 torchcodec-0.14.0/benchmarks/encoders/bench_multi_stream_vs_single_stream.py 0000664 0000000 0000000 00000007051 15207546147 0027510 0 ustar 00root root 0000000 0000000 """
Basic benchmark aimed at validating that the new Encoder class isn't slower than
the existing VideoEncoder and AudioEncoder.
"""
import os
import tempfile
from time import perf_counter_ns
import torch
from torchcodec.encoders import AudioEncoder, Encoder, VideoEncoder
def bench(f, *args, num_exp=100, warmup=0, **kwargs):
for _ in range(warmup):
f(*args, **kwargs)
times = []
for _ in range(num_exp):
start = perf_counter_ns()
f(*args, **kwargs)
end = perf_counter_ns()
times.append(end - start)
return torch.tensor(times).float()
def report_stats(times, unit="ms"):
mul = {
"ns": 1,
"µs": 1e-3,
"ms": 1e-6,
"s": 1e-9,
}[unit]
times = times * mul
std = times.std().item()
med = times.median().item()
print(f"{med = :.2f}{unit} +- {std:.2f}")
return med
# --- Config ---
height, width = 256, 256
frame_rate = 30
num_channels = 2
sample_rate = 48000
durations_s = [3, 10]
def encode_video_with_video_encoder(frames, dest):
encoder = VideoEncoder(frames, frame_rate=frame_rate)
encoder.to_file(dest)
def encode_video_with_streaming_encoder(frames, dest):
encoder = Encoder()
video_stream = encoder.add_video(
height=frames.shape[2],
width=frames.shape[3],
frame_rate=frame_rate,
device=str(frames.device),
)
with encoder.open_file(dest):
video_stream.add_frames(frames)
def encode_audio_with_audio_encoder(samples, dest):
encoder = AudioEncoder(samples, sample_rate=sample_rate)
encoder.to_file(dest)
def encode_audio_with_streaming_encoder(samples, dest):
encoder = Encoder()
audio_stream = encoder.add_audio(sample_rate=sample_rate, num_channels=num_channels)
with encoder.open_file(dest):
audio_stream.add_samples(samples)
if __name__ == "__main__":
tmpdir = tempfile.mkdtemp()
devices = ["cpu"]
if torch.cuda.is_available():
devices.append("cuda")
for duration_s in durations_s:
print(f"\n{'=' * 50}")
print(f"Duration: {duration_s}s")
print(f"{'=' * 50}")
num_frames = frame_rate * duration_s
video_frames = torch.randint(
0, 256, (num_frames, 3, height, width), dtype=torch.uint8
)
num_samples = sample_rate * duration_s
audio_samples = torch.randn(num_channels, num_samples)
# --- Video benchmarks ---
for device in devices:
frames = video_frames.to(device)
dest = os.path.join(tmpdir, "video.mp4")
print(f"\n--- Video encoding ({device}) ---")
print("VideoEncoder: ", end="")
times = bench(
encode_video_with_video_encoder, frames, dest, num_exp=10, warmup=2
)
report_stats(times)
print("StreamingEncoder:", end="")
times = bench(
encode_video_with_streaming_encoder, frames, dest, num_exp=10, warmup=2
)
report_stats(times)
# --- Audio benchmarks ---
print("\n--- Audio encoding (cpu) ---")
dest = os.path.join(tmpdir, "audio.wav")
print("AudioEncoder: ", end="")
times = bench(
encode_audio_with_audio_encoder, audio_samples, dest, num_exp=10, warmup=2
)
report_stats(times)
print("StreamingEncoder:", end="")
times = bench(
encode_audio_with_streaming_encoder,
audio_samples,
dest,
num_exp=10,
warmup=2,
)
report_stats(times)
torchcodec-0.14.0/benchmarks/encoders/benchmark_encoders.py 0000664 0000000 0000000 00000016230 15207546147 0024033 0 ustar 00root root 0000000 0000000 #!/usr/bin/env python3
import shutil
import subprocess
import tempfile
from argparse import ArgumentParser
from pathlib import Path
from time import perf_counter_ns
import pynvml
import torch
from torchcodec.decoders import VideoDecoder
from torchcodec.encoders import VideoEncoder
pynvml.nvmlInit()
handle = pynvml.nvmlDeviceGetHandleByIndex(0)
FRAME_RATE = 30
DEFAULT_VIDEO_PATH = "test/resources/nasa_13013.mp4"
# Alternatively, run this command to generate a longer test video:
# ffmpeg -f lavfi -i testsrc2=duration=600:size=1280x720:rate=30 -c:v libx264 -pix_fmt yuv420p test/resources/testsrc2_10min.mp4
def bench(f, average_over=50, warmup=2, gpu_monitoring=False, **f_kwargs):
for _ in range(warmup):
f(**f_kwargs)
times = []
utilizations = []
memory_usage = []
for _ in range(average_over):
start = perf_counter_ns()
f(**f_kwargs)
end = perf_counter_ns()
times.append(end - start)
if gpu_monitoring:
util = pynvml.nvmlDeviceGetEncoderUtilization(handle)[0]
mem_info = pynvml.nvmlDeviceGetMemoryInfo(handle)
mem_used = mem_info.used / (1_000_000) # Convert bytes to MB
utilizations.append(util)
memory_usage.append(mem_used)
times_tensor = torch.tensor(times).float()
return times_tensor, {
"utilization": torch.tensor(utilizations).float() if gpu_monitoring else None,
"memory_used": torch.tensor(memory_usage).float() if gpu_monitoring else None,
}
def report_stats(times, num_frames, nvenc_metrics=None, prefix="", unit="ms"):
fps = num_frames * 1e9 / times.median()
mul = {
"ns": 1,
"µs": 1e-3,
"ms": 1e-6,
"s": 1e-9,
}[unit]
unit_times = times * mul
med = unit_times.median().item()
max = unit_times.max().item()
print(f"\n{prefix} {med = :.2f} {unit}, {max = :.2f} {unit}, fps = {fps:.1f}")
if nvenc_metrics is not None:
mem_used_max = nvenc_metrics["memory_used"].max().item()
mem_used_median = nvenc_metrics["memory_used"].median().item()
util_max = nvenc_metrics["utilization"].max().item()
print(
f"GPU memory used: med = {mem_used_median:.1f} MB, max = {mem_used_max:.1f} MB"
)
print(
f"NVENC utilization: med = {nvenc_metrics['utilization'].median():.1f}%, max = {util_max:.1f}%"
)
def encode_torchcodec(frames, output_path, device="cpu"):
encoder = VideoEncoder(frames=frames, frame_rate=FRAME_RATE)
if device == "cuda":
encoder.to_file(dest=output_path, codec="h264_nvenc", extra_options={"qp": 0})
else:
encoder.to_file(dest=output_path, codec="libx264", crf=0)
def write_raw_frames(frames, raw_path):
# Convert NCHW to NHWC for raw video format
raw_frames = frames.permute(0, 2, 3, 1)
with open(raw_path, "wb") as f:
f.write(raw_frames.cpu().numpy().tobytes())
def encode_ffmpeg_cli(
frames, raw_path, output_path, device="cpu", skip_write_frames=False
):
# Write frames during benchmarking function by default unless skip_write_frames flag used
if not skip_write_frames:
write_raw_frames(frames, raw_path)
ffmpeg_cmd = [
"ffmpeg",
"-y",
"-f",
"rawvideo",
"-pix_fmt",
"rgb24",
"-s",
f"{frames.shape[3]}x{frames.shape[2]}",
"-r",
str(FRAME_RATE),
"-i",
raw_path,
"-c:v",
"h264_nvenc" if device == "cuda" else "libx264",
"-pix_fmt",
"yuv420p",
]
ffmpeg_cmd.extend(["-qp", "0"] if device == "cuda" else ["-crf", "0"])
ffmpeg_cmd.extend([str(output_path)])
subprocess.run(ffmpeg_cmd, check=True, capture_output=True)
def main():
parser = ArgumentParser()
parser.add_argument(
"--path", type=str, help="Path to input video file", default=DEFAULT_VIDEO_PATH
)
parser.add_argument(
"--average-over",
type=int,
default=30,
help="Number of runs to average over",
)
parser.add_argument(
"--max-frames",
type=int,
default=None,
help="Maximum number of frames to decode for benchmarking. By default, all frames will be decoded.",
)
parser.add_argument(
"--skip-write-frames",
action="store_true",
help="Do not write raw frames in FFmpeg CLI benchmarks",
)
args = parser.parse_args()
decoder = VideoDecoder(str(args.path))
frames = decoder.get_frames_in_range(start=0, stop=args.max_frames).data
cuda_available = torch.cuda.is_available()
if not cuda_available:
print("CUDA not available. GPU benchmarks will be skipped.")
print(
f"Benchmarking {len(frames)} frames from {Path(args.path).name} over {args.average_over} runs:"
)
gpu_frames = frames.cuda() if cuda_available else None
print(
f"Decoded {frames.shape[0]} frames of size {frames.shape[2]}x{frames.shape[3]}"
)
temp_dir = Path(tempfile.mkdtemp())
raw_frames_path = temp_dir / "input_frames.raw"
# If skip_write_frames is True, we will not benchmark the time it takes to write the frames.
# Here, we still write the frames for FFmpeg to use!
if args.skip_write_frames:
write_raw_frames(frames, str(raw_frames_path))
if cuda_available:
# Benchmark torchcodec on GPU
gpu_output = temp_dir / "torchcodec_gpu.mp4"
times, nvenc_metrics = bench(
encode_torchcodec,
frames=gpu_frames,
output_path=str(gpu_output),
device="cuda",
gpu_monitoring=True,
average_over=args.average_over,
)
report_stats(
times, frames.shape[0], nvenc_metrics, prefix="VideoEncoder on GPU"
)
# Benchmark FFmpeg CLI on GPU
ffmpeg_gpu_output = temp_dir / "ffmpeg_gpu.mp4"
times, nvenc_metrics = bench(
encode_ffmpeg_cli,
frames=gpu_frames,
raw_path=str(raw_frames_path),
output_path=str(ffmpeg_gpu_output),
device="cuda",
gpu_monitoring=True,
skip_write_frames=args.skip_write_frames,
average_over=args.average_over,
)
prefix = "FFmpeg CLI on GPU "
report_stats(times, frames.shape[0], nvenc_metrics, prefix=prefix)
# Benchmark torchcodec on CPU
cpu_output = temp_dir / "torchcodec_cpu.mp4"
times, _nvenc_metrics = bench(
encode_torchcodec,
frames=frames,
output_path=str(cpu_output),
device="cpu",
average_over=args.average_over,
)
report_stats(times, frames.shape[0], prefix="VideoEncoder on CPU")
# Benchmark FFmpeg CLI on CPU
ffmpeg_cpu_output = temp_dir / "ffmpeg_cpu.mp4"
times, _nvenc_metrics = bench(
encode_ffmpeg_cli,
frames=frames,
raw_path=str(raw_frames_path),
output_path=str(ffmpeg_cpu_output),
device="cpu",
skip_write_frames=args.skip_write_frames,
average_over=args.average_over,
)
prefix = "FFmpeg CLI on CPU "
report_stats(times, frames.shape[0], prefix=prefix)
shutil.rmtree(temp_dir, ignore_errors=True)
if __name__ == "__main__":
main()
torchcodec-0.14.0/benchmarks/samplers/ 0000775 0000000 0000000 00000000000 15207546147 0017667 5 ustar 00root root 0000000 0000000 torchcodec-0.14.0/benchmarks/samplers/benchmark_samplers.py 0000664 0000000 0000000 00000007170 15207546147 0024106 0 ustar 00root root 0000000 0000000 import argparse
from pathlib import Path
from time import perf_counter_ns
import torch
from torchcodec.decoders import VideoDecoder
from torchcodec.samplers import (
clips_at_random_indices,
clips_at_random_timestamps,
clips_at_regular_indices,
clips_at_regular_timestamps,
)
DEFAULT_VIDEO_PATH = Path(__file__).parent / "../../test/resources/nasa_13013.mp4"
DEFAULT_NUM_EXP = 30
def bench(f, *args, num_exp, warmup=0, seed, **kwargs):
for _ in range(warmup):
f(*args, **kwargs)
num_frames = None
times = []
for _ in range(num_exp):
if seed is not None:
torch.manual_seed(seed)
start = perf_counter_ns()
clips = f(*args, **kwargs)
end = perf_counter_ns()
times.append(end - start)
num_frames = (
clips.data.shape[0] * clips.data.shape[1]
) # should be constant across calls
return torch.tensor(times).float(), num_frames
def report_stats(times, num_frames, unit="ms"):
fps = num_frames * 1e9 / torch.median(times)
mul = {
"ns": 1,
"µs": 1e-3,
"ms": 1e-6,
"s": 1e-9,
}[unit]
times = times * mul
std = times.std().item()
med = times.median().item()
print(f"{med = :.2f}{unit} +- {std:.2f} med fps = {fps:.1f}")
return med, fps
def sample(decoder, sampler, **kwargs):
return sampler(
decoder,
num_frames_per_clip=10,
**kwargs,
)
def run_sampler_benchmarks(device, video, num_experiments, torch_seed):
for num_clips in (1, 50):
print("-" * 10)
print(f"{num_clips = }")
print("clips_at_random_indices ", end="")
decoder = VideoDecoder(video, device=device)
times, num_frames = bench(
sample,
decoder,
clips_at_random_indices,
num_clips=num_clips,
num_exp=num_experiments,
warmup=2,
seed=torch_seed,
)
report_stats(times, num_frames, unit="ms")
print("clips_at_regular_indices ", end="")
times, num_frames = bench(
sample,
decoder,
clips_at_regular_indices,
num_clips=num_clips,
num_exp=num_experiments,
warmup=2,
seed=torch_seed,
)
report_stats(times, num_frames, unit="ms")
print("clips_at_random_timestamps ", end="")
times, num_frames = bench(
sample,
decoder,
clips_at_random_timestamps,
num_clips=num_clips,
num_exp=num_experiments,
warmup=2,
seed=torch_seed,
)
report_stats(times, num_frames, unit="ms")
print("clips_at_regular_timestamps ", end="")
seconds_between_clip_starts = 13 / num_clips # approximate. video is 13s long
times, num_frames = bench(
sample,
decoder,
clips_at_regular_timestamps,
seconds_between_clip_starts=seconds_between_clip_starts,
num_exp=num_experiments,
warmup=2,
seed=torch_seed,
)
report_stats(times, num_frames, unit="ms")
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--device", type=str, default="cpu")
parser.add_argument("--video", type=str, default=str(DEFAULT_VIDEO_PATH))
parser.add_argument("--num_experiments", type=int, default=DEFAULT_NUM_EXP)
parser.add_argument("--torch_seed", type=int)
args = parser.parse_args()
run_sampler_benchmarks(
args.device, args.video, args.num_experiments, args.torch_seed
)
if __name__ == "__main__":
main()
torchcodec-0.14.0/docs/ 0000775 0000000 0000000 00000000000 15207546147 0014654 5 ustar 00root root 0000000 0000000 torchcodec-0.14.0/docs/Makefile 0000664 0000000 0000000 00000003221 15207546147 0016312 0 ustar 00root root 0000000 0000000 # Minimal makefile for Sphinx documentation
#
ifneq ($(EXAMPLES_PATTERN),)
EXAMPLES_PATTERN_OPTS := -D sphinx_gallery_conf.filename_pattern="$(EXAMPLES_PATTERN)"
endif
# You can set these variables from the command line.
SPHINXOPTS = -W -j auto $(EXAMPLES_PATTERN_OPTS)
SPHINXBUILD = sphinx-build
SPHINXPROJ = torchcodec
SOURCEDIR = source
BUILDDIR = build
# Put it first so that "make" without argument is like "make help".
help:
@$(SPHINXBUILD) -M help "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O)
docset: html
doc2dash --name $(SPHINXPROJ) --icon $(SOURCEDIR)/_static/img/pytorch-logo-flame.png --enable-js --online-redirect-url http://pytorch.org/vision/ --force $(BUILDDIR)/html/
# Manually fix because Zeal doesn't deal well with `icon.png`-only at 2x resolution.
cp $(SPHINXPROJ).docset/icon.png $(SPHINXPROJ).docset/icon@2x.png
convert $(SPHINXPROJ).docset/icon@2x.png -resize 16x16 $(SPHINXPROJ).docset/icon.png
html-noplot: # Avoids running the gallery examples, which may take time
$(SPHINXBUILD) -D plot_gallery=0 -b html "${SOURCEDIR}" "$(BUILDDIR)"/html
@echo
@echo "Build finished. The HTML pages are in $(BUILDDIR)/html."
clean:
rm -rf $(BUILDDIR)/*
rm -rf $(SOURCEDIR)/generated_examples/ # sphinx-gallery
rm -rf $(SOURCEDIR)/gen_modules/ # sphinx-gallery
rm -rf $(SOURCEDIR)/sg_execution_times.rst # sphinx-gallery
rm -rf $(SOURCEDIR)/generated/ # autosummary
.PHONY: help Makefile docset
# Catch-all target: route all unknown targets to Sphinx using the new
# "make mode" option. $(O) is meant as a shortcut for $(SPHINXOPTS).
%: Makefile
@$(SPHINXBUILD) -M $@ "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O)
torchcodec-0.14.0/docs/requirements.txt 0000664 0000000 0000000 00000000502 15207546147 0020135 0 ustar 00root root 0000000 0000000 sphinx==7.2.6
-e git+https://github.com/pytorch/pytorch_sphinx_theme.git@pytorch_sphinx_theme2#egg=pytorch_sphinx_theme2
sphinx-gallery>=0.14.0
sphinx_design>=0.6.1
sphinx_copybutton
sphinx-tabs
sphinx-sitemap>=2.7.1
sphinxcontrib-mermaid>=1.0.0
docutils>=0.18.1,<0.21
matplotlib
torchvision
ipython
fsspec
aiohttp
joblib
torchcodec-0.14.0/docs/source/ 0000775 0000000 0000000 00000000000 15207546147 0016154 5 ustar 00root root 0000000 0000000 torchcodec-0.14.0/docs/source/_static/ 0000775 0000000 0000000 00000000000 15207546147 0017602 5 ustar 00root root 0000000 0000000 torchcodec-0.14.0/docs/source/_static/img/ 0000775 0000000 0000000 00000000000 15207546147 0020356 5 ustar 00root root 0000000 0000000 torchcodec-0.14.0/docs/source/_static/img/card-background.svg 0000664 0000000 0000000 00000000543 15207546147 0024127 0 ustar 00root root 0000000 0000000
torchcodec-0.14.0/docs/source/_static/img/generic-pytorch-logo.png 0000664 0000000 0000000 00000105700 15207546147 0025127 0 ustar 00root root 0000000 0000000 PNG
IHDR X X f IDATx }ul 9 A 0%T$eE["EJ,ϒUeoo9)X()|mYV(f19 @HL
3ݯ흝 ~?3= ?zq B!T
!BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !BB!DH` !B8'R{ B!8$0(O%!B x>|UB}
'Z PEޟψES8 z!#1j.>C%*abشy^yaC;gF&@8ڪx >1zuT$biXzp o-PCeB}۷{7^ף>|L4ӧOǴBC<O$!1 wރo X߈7mo3# N `rdtuuaԩL8Q> %&r!tPC9
%ԃχ'rW Xs+(%
X+[÷h`[d c4<{eRţw-Ƅ8eERTPi (^m<)):Ll (H`5q0Qz*%»vD? %h v JoZcf$G@yo]P9J3-ڋla_ CM@ 1p (;@H\M($?ߋQ/ »^_p6nbC&.ev(_
^|E˶k\ =xKQGG&OI&kr;IDVb՛ ç*dj3 f.E"P H ymxg0t?? \͡~hkoܹsY UBʏIb 5X9SQ1Pϝwx&0W Iܛ..a [oW_ZJWq qZ[ q?҅)]S,bƌ3| |Xr\6zYٸ[֬Y'x饗> >
hpoboý'67~w}?o>O x 1R,W 1$c)GD\] #P, $[L&tiY|7
nTU2
*rlp:o<,Z s 9x fΙO a>Ľkoc=f"xqVbi
z{biXB Ukv?ٶ1tm"&0Xc2u7pտ h
z!yW1rӦM3Ew^Iuwi)?5O=߽pB,;0}8qЁ)(.5?SLɎnWZ~-yLAUE[[bDMPCk(3H >B/JuR m;IIgkb5}r,[ o{pyCE9-ɢ̶<g=N}wܵn+
yn,"K`,X"5*+L~>Ja{0qUz:R ցFDTDTvtitO#j "5vZ_
j&OJĽnkiʹOÉfbːGP ;zj~Kpo9s2+ 777-[y-zŜ5}&f1Tuom} z}HvŒQ& X {
z 4^
7Q_u%FRN}Ӯ8zeqhV~x1" %4-- J:;'w&|-#wa TO..ch٢[nk NJWc ;!q?v'`γSCuW$NZ]
"w0 j's!
OB!Fkb UJq*n`ōV
{"*yQMs1ɼLX @"F-09Jch"{% d~^=؉My@D7 +;WY=N=Zj_xgqf}#.3q3T+QXu
aɌ Vm۶9wV]}&oT̗J7]%' X@GT Η=tS"pP%bV&
|O<WFibĢ**Gg=:-Bi|.Y blhզ 4P kVR /̙337Hyi@*߶c
;9isBe'\q2gL[nJƿ}}}uDpo?\v٥ؼyFQ̝oAغaժUUTJA.I^#<G uT4q16C+wY)͒U')0k!f*P[o_DkM;`MӇ_*8MêkUf/'}C@ ޤ hI ;to ??Ʊ kK=Da ;RwE;
yf<)<EIoY{%˒a>>|_ƴSqI'3mV8 vaYu{B("[;5y&z ˕·ܖYiE{qꩧEttvص[BT
182QG1)Ĉ:,^ML$
$4\Ure9{!{nu`
>U_'@Q
T9kT4s!KuCt"7y}[k_G2Tyt0עs2&u!b0y!MDahbN
*@tLtVlݲ/mx ֭qgEV5 l:qab9l2fһ Oe"W;-yi?Xn+'?(ݹ~d{m`{!3v_gv H`H+#'+r; 3+nLx*Uo:`! X0s ,KQkYݏbEWGP
-Q]6 6l;2sy.-;ʶ 1sM@:SΥ- ;(iSO;r-oV-NZqa!Λn ?OL>ݬfU&h4kDZئiɑF*T
9kQ ) RJ(JQ ય~%:O
ѤW&U#I./\YCW :VQղdE5+Wz3*GSxPkG{9WsIqO/1[2| Z9 җ/8 _;7-h_,.ʉzT=q|#p;e\vg}7H\" ZDV,XtBr7/ [-)4&X!]x j qLWTMP}ˈ#:x^W ZET.!n 7oF{Y^
(F,0:m. tAX`
[n@G 7n܈_8qa&֗0+FF>lBLtD/*{hj(ݵ w xw_2EVEIbnAy .4Y:GqU>ĀP헽תoojɊCD0 (.sĊq^A g/| WZ+»n[_FVo
_Bc H>y ~3{)IQ"( _~xTly>QR/Bg څa[[ oA m6WdO/~ ʑH,땣>JpnqaX?m4|ӟe]ffYV
1 _: |AkVld2h J^<WiHhIBL,$(,OidûiѤΕQ&Я~D4 [#~vMj\EgUV~ UZ#ڻw5VJc
.CxgI멭̪ +
( šc3ZO-,BʌAZ(Ys (<,XY[M'f "- .(v8rpǰ73&v2 k{KxH[ rb w+ d
$)M^ =;?C&eNlQ4ne:!jqA(k7Iz5oӣ1W}?#|fBZN`փV-jBc.XFоeZ&)5#QaiX=^t>/n{ Mg98"t XJ.&A=A"yz!jQΰrY-N>@4Cq<]Aa Cz FeK6իVWWW_K/% N:$+C|cɆ!kX8(|Y}89Gv|Mjc*;;"[8 ]'Gv=^Yu4v^6(OYTXTQ JD³
{!xܽ-yl^9PΏ
>q'yk&DY
ފaj^crM|/5$1CΏ P ՠn};?|-Ԃf!5>m , `+ xe^0/FI`x/}KGT7Dx;"!ہisOb kk(
&kC;8YzMeMŏ9QbɢIqOAZmv((|IlٺfJ&Ԛ-7hƹއ_us5{pt櫪{I?P(&!nr$ /db*yMjgσ7pnMIJ)[bE= "ā\,QgKhd{&vnd~il79Oq~w?C%}8{\AL7;SQ e4r(WHws49ɗ"Mnbe>zQgv ._ R{{;U>'=lG`.O~]s9ǚ/۽ j,0Mb?@y8H Bc*XX ڏ=Qj%S=T(&a!uKc$p, \SN9
@#rk6:Xfl%($;>pL[qe/G
ԞxXRSLVf |k5'(! 0t$̟V$w٘1};n{~:igY6llmIf]lnl|e.6m ^#sgC9g L:
r7?9[r"o^ꞌe6]8wg\yHRv+Y^;&^^8 bwX L
,rbא% rp|xFLAUg} K@; P]rBcs&Wl?kR^2*%6vc(bL *B/iZ0G=?G3xKOH3
6hD5rN㈑j-˕~ .caQ!qÁV
6`ٖ֭m[vtw[ˤ4,2?'rln>tPP&rn׆ IDATz&| ȉ ޝBAbZRL=#x3$ӓ4# NcZ\[h?xkO ͯ_άEY↞==6)m{媕x5xX:|lj'8]ƶPazjp
v(eL8;u1X $ܙ@b
ݙs ]mxyVZm^9}L!H
<-]jp)&1¸C,MT 'oSO>eǐwoV 4w\ }8m2%sJݠ
5$ f ByJ? vhdLPo xŗXbĤP ƄuDe"bb{JD rKk灑A TO4W@ ?考I8(b\hء|Q5l ?>P 7+Cb{;۵K'^zz{pB| yh}]f
~{5x}&tyG;qwC]
p oh78CȓO=+VXYZ,YB(*{V8k֮ÖēNC^]W15˄|;xM(rfZ
=}=&H\(Z~x8;4 ;
n7]8Shyft| 0|P0b"G kB/&ִtձ\~ d\VdҥvM_v8`Y深KcR?7 :ޞ znO ?`rR 0^7MD`CdV$ۛ??nڴ?lpjZ8؎9 `\tECviD!61fOZZA(}d`_Ϗ@K۫U< q rc.ZmԉsfC>H\\i_MW ǟ .x> ߞ[
]őx8.PQѲ憎R "RBcq^%F;\{6Ln __˯(6:ּ%"omIJ<&Z]TS57hQo
v { I']g 5< rTX
+a ɲ ] /^3<w}mY r;f9*zanMn*4m+vZ8r2DNqrf {cQ\f 6ٸ BBk=-=y84KsŰx ߕJU4i)>f3
x]wSW?Ms_ݾwƌY&x%%)
l??sҧ g/Kݣ ˬO] GmZY܌9x}0cu\q?pϘ,?nvY&4,b4Xlijdm) 9rV9,'7Mlݶ,sΑ
$Z>$s
^[ ~O>ԠdQ<3j u "/s}/~,x'} ->1&#s LM܍a؍ O݃ҊkGvD8 P Ta,"I ?pX9 jx;bPp+w j-op 5jx{V`chb>U`ɝA 4i.F%"n·-[sr$X3͉=w݄G!ۦSyE Vi}{V _LT؏bW_-/xxodl^[ 4dWnvi}-"Zo Ew" Y߹oY]7( Ƽ0qV!D ?\QI{%Ȥm()`ncgw|y+~Аe,VYuB{ W|
.kH`W ,ס'F { K~Tl<(ڻ?snY0G@ V ϯ-.BϢ <ٮ jַ-W~I.j%FarX ц6 WcPLx ˖-g 6
¹k#z){|>I˾;kj*Ay5 i 1|9S{Ke6{mp%^Q~ƹk6 J^RWnrdspsx3䬛<:]ZRwn"t9X<8 O~b+υ3܆^s1A\<}{?<^y;GɞӢjRAϭߊkᄞ +ޗ]=?# &*u+]p 'O7 oWwG
_ƊW 0 H
nW~vMu2wkgux)r,w''dݽyf{:G>,gAl;:~8ȣ߂ٶ3d.Z?bYGD%#7 uG{jML {c=f;%K' E9X-`LLЇԉI==n$'L'q7"2G.jf!(IVaqƘKG! YQqB)*] ~xa# uE>bTzj{ /=0̝;gL (? y淸[- |(Ok~[|Svb ͘-c4m~q݆4M ^TB_ ֭'Sc(G@-E \ X 3*|ÊF%\<%h|{< o9*J'0JuuF(H>};=Xtx8*8 oVJ`{v
(D n gP8# * 91PT` ܶlbK0w\ۆWghz3%H]e8" +'_eǭ"sŶ1"K>\1&^LLVf'ٳf[v;NV8+3C)j GY>eߍ~&>`{fg{H+j*&=tM7=}5@?,0g8pGᜳ¶n̝?/μK+M {1sQ\)n\pmn%Eh?h<-`y2IJ gy6{9yFG+"V
[p;ou;
: q?3gʕd C 9bSByos~w8spG |)3H`V ǒ&`y\j 0G߲cVV)Mp/Ӷ D``(݅m{3~ 2
"1+@Ax!Jk:(ՆϢO\q} 0 B&mN$nߎ9EWt, > Y6mwUxO{ل_J&+Ndb/gJBA 4p\MN<23=D- \H$i&s]~&xv7wo~-@jrf+1mʴAׂb .N |U0bi7vmBĒvM ,(ן{ݛ~R\+qahͳ2;AQq1 kPW+WR"<7Y3f?O8~uOB NiB <>uT =h yđ&hկ~YRR (hHX%=ر}=gB\xхqV=a&y?7<c˖e5 o H`չ)^yg%J^Q) #a M看gOb1A?d= Kp0jO܋`+_x;⨊aeJ_Go :kBk6)CD@6d,;]1+ͬG.+ G+V^dM2 O{w,vZŢZ8ѺV:.,%=.8 6
y M뤹Fy= [|' (h;;_O|~C5 wɴ"1ی}q Znf\R GW//0gEZ+,Q oWE,߿,/{}.to}3RtѺ^f}
!}#\J /,t͜53=j &VKa {)+Z[Vc+W:I nגQ>lի 1+avZ,U ; O=hF@P͖M?oCuց@%'>uW\Ƀ3pwXvY(N;RZ^ uWꨗ<}qg u,cMV
eh 8{i2։br""8 n;uG?K
=%֔)}MNtS >'KU.ilrֲa_ ^"˨sus%hE ?pBV,6_Nt>e' Eɸ1.M''vH0Iۑ[f}]I߹E ,ek^3ގ>xFVK< [<(i
p*& t
:lqrU
bݙN, 3/ Qb9 e xѴ%_f4p?3h/2~ p\|"?. 4/i?Xh$'*33+g-'7:%2q+W}8]1<;ٟ 7Tw,Cy~g> Pk| 1% ZĂr(:,0~0J۳A1{ '3~`ʕYп?kC 6qEK_fZga oojTϑU{ tt3+ӟ42 Yk\orlݧD3ۓ㓍99{Y'#c=;ced2 e ]yՈ9mjj~_\ҴQ?ٳ"T.[Ulk1I@'믻E]} 7к潌XผƸwXlr-J8RFWM(Ҋ
tgblb\ q)
So7WSc] GpYgNfqJ !̚8?fNa"97d҂s=0-5-I >vY?Y8̟Z܂dw+{+q-7LݜE?$;z-Kʼn| >d9QnEUeiIZ2ϙXSno ӕAswQ͘>OhKy0ې?=ǝf<0xqZŪ10*ad-G#:8V!FhgJ Q7 jE+x>' oRR AݕiFZ %
yě_7ቨ8
Q, ?V|Ob-9s
l`':\k_ { .\m@hc\Gg tP&p,Il[( l2
#*[r<`JV6wLE3gu =+
k8ڃ~>Y3gkΝօ t3. >@xoqU|/M A6}ʿkHa-E .wyGxi/s3۽k\]wiV$]ึ xl7?N%
s[Y,FUwgƻ,?,^%,I.5-=Iz5kX-(!Byk&,,{ig,bI^-I /.|~~;.5úro jy! 6m/rf)VL: tPV?/\ 3 _`} Q-@uGrTWug D/ ]LqqiJar̟g ~+%.#xā.hx/X8qM NpdHaePq"` y瞇L\(
bhuNX'40 oKxusi{qKDK8#|6nޱgw|"͠ {X5sf, [ d!O!Պ =M@/qRZ.H32 >L$C>0)nb46~d/_&QdYh57O8ɯX¦ 3 aI],8N2[Ќo{du- kXcun<|n8 [s 7Ic݆+k1ث y\ZN7-5&tgxў9sp'XpTHL-O\cr[ 7+1i -'\0x͌vY'#qXSu]dX"Պ K ~MA $i/Ӿ`10}.?$[Ya$:Z&1 ^W~z"uL)+rnٚ=Www*k?I }# VDyCrffQ0&˹ۥ^-rEkZjߠi `yWߨG ,M$ nd֏h`_x
XESO }V_.Nᬮ۶lUVe}'8M1x,+;}Yw͚5&>osf)l!{ QVl|" l +W*K VX%L$$Z99+Fq^\>X 5xEdũ5.C+m3 h=K+nAuăK.;qIҋ/Y|_ ?O].6otU>f,s
Xo.^$ v_&ѼƉ9Vug/XrpMlwt DܢЪVV [l1 ;YZf3'|ǚys癫B7 kF++cvI{9lO2iBD5'ȢE5_,-f-\cNWtaZ)XiWi$rю zս9
Zlzs8iekqZ z*7bkde k4dձ
,{,
h?LLҕds:ZnK
?VntYU^"T+ս ?HlN9^ڴyS=c!1Bxh_wqK!plLCN$_lZMF`܄*\GCR~[P*#Bcq<#DfuI]nL=ȡq$XVvnv K/ ҂=&gw%b
s_+upo[ɧEAYgB$؞uV2g
@.}ٲe8c|'/i3Ć04C\ثQ=xV.W37P`d#k۶4QOq"esB>d" #3|/ę H&qZ^=D0kXũPQLmWۑg . {"[P vO'7j/ͯһ5t4$1hlPIV1[YWTɖ϶7`NquԑGYAsq;HJ1g݄@uc 3ݗ/_n,)y&ܱ| L_g"hHum&;Dw '3Wѝ[
# K|>;UHBʀp^wFYJ}Vh.^\hE5m8;#D /.Yegǐ wMDC#aK$Zwox+U}>=^]Y%qG͋_uxq=eGT6Wn~+)mEa !D{T ݁糚 ߺՊ.X]5 ^ghf @CxIaҺ )sHS(X"łv֬Z˖&&1uHR]W|hfHx?P`rrLLA.6ol ; Lk wuZk\)h:? E8&Y VtMFh?AQHnȫwx+z yXTdxuzAF1P-ik& ̞WõAZ|lۭla`6O>~&fnƕ?Ԯ
Θ'nuXnq gL y_p4
Ȭv^YϮ{7tu٤T9zqÆao߱bN=AAƣ >Ѽe(KH!e '}
&\"®27 #ѪiA :6a ټ' uxN`"ēԻ
zԆ8G{Oz=̘ͨ;.~7 Xe
^x#];u"zMfs)o1KPGK1ji52`d[O\
=]f'x"> U%< V*_Dtn?Аj, &շ{t%^} XRԲVGŋQV4W&Ј͍:f]:0睇l[(UрznAT* {]CY||Ss[釢鞲+
{
o ճ ,䄈%nfWRiPiR3pGOϐmL,c
͊ՅHP/=kefu2_
xq}7q^olV lߐZ9"w$m%D,B0e:,9U%3ƾcB \$BA8Z0N^,%D(˞~Q$ꪇ F:䳍lټ< ֯_/=| $itgYjY
۽Yз67gulC/± {Ѳ\Q&cq%tk`5>W;G)z
ʜZU
p}oul-2^sTG'GN̙ 5=_ 8h(j@e
[;!
i7>; D{C?'>ij ȂYs+ߓp.xkE3w<;漋M.3kXqϹ()xq1e8sƮ?aǘt<
Njm˖IlNZzRt
%kLXBT*-(qBzm=_ A_E(+Vޠރ"vDq\u
#Q' c]W qbओN>tdseP/hnB hHPlY~VWjϳ#FO>V 6 /Lį?*F6XM7iig ?v.{z{JN;47^`PcOGHor&};5& Gg}YV᜶D P2$nʹ
n=w{A?U3 W4`j+RXMxf-
<|
{IV{To)V(ZqjXϞFx wE]Q?=⃦&۸і-[ڟc=:㏛]:=qV=kO1^ƛiXzR7gAz ,H 'ݏ^/ X^CTh2(j_: :묳TH6e#O?(Q:2#GYe\m k JZ6~UJλXY-S2Ч(̮8RSq[qO(/fAG]J\I7sV +Knʳ8U#=KfjmEͺlw䎥wT*c6d_6[, VOUziv4lcU]. vLjFKXTC{QwvDW7>fBZGkm'SO;9<X{b@udт?Wl.DI|y>қVFFUN":۳o=\ T=S(lebXiܭ(dۛ?ޮ=Jա}JbIk%JKOKId[U]/nuذz!ŻLQ5[R}w18
_xzTO; d0R㻑Nok!=knE"b饗_c?>䝿ժՔc N7(ޙ$S ofJ3rY_Ps{cӉs8'ue^LEԚ.vѩ ܭ^f2 ?x]͞3g
D@x._FmG9k>["+V#^4; IDATiZb ;t 4m{ٲenjF`C+oʂM[[?t ݭ~3k[ۛ۹9 cIax51VzF}3eȍ$P:v+KIm տ%MykQ]q_W Wq ISҼ}DY溵wcWʶ+^ŭ)V# h\?}3窞
LS$w^:*[zݟr)nh] 'Mg=L1e![
VNv:]gsZs
v9#ÎA?EOj n.l%L^a<߯o}[ڥMf._՞6<6#ZF3p/9n]=mÏ<cONf1 ^yR+bʯx զ'6s&QMS]Ž
%Zgʾ\{-JTo%0S呢_ UKǿ 1+|:rkMQ ni|A%UnVG辆4҅X9-}T^mŦlohk!G qEV lmp&*/F =">
6Z5 Sۓ;VN e `M_'[ڵ-v"7;uT]qZhqy=-b}yd
7ےDFl%=2YwhDwo5j
k i$
ԥV)I3MKsWԚ2G} A$RZH{~:9MĈDpMI#'6EtRtF { ^>n]k6tJY0_-nl.x PiBP/v!K&vLmks6dC1n.!nwWCXVuD /U8Eź Ӯ.YMӆDA x~ӟ" w*[|05ȝHs5kV$TTSW6teEWZj}l (QmUZߣ(ip7'}s5ϩeioQ @bu[o?KOn ^5^fi {/y7ܢl"N S^:^OƉ4oo]c9*kT19sE><و ;o[[ƍ^ٿ%vh6uK]RU_\. mH_GDV^R]{i j7k;~Z}j}韾 o(VdŴTVq m
vtv$\tM_WtAc[W|6:ko&wV&썃}s=oϝtF9q.+Z
z$ ,OXd^+j וB>] ~Tq.?oRE9NqpY6w)Z
O땿je˗+^Hfupq$ʍmJ?:/rcx9Gq+uyk4*U+o`oLo)m:.tg5-6|;EOۥY I2\|źo͚4], s.:n:yvQ3k.|N:'XTllg?3 F' i~*5y ԏ {XbdY;[o\//G˚*;"bjяc1`u¬/?mG?Q} }ҡ6&˫h`k#[.V wG62*I69=zc tmy5e_~+ ,R},)ꔲF5;MğLMjѳWk۔nܠcQ۪oRt&T%Y WrۻYzUoKe)V^kwwڲstr'Nn~A͞ ѲC}) /`hSvHwrH;^v$.B}# ƍ]
Ǫ [g>.X>tH۽YɥXZj`狺U6,o˦oٰa {a+-!?vx
kll #[pߞ7vgݞ2 i{hsʭ!ixٳW?y%ǟ3Gl#EC(u ="b|RPSO6tRjuS-)+:%zNE~KWS5%҂NRz"5ƴ)JtBz+ڛ+գ[^6giWДV6h֓$v)ձ;VҌ5Rݞ٪[XrVۖ@'c0FCn9[^ De,zhyt7 ;JM-
eƊmdꞪ m$[\+WtkuY6Z|aOUWP
kK_k۶m.\1#[^~[ k4W^y^yU}3 1mJlB=.wu֯_oQW]u-^FfGs,~#y?nS|ox w?m߶fFfqMB}@:,QnMz gb
|{Fx[ ob[k]m$=_X+. )hhw@uhXҸD}l:ޡ VcJro^I1;9 m(ktSy} }n[ڹWڹ}M,Jq1g6QjVT*~l<jXЪ>7 )~WZL,vYG P+^c}Xc5@]v(ۨF5`UWaJ-]:XH\xm;
F?>j,آ#6jbvm^+{]ص8k`#'#]= 6kev~G̳ϺgiL?gu?6e_ OF =Tvmp ';vz]0gӌ֊ ?cNB={h ۵uVldF^yw-sn /=څoFJ-hxCuejFً8լ'+7Oj~V G!S n/(p)Z]-}r;$+N/˺QL %}iZ S|oT˺x>k=x:?roTSݫi+s}1ֱ>j55d - kHFIm!eݐ=/F4һ_VlX~:ۅzŊ~XM!ڱ
駞..Ʈ7goۦWl "viVsn#&(lt}u >h|~E?YKkq5 -9[5h B#5eu:\,ҳY4ctM6 }ھm g,ڴ**E6mrΪQs`Oc9 0ѣ܆U?yl4
5/TOI^V]?n(JT^K"`
+ 5(Uuo-2Rۮ, W\ c ={uZ Uo2رCn}PXW.6hn߿̦4\X!SvQ<网EݒXٌY?xtj==վEe/?VETLU蕵/Efr'USdɧ wϔ=J,oY%{= D-5^6 *NP:wO*)EZRt 'vNj
w, vd}V_8Bbv?aNOCI+oA
iPZ§ۨljNԿoV75[&jѧFf=nC=& 5VGK[Yz8jk|V)ݟ$%d] YU»k_hxz~vǯ~o͊?v[|-$O>M7w.7#;3 Fxc_\qDP7wlOMK1u8Zw,dJ]jՕ,:Mɬj.] ݧ##Ζ^E*KY ᶼq'
+NjkW_i41VG1ۭٔ㬩>}KWJg-SǴbz!(BUT 4GH_ ^wU[n _z%>XaMZd8aM֑ eVn/w YK,Q
5
mpT;#t ~ H2k9`A6h+kt9gzU>
6ѯv a,/-ZƿF.PEŨcT FyGs {cWVrx7Z6 w{ 1k}QIhaW-:B*E˕GMU=yU)iEOޯl}k( Q oY
GגVCq^d-e{ayi98݊sZKϓ,S5e
2UG~,?Ldgt9 6/ |2Ns
h\e!̦仗m.\VWɩ<| վRGrm} o9FW^Znn^6tmDT/XZ*-[1g7
7ܠ/Hqed88CR {m;v,8so 4sߎ5Y{.zk`Mb(7=Ekw\k4t|^CM=lCn7jvYgm|.]Rg.9-ZiK^Te[r]VOG|={HbY{_}}z PWgW;tsJˆ+̍_ٓ%n=UK)ҒN=WRo6
Oձ gfzW
1$u8[:ct%j-8KʓXXW]JQvM%#@4$gDHFZzzת-Hf.o?=b;!ů6Y}gb;SW_}[e1w"oJOY.v,R7oH~3tm'C=ym|e}=/}A=w}ꔭ=N~='续v g#Wv6Z5(8Z?믿)nDF{뭝nh2rS{~|ZyJ"4vR֓Ɏ{n ZMW k)m_cP{s0{l7B0Ĺnnj}+sC}g^v?9Y;z?F÷=gcb7nj M3lԫVԤG*## |ؾsAIQQ} 77Gm/|_ըۨQD*+ªn$ C68epAܷm}8R-jƢΣ<7MkG`FNPꪖb;b>wŽPy ' Tv" _ }-kiDkk۴яZT^;U*l1&ׁW<7Gկ |&Ru'쏀PV~o߫SzZm?>\XfJ F{! M#qGZTI۷n׳K}MZՙg閆 #`Gъ?zkΚ#Z~ P \HZ~vgJmuXlzЖ c 8Jْp̿Vz -`Y(Nc% EB~ogc
mkz_mܸqvA=CV8'; 8JYv=ٍdY٦;wtm1=7# qĶ/ mc[X{_gT WIDATnʶYj/^\|oٝ"w uS_̈́]q{e5,\P/\Z+-g| LyYKUGQ ʲAvV,`]|:31) 0Il6;rЏ^]m۶n_gZ3|N+twj 0I 6GXzջݻۭϰYf/ׂ ʳ}[L6bw^OޡZ~tMQ$Ş6eE/] "l6ze5XqDX$;tkYgήgyVg5khڔiAͲﵩ9si\VFn5 gyYSm]wݥ^xA=.YR-[
- :ab:G{9}_wVV-z-zjCQ l N9 ,` %YZn^uWnl$>oM@;i [Rwemڴi[um@fe:/y mqdFiVH.<=ڰaTe Z6e 'gV[[iumc On~D`q#X[ў=v:mڴMZukՐ0 X=%X6%^͝=Wԧt-,͊-xl`L 0\ߩJSgG.t^۲E?ɏ}vڵ,&?}Z62se#UV?nAӦO+!R1bf]yFoa 'Q>Vh A
NeQ7ovuXO=֮][6kXjjA UsŋukժU:s׳ 0X
ն۴qFWW]wFǼydviZ0t1H E&)ٲ鿴ɲګ.::;Jö'Fܭ(t3QR1D;0 Y/^ow{/RW}=*kUYG\}C` `*ǒ^YCON^Xy8)>ny b 0*a #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F 0 @`, X #` F *3B# IENDB` torchcodec-0.14.0/docs/source/_static/img/pytorch-logo-dark.png 0000664 0000000 0000000 00000066163 15207546147 0024445 0 ustar 00root root 0000000 0000000 PNG
IHDR O5 pHYs .# .#x?v
OiCCPPhotoshop ICC profile xڝSgTS=BKKoR RB&*! J!QEEȠQ,
!{kּ> H3Q5 B .@
$p d!s# ~<<+" x M0 B\t8K @zB @F&S `cb P- `' { [! eD h; VE X0 fK9 - 0IWfH 0Q) { `##x FW<+* x<$9E[-qWW. (I+6aa@.y24 x6_-"bbϫp@ t~,/;m%h^ uf@ Wp~<5 j>{-]cK'Xt o(hw?G% fIq ^D$.Tʳ? D*A, `6B$BB
d r`)B(Ͱ *`/@ 4Qhp.U=pa( Aa!ڈbX#!H$ ɈQ"K5H1RT UH =r9\F; 2G1Q= C7F dt1r=6Ыhڏ>C03l0.B8, c˱" VcϱwE 6wB a AHXLXNH $4 7 Q'"K&b21XH,#/{C7$C2'ITFnR#,4H#dk9, +ȅ3![
b@qS(RjJ4e2AURݨT5ZBRQ4u9̓IKhhitݕ NWGw
Ljg(gwLӋT071oUX**|
J&*/Tު UUT^S}FU3S ԖUPSSg;goT?~YYLOCQ_ cx,!k
u5&|v* =9C3J3WRf?qtN (~))4L1e\kXHQG6EYAJ'\'GgSSݧ
M=:.kDwn^Loy }/TmG X $ <5qo< /QC]@Caaᄑ.ȽJtq]zۯ6iܟ4)Y3sCQ? 0k߬~OCOg#/c/Wװwa>>r><72Y_7ȷOo_C#dz %gA[z|!?:eAAA!h쐭!ΑiP~aa~ 'W?pX15wCsDDDޛg1O9-J5*>.j<74?.fYXXIlK 9.*6nl {/]py.,:@LN8A*%w%
y g"/6шC\* NH*Mz쑼5y$3,幄'L
Lݛ:v m2=:1qB!Mggfvˬen/ kY-
BTZ(*geWf͉9+̳ې7ᒶKW- X潬j9(xoʿܔĹdff-[n
ڴ
VE/(ۻCɾUUMfeI?m]Nmq# =TR+G w-
6
U#pDy
:v{ v g /jBFS[b[O>zG 499?rCd&ˮ/~јѡm| x31^Vww O| (hSЧc3- ;'iTXtXML:com.adobe.xmp
Adobe Photoshop CC 2015 (Macintosh)
2017-01-04T23:25:03-05:00
2017-01-16T15:09:50-05:00
2017-01-16T15:09:50-05:00
image/png
3
sRGB IEC61966-2.1
xmp.iid:eacdfc19-397e-4f5f-aabe-0bf5cfa68232
adobe:docid:photoshop:21886c52-1cb6-117a-bff4-aa7c7148961a
xmp.did:fef6a0e1-0f89-468c-a40e-932975d64830
created
xmp.iid:fef6a0e1-0f89-468c-a40e-932975d64830
2017-01-04T23:25:03-05:00
Adobe Photoshop CC 2015 (Macintosh)
converted
from application/vnd.adobe.photoshop to image/png
saved
xmp.iid:eacdfc19-397e-4f5f-aabe-0bf5cfa68232
2017-01-16T15:09:50-05:00
Adobe Photoshop CC 2015 (Macintosh)
/
1
3000000/10000
3000000/10000
2
1
1024
214
, cHRM z% u0 ` : o_F &kIDATxuHG~0 AӬ to4 H@ThKnZtS4;#L|(d[ H B[UGV & H (y7h _g3ICIi mR @I:9 @ =
46?t њ ]G#< h6 LRȟu
@1 Y_3-t Y @{\wI4 D PRgǿ~V DT H @8GX Fa
&4 h 4:W3Z {R ~ = {>- |^z]SH3 @ < %}t#k>}5HrC ?4᷻n*N X'\U H Du>%4 @#銖 $ 0L([ $ 8t.*8?spu |xM+X 笹 /
?yN1ZKc61Zr
g7||^=% @Ro>,+y 9W ` C7~E K @ m۟舖 `URw 6YeU>
H `]E鄭 H `_R b7T5 ի: Oir phG<:e z 4Y2 8vK 7 ! iZv\ PL ,V9W ) < ! > K H P|~!>W wz]. @y>p
N $ (G?s醄 ,> "+v ֏A/ } N%-O :6b5V ^ϿytD Z'bްmf~k-4Yr n =j@>]K: Am @= ·ϯo%?lj 6"]G*nD3-4cB H P:cHTT\ŢJ u&ٯ !p
ĩ&*:΅ ԯѹvA/- aQWOS $ هoU W{DS @\jV ö?&s. H EI?PT & \=d DmDTSn1 /mZxvo 2 }G'(x,VK6 DԫI69 /I(XvK } A/trB$1w`?
IoR)d!!ERϹ*hs#_m& ].ncK6:{?nY{anUsd
(Ry !vAo,VAm [m¯\ҟy;̻{4:cO܃/7X!c+ o >E*8g-Xpo_4B7N+aUT*mNrQ7KU}l$-%}d] m?9b_p!lx2`- &FF%_oPA" dzQ UQ
4pcWfA} ι,$\Y4 ?xI望9M!#j vC`hGmRYI7aRQkȶ1;ט%q
C⍊cMNp
^ Ц%[jg!@nqmb ~;ZV }wAF6I > p%˄D@4Dҷ0n4ޛ *t㜻n`:R1bP;k
.E $J^~$~5@>uAo[ׅ* 枾M %M| (gp٨U
iX\qUo8,P675 TՉV`J]ݽK3@ tBې$DNf
x KA[5N>:!)J
* ^ < VÆ| U5Qxqn MqбI'lvw,z Nh s 9Wak
*z{1 -&0=WW)Mq\kP˽.M xhB0Lش,p>Cv"{?k?l*{$ι0Jhuýק)@ &BQN֪u}v$Y> `*1pH3 tUo|DŔs$ lb~UmFl2a'1 Hwh]5a/H |i:KLHTé L,58t潟CQ %2(# ώGf*^Hp$@ v|XLΪ_yWaʇՃץp`,X$ $@ v.ꜭSFZ
V{Rx 3 HhI7"J $ ?F*ZG5O:$x
;6 Fea|` D"QbO8Az!Ȥ ~q}P %'7L i ԚOXn*@ɓƃC{kH냶jT (> e # `
[ZonF *Bph Li\Ig Pչ\(L4:Ecr%FZ 넶;kc@^" <~Q{K;/4 @y9 `- V0TXkzꢫW!Q8y?HWU, lU` ѡVI
M9m u
63I_Bп~š+DRqo $ |28m ~BRm;9wi_92ѽ5f*%#MP+CIv?]i풄k?t i*,iZHlϼM UT6ο n 塣L` QGqΥu.Os T}&obtOueSj| ՠ?P@6JCYy#P$<F6gj3rk-ce ޘbc)T)]~
}_zcIϜ& \0\qբ.7lk JR 1Z{CНGpj:oR?S$zzUV*1icߧ1\ m~h(6Ƨo@{eW]С[ 4庡」~,BFV^Ժ*{v?`W 4PT@*Ͱ>|k[0M KV?$W&ئeJU N"ޏd8ICslt*3ܰmhQ_ki>%4CcVsUVIRi
HNg-m:O aG1ݢ1iC03jDKV X\?D;&X'҆5i,[M[MMé}xx$ $Ʋ)6hХlU藹O>e\YvevB" k1ȹXu
k7@=X[6anYe[Q )-Qk\v$2T6b\5 *>\ a ]GƀGy*Wu5.U˕NC+em_$FD
MbSIgwj~hкp-C'(.յ?c>*'*M [^sOwUDZH`TI ܩ/\ A lU:
lÑk_ {. JǃD بX_SXj ,VA6`." ;Ǣ,*0fJ< FA}o\6sFh&(S v{@u%][
2a* u7
x1LI짢u7w*'Г,F h/FvoEh
- 2(5oSn8poYUR1E0w)@7i,my?{?/z
nk W%ιk࿎onH 5t}͜ R ec^
s aH?M wZu bS'@4snexx
?os[9PF\_`/E,Vm~cw㜣 & k ċÏ'
S}ƕj L:8{Kt H !T]r
zVboQO<{w& H &B|[!C2* @9 0QYz\9 Ý?[t c H (!b][*;142 5 @ ` K-*]cf}.c Bs_(E2-Ј@b-F. @ %>@S 1 H D OXw]#>XMsU~ ()ۊk*@潟rYKis<IUUkn! 3"?PES= P@ 5"$ @f82T(dA}*Z V<#O] $ < W9:R ?
oQ |R>˭42KIsn)_/YhM"A/1:4}+:H ,`%_$ 6V
AᔦHjU
V7s+B $ T.Yr Y;)@pZ[:Ҫ L(\1YǑ7F * w:ǤoEq˂ꗔ
h. TI /cviY$C4{km$UP
*s}n=pE T1;P. R~㜫'b }H%}dw 0m8, 58 㩌Ijo} H @E9R2 К o[K/:ι4k &SH c sCi}3:osnB3
%%- |lݞڭ IWs)xs\fKp X$] z`UECv%]W=c>$ ΜsS `
@$U`k^* lޑmF:$* uaɦgN%Kc k,yK ,}6:n'̇oyg&iN12s.q
@ vo]^[ˊѰhmF :s PXmd7?Mj%}$agԀ~#M"^\$?gaC$?'b .#:|лxLq,wkm&e,7&oE$\q$?%9 ι9wYiH $ x#@LAd_ҷ^,0wZSpg ѹtU,xD,ߘ$ 8@ )9%*vak%Ȯ:,zfδ_+
(I:W$9:,__r F[QUL7 1MX0b a"i [b;D@w'F&>@ &0$J- aėkn\6U &
WCk\?
oEx| h& ډRIi>]HV||<`*DśN
F-c܍qϝsB|y7ιO/ӽ/"P **|TU >p ow]I
1`dFs.SڕS_:*Ǯ ~)$Lp#}i
>k.JA R |)
<7X6I97 SHx?֏MnP I$db5x
o,[$@ #VPj+D,;élgĽ>{O?H Pm\oC๑ι.W$@;cm
j!<% `;N+ڿ A]` > }^H $ 0Llhʍ [t1-{y 1 "H UѮs*d >8# J0 #) @W<K$ ж)Υvpu
?'oyl DH+M٭ I3DH 6_ @Jf-p/ddcA\Ś(Zc>-?H P$TyX
+, X$:$<o74H P$@bx\b sc`vĂ]{?ۤ{?
+$ i OSS=/K D5s*Ji# AigƃrId:M
́ \GS`A S٥ @C P
0 -Uxap9]9 Z*=I:/@ '*lySd*{[tA@ʼn,$F-LlT,B @˒ yXMB"`A(g
3Gm㜛p^wU,yú ZېXEp`cc ^{>S5k׀L x0`y*JY +'OTC\Í~?^EBFE"8/h74 %$M$MAo(郤OGEF K)>X2`"q8
- `'. $
~2}c1h*'}assI|Hd@7)SXblsиq e @ _$elWqu[lT¨G<:R pQ]2$ xIJιDER+_῝'!5?Pn$ 8!x˽oi /2Vqr~"6Qk$@ KJ%p; %y/_'a t 4 @ -'
torchcodec-0.14.0/docs/source/_static/img/pytorch-logo-flame.png 0000664 0000000 0000000 00000001762 15207546147 0024602 0 ustar 00root root 0000000 0000000 PNG
IHDR szz sBIT|d pHYs b tEXtSoftware www.inkscape.org< oIDATX]h Ug66i
R661R*4PM,L6PyD@!O"S۷JfPJ[PĆDnՒ&LJθ=OsqvfU>xSGj4 U`SD͇ Zk٬ӵtkl
@AZ#mX1Y, Z -C73ÿgpH-C \!&} Q嵒Uxkۭ{ aC1jF
oΎ&T~6$`W͓ %2hu#ewQ8$ [5J/ED^+=3O< D-KǴ# p=lnni`OAx ( sy@~ 1
ʫL'SO N=aSmrL;- @0CuD@shug/(@U*πPiobQC 5Ѧ[z F7 0A2lnbi]I84;8'IR~'L;-Ao:]m(5Zh D>(MQkA ƒ7
>
%oӉ~5tbH%W2iky@"e xcD1?̤Z+^Os M'G}g/䳟g/
V<;>4 m9_>N> :
<\=S^g 2 >\S,u_X
3/<yz] IENDB` torchcodec-0.14.0/docs/source/_static/img/pytorch-logo-flame.svg 0000664 0000000 0000000 00000002725 15207546147 0024615 0 ustar 00root root 0000000 0000000
torchcodec-0.14.0/docs/source/_static/thumbnails/ 0000775 0000000 0000000 00000000000 15207546147 0021750 5 ustar 00root root 0000000 0000000 torchcodec-0.14.0/docs/source/_static/thumbnails/grumps_6.jpg 0000664 0000000 0000000 00000064767 15207546147 0024240 0 ustar 00root root 0000000 0000000 JFIF C ! "$" $ C ,"
} !1AQa"q2#BR$3br
%&'()*456789:CDEFGHIJSTUVWXYZcdefghijstuvwxyz
w !1AQaq"2B #3Rbr
$4%&'()*56789:CDEFGHIJSTUVWXYZcdefghijstuvwxyz ? 5Yq;@U$KAimֺ V @֓%FЌc}leus+~ g"+">9 μ6&HTbU%6R9@b0Eo_6n5a ۚ 1 րoތsX4 J#Ł > Si ǀI
D5$ PO@x44 AB iz#&C ".HX
zsڟ:qN3 E)n(l2nJI;HήO2]E')$>}PP
He pp:sPYv*Bá5B<Fն6xGSցd(uknLc t%|F{9 LQ@j xPUAz ec<"GcDJ s@\fd6V.N*)V@QbVMne& ㊱8"1(eX
MOsxڹ={А !|曩HDSI?OWvXxM.
GCޝR;795$`5+'ʣvϔ8Eq)*ҫ\
@>vU̾+s}4XW .(OjӽI(iݙ,Il'eLdֶ W/H@1A
k3 NpI G5FR̽qst 6&j8\*ݿZXj0lr*g "!#X9c+"܊6
6DG:azl):ʞ@=piHT $L8
AsJx*G'Җ #z(rpGɤM-3b%n9銎=`4wn_*pcxVV'y=R@>a !U|L \⠑ zJB6|v`2+נHll`X(pj1$xT[$ ɫU\qM8rܠ-@Ns8Ux#Y^g8>q}hl" )K16@7$D 0{҈ TsuZBB+9RF)6;an:OMA2MhI>iKn2˷>õC/I''8*xmT˟ƪL#wd$Հ1
Đ:ͅjվ\gr=(esgPCS[H VѮɦh ' |51ґ\ O'%$K,7y$N@Dݞ%E# 8MԒmHQB
jf !@[Ke$*)#2 &ٸ#i?Zǀ6> } %#G jm`gP9l^ "KV j4 f=-ۃJO?J"*9l(ȠC*Ieڧ9늊8$E r:оg4C<$
ZN @22Z9te {,H sPn ݈Ө(T:RlrpsҞ-#u& 2H Q۰xD1:8 q 3եuss&KE74 (2IoB, !8 S˅$]wo>XAҠ$rs<O;mMJ"Lɨ$~ ֧쑗? LD1Hw0 pr}OzK4(;:ifV!bXjq`㎕fOjmHI: 16eD2*1n8 8 -4_22'O@sZF73>B۷9'K bnjUVre'#RHqQ$4uћ mMsԡ#`܊/"oj, V퀭V=
z東 ֊p6Dž*dJDF yYl+nn*1x#R$, H< l9-@
cֲn *ďi-Ad*Ι] sɣk;I9'[>@;@Jã\,2{Ւ"z 4H8> X ~&
mG! YRpTmTQ-0Yi
&sfe!ګq֢vi$bFp\$X`~˥? (S 0G0}Xե vMW'1V2{@
- ۭZ<ژB9sH ?63tҒAKpN昆 s!*mQ6ՈDɑB
1s& h27 1 i7GS5-£Ĥ. 19@H\qsZW ؔ# V:TɎO$V>Q2"ۆZpsA^ŏ".:95vLm5Nq@
fُʆ)\#L[_e F86R frEy8lWcdҮEq0};VȍL_`ziH eAR; wێc#8ycׯʋtc!ִ͇ulsޙ3hr>`}5F AV;*Rq N@ zqN|OQ# )/6r+w e [l5&K{㚖Ԉl`d* tu&PΪaLQ؎(QԆ`6O(̬I})$ym@Rr{
];?`kif<ao.%p@ V$Nc*' vI *}8 $fS=}):T^FOSe܍bO?t{ _rHsM|N 4K%ea^DjJB7:
pǩ'5^H0M&4D~f f;}Q Gn)^~SOZK'˵l]:dMYQl ?:F95eM0G̬~c*$WrqHn$}I䂹; z]ȑI Y7"5 F 6jktU(j=*OnD+qx AC :rV7/]^Y
wP¿*$vR.-i t6ۑu7j0Y;uo 4 -6fNcos d0p[ 3)W.)b
tP[ J$"(0Ei?Ǎˍ+Э1FD2HqQo A47pZlhde+ T 4ʹu ':Hc4qgք yWCWWH|*ռ 95Ek*P04l̺ن%9TQ|hYs֮#QL}N&<qOq>{Yn=6M.֚Otp6Lg5# ,HsI*2goM ȅOm$;T{ł/,GrM}1+99ߩⳖ M-*8cRI|! LVw6V{s9%Hpsk1SUNLV X94ß(M"_C˵XA>~"ߙ{EY e kcEFi&팬qcH˸ &84*l -=0ɢwǯT>fidR&6'T L.P1Ml2ga=*gH0yJySiJܳ
T~e$)NܜvDH]ZX5? I)x {j Cs ; {q֩2 L` eG>ڦ8c* .p8S҆Ƒ 鼅2ncL?A$S hqdcg-*cv1BJ9#9
W2* HGKH1:{~ЛpO*+b3 72nspP2=
FB ;qO
%0R{Wxim
A#`R PI"`=O}: Wm0о:浾-xFXl>u;NyHl R= 591d g 5){p K- )WEǯҼ:=8BȺ̘,ݪ"X~]^@"~ 41(z/Inh
1ks qvdI\RA8re^jh` +b ( g4F L*APe ֣ Ly4[(WTeX
#" K F4EP
˶j#,hdv 5O1 88Q)m ~jlwf9H3梋h6K929 .y9B{7cBoJٷ r,tKm*HvA օNm
7OT">SP Uc#ڹOƶi=FFBgwk h琒
ܟNaźGd19YxK /#.
kI0Bg8:+_|WDpnc*zMͦE4R#'?Qjw3aSVet|qzC?KJC4ȷk~gC d^$lkWR][0Ϟ9r6<`YT3~ <'qV%֭\e8I]f)eٞB;Va <ִ0cH"of]th*Vg/0qTy0 `^qSt ߹A -@5s,y䌱nz,q K \U݉ g eI FH-`^@Pk' ~ o S#ָ
#>#| 9֧NYČL{HU{aϕ;7V3v<[9(' J.'VUIHz$5/&U ;ðL4syi;>NB@Y@HhG!x7 (/nc>qTu鶭{5[2}1NNoG>$* (OQ8 _ ?z2jݼZ$[ n:IONWj8:Jr$v' >S
98R e HRNTV "<*|(R+ؕ9I<>evh=#%H ɨ<'dڗΖeo* 9nv-lLpU>"2tG?`Wf6a,Eldx6֦-Cb]3g?6tߎ,Z}+#X9$sb&yW-zZ };{X-` @W > m.&!
QoZear\n?19L~R~lRHaYـ'' U)B('҈ԙ݁H>
ʻ] *:jY#Pr:b[?2y$88h Ҵ
y@]{j US4y#I i9ui1̀+;ྗ{ q ڋ?Sw*c^ym6y }j>sPQ3y5\X\3LsjOZ :gwNV9>{?r=Օl:B9v 7\~z%]59X:5[YY#
TD=x~D_ άf-D! jkR0I&n;
q>) v!m *H_:w+2~EW@FMC]3m`pI7S)$&X0rO qd"@e`zKER|)?)IZLȄjռn.a|G"
z"\d 0M^6 .$~#<~I@Џ]1P=>V4;$X)6POM &
zjEfY1@-9ulo.T1o/?T 0A; F6_"9e;gR^%i" 2crYq_A`6: vp9= w֬sw ^] =/_{[Aֱ>@d`#sA# jj1$Ez Vc5? &k
g 8FfA#pYXSk>^y_Y}@WӜ +:)3rx qOQ Nl,`oʈobEdˢ-],>L"|<kd\l;:e[LqDDF 'Mhϱ =Hg zԏ%Y[LU`L鏥$063q
*`00)o-"dɭim-UCP[ZHY]rIlV2uVbszJV N ULhy$f_e5q,dX< H=0)ЕgR6Ӯ4
X$D[#xw.~g# 7 f;ıpA+#e;= T7-9b 5ݪcaMu-i31$t䓺6oo&lV K1TG?;H֠p,8uƞ)qR)g$Cuesʞ?i8J/Q' gC7]jGpYIbx>':EM\U=k< `ω4巒ᣎUs^q[4" Nk%-Ǟ.A qQ|=8W\74xICAl CRAiԽt#5^ԷZn%vvߪRsIF ^oaԵ*!dNw|1yxo2']HC LvZO,ːݍRH!AIaup@AX$M fnD3F3#4o]c].W BhAo d$q�hH\d) z*ТģѤ4 I2}"}SOdfP/$=k(@=A
Zk:DMNq2 >z uH%@I=;$̝sf[}Ͽ\a1%N,oZ08 @z QX 4麺:DZ 3/Fy_<}+4_
XAu" >٨p} G{58$4ycX=y|^r;zdK}8 [~# Gy hJl10qi+ > ) s+ gx , ו/e8"0+=Q}{-"H
y5<ܮqtg8:<˨[&?:^S<<7^ ,Ě}Txm[\ JTnBG[f A4D7 + E8U?
ӄwges
tpNnyE, b|,dwﯴpNۅ7.~FTp3n ^R2Hc i^E$
$~|ƺ 3ĚEMuhn.ع# /s4pIjU-(8/S=7S6͍)&?CW`!h< q5' (i^/0CpA?>`I=56G_Mkqu8N T r+w^.n,L2(W].3{@OoJmIo>\e )jΘŽ]]&OغvaYaa+ER;9\G!|?κ/qU{{[s>iYJmi'|`>^j ]?令m} d%! (5%ѠO2].F~v-C[vZ[OeDIJ\xbCӏ*ݣR'߭sh뫈e?+ f91Z(.s+ 7Fb~{Y99T}'Dvk.ʠԮf>FGUfN?#^Cu;%tьRnX?qsp#
wouuM'{ =9Jhάҋ: ٥д[ qob IGOν_ź@k Vam 98Z_M, {Gw )"? @5N*[ m5CɞXFU2cu&Dxm!Y*7xϸ5?Jј~nǀeK8ۆOL rk vqyiLmlP3^p݄:^RyA3d9' u5Ltj+0r21}ku>t,|;LcֹjѕU;E-c)>nGy3D>Λ ҩ[]:q,\Hw9IloZR8=G>S tVVcT{lFo
7#^ñO6<` >a}]m0I'וZȢ' <6֔qJS*Eຽg3 S+}3Qڍׇu/f \?#vVcls ៈ:eI lۿ R7^GK٦\8b}n
y7IgZ* Uċ'[R+ <Of.2aBKs3n; μ%XXx=;W mkKQD
0үe9!?C_ D.qGpsq\Z3hꮋiw(nf8%Lڻ=/Im{#iavqQj3{t?9f^UKCц-gg)6aOA}mh1ZC5}\?ʦ!~"Cs#S5hVO?0q$}=Ku 7Jmbw v!䚣{
M6 %y]uy3I4#$T2rl kV&E &/&Ȓy_>MdGԧ42ɺn $R5B S^j߆Y_R;5hLɾz+ +Z#Z yWns2:Y.L*7dzAፒlhzG˄Wq ݸТ~Xn8hp>ƻ_Z28$W?Dk)p $:Gj6@MQ^}XX脏N Urpzgަƹ.{5WD Va)o:aשC+(rꫲDXd"R$E>XS o^2BH ]1vCITq@ZX-n8Bzֆ o%Pި}MCg/ԋYFɷo..
k1s]gVn+1ڋ JT9Sd,Mxoʀ=3+S7BzqUi Gy>) :nF>aӐ yV0BA|>Ow$# y$K?8 \uW44z}*Λo7xk?jg￦q4Rm-74VPʙ8ҬZ<%s~#&=y=zQ$-mJamZ_=mI>ZiyY1& J۹z&$W p1 [*ѩ8#6WT~9郜udwaO97H&'5?A ypDuۻ
gDRzg +5_ j0ﷸm%vBgLXӁhv+XaFG+⫟qa s4,xOƵKӬl4= LKkxEy'5۟& ڼp@ AMvr CdczN9+2݀:g\jBs(#z|M4 p v8.s;v_.mpQ)Xj7{u5-q/!U22:dt5ֵ*䝛;rn.<̀*4,ꚬ?̎ O %8quQ[Ls!Kdn==jV@B qzjYWYF'(z 0k(9f* N@\9R.'>4
=c `7U kq Y4L$#>^Չ R0U*)*2}CXɡh
6p*[" hA
ʼn tӸ/F0N>|3RnͼR A#㫹Mx+
lZ( =v 8t"/<&'S!R@F
v?u InL z ONdifz
062uqA5$a$|)Cc#,K