-
Notifications
You must be signed in to change notification settings - Fork 227
[BE] Add sccache to manywheel binary build #1169
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 our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
|
The ROCm failure https://github.com/pytorch/builder/actions/runs/3301097112/jobs/5446221931 is due to the recent change in #1160. cc @jataylo Could you help do a fix for that? It looks like 5.1.1 is missing from the list ( |
@huydhn Will take a look at this, thank you. |
Push to https://github.com/pytorch/builder/tree/add-sccache-support so that the docker image can be published for testing |
The new docker images are published for testing cuda11.6 |
Not using sccache for binary builds is a design decision:
|
I would also add, that for binary builds we want to have predictable dependency versions from what I saw in failure logs it failed to return some dependency - dependency was missing or incorrect version. This maybe a same use case as @malfet point 1. |
Manywheel binary build currently doesn't have any compiler cache, so it takes more than 2h+ to just build PyTorch https://github.com/pytorch/pytorch/actions/runs/3285968556/jobs/5413580528.
I'll make a similar change to libtorch binary build later in a separate PR (lower priority cause building libtorch is not that slow)
Testing
GPU_ARCH_TYPE=cuda GPU_ARCH_VERSION=11.6 manywheel/build_docker.sh
pytorch/manylinux-builder:cuda11.6-e7608179efd287af102e40941fc24abff8d8a5bd
. Here is the exact command I run inside the container: