- Notifications
You must be signed in to change notification settings - Fork371
feat: Python Runtime Windows Builds on TRT 10#2764
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to ourterms of service andprivacy statement. We’ll occasionally send you account related emails.
Already on GitHub?Sign in to your account
Uh oh!
There was an error while loading.Please reload this page.
Conversation
HolyWu commentedApr 21, 2024
https://github.com/HolyWu/TensorRT/tree/windows_CI_trt_10_fix |
gs-olive commentedApr 22, 2024 • edited
Loading Uh oh!
There was an error while loading.Please reload this page.
edited
Uh oh!
There was an error while loading.Please reload this page.
Thanks very much@HolyWu - I've cherry-picked in the non-C++ changes (the C++ runtime will be ported in a different PR), plus some test modifications, and will see how the CI tests go. |
Add build-test-windows job9.3.0 Upgrade and version fixingAdd bazelisk installationUpdate installation of Torch-TRTBazel install fixPython only for initial Windows buildConda install latest cuda toolkitCUDA Driver UpdateCUDA Driver Update 2Get windows version infoAttempt driver upgrade within systemSwitch installer script to curlSwitch installer script to networkTry testing E2E models on WindowsTry linux testing action for WindowsUpdate Windows build/testUse local version of ymlUse local version of yml 2Use local version of yml 3Use local version of yml 4Remove CONDA_RUN referencesUpdate Python sourceAdd torch-trt install script packingRemove extraneous python referencesReplace pipefail commandReplace pipefail command 2Replace source commandUpdate cat command for PowershellAvoid script overwrite in runnerAvoid script overwrite in runner 2Update path concat to use echofix: Use forked infra repositorySkip things not supported by GitHub-hosted Windows runnerUpgrade setuptools and wheelRework windows testRestore root WORKSPACE to originalAttribute naming fix in YMLEnable one test in WindowsEnable GPU AccessEnable GPU Access Attempt 2Enable GPU Access Attempt 3 - revertSwitch off miniconda setupSpecify Python 3.9 as default for testingSpecify Python 3.9 as default for testing 2Revert py39 changes, make changes to infra fork, re-enable cu121 installationFailing cuda installationFailing cuda installation 2Failing cuda installation 3Differentiate pre-build script during build and testInstall trt bindings + lib manuallySwitch back to pytorch as repo sourceSpecify validation runnerRevert "Specify validation runner"This reverts commit6d664c0.Enable more testsEnable more tests 2Remove Python runtime specificationClean up and enable additional Windows testsUse .bat for driver upgradeUse .bat for driver upgrade 2UTF8 encoding for windows testing
Remove serde and extraneous tests
| source "${BUILD_ENV_FILE}" | ||
| # Install test index version of Torch and Torchvision | ||
| ${CONDA_RUN} pip install torch torchvision --index-url https://download.pytorch.org/whl/test/${CU_VERSION} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others.Learn more.
Actually you can restore this line back to${CONDA_RUN} ${PIP_INSTALL_TORCH} torchvision like before. The reason you had to manually specify test index is that, when you were usingtrt_10 as base branch, theset-channel actionset CHANNEL to nightly and hencePIP_INSTALL_TORCH was set to install nightly build. Now you have changed the base branch torelease/2.3, theset-channel actionset CHANNEL to test and hencePIP_INSTALL_TORCH was set to install test build.
| python -m pip install -U numpy packaging pyyaml setuptools wheel | ||
| # Install Torch from Test Index | ||
| python -m pip install torch==2.3.0 --index-url https://download.pytorch.org/whl/test/${CU_VERSION} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others.Learn more.
This line can be complete removed for the same reason as above, since theInstall torch dependency step already installed the desired version.
narendasan left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others.Learn more.
LGTM
Co-authored-by: HolyWu <holywu@gmail.com>
Co-authored-by: HolyWu <holywu@gmail.com>
Co-authored-by: HolyWu <holywu@gmail.com>
Co-authored-by: HolyWu <holywu@gmail.com>
Co-authored-by: HolyWu <holywu@gmail.com>
Co-authored-by: HolyWu <holywu@gmail.com>
Co-authored-by: HolyWu <holywu@gmail.com>
Uh oh!
There was an error while loading.Please reload this page.
Description
Type of change
Checklist: