[release] Bump torch version to 2.7.0+cu128 and torchvision#61328
[release] Bump torch version to 2.7.0+cu128 and torchvision#61328aslonnie merged 9 commits intoray-project:masterfrom
Conversation
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request bumps the versions of torch to 2.7.0 and torchvision to 0.22.0, and updates the CUDA version to 12.8. While the version bumps are applied consistently across the requirements files, there are several critical issues with the URLs for package indices.
The public URLs for PyTorch wheels (https://download.pytorch.org/whl/cu128) and PyG wheels (https://data.pyg.org/whl/torch-2.7.0+...) are currently invalid and return 404 errors. This will cause the build to fail. If these packages are hosted on an internal mirror, the URLs in the requirements files should be updated to point to it. Otherwise, please ensure the public URLs are correct and the packages are available before this PR is merged.
Additionally, the requirements_compiled.txt and requirements_compiled_py3.13.txt files have been partially updated. The torch and torchvision versions are bumped, but the --find-links URL at the top of these files still points to the old torch-2.3.0 index. These should be updated to torch-2.7.0 to be consistent with the other changes.
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
| @@ -1,5 +1,5 @@ | |||
| name: "oss-ci-base_cu128-py$PYTHON" | |||
| froms: ["nvidia/cuda:12.8.1-cudnn-devel-ubuntu20.04"] | |||
| froms: ["nvidia/cuda:12.8.1-cudnn-devel-ubuntu22.04"] | |||
There was a problem hiding this comment.
wait.. this might be a bit too much.. we do want to upgrade ubuntu, but if this is upgraded, the other CI images also need to be upgraded for consistency. which I am not sure if we are ready yet. @andrew-anyscale how is the status of using everything with manylinux built bits today?
There was a problem hiding this comment.
I see, I only checked the forge one, which is 22.04, but seems other base image using ubuntun:focal which is 20.04 iiuc.
is there other recommendation that can be used to make the glibc happy with the upgrade? I think this premerge does block a few PR from landing.
Thank you!
There was a problem hiding this comment.
we are working on upgrading to 22.04.. I will let @andrew-anyscale have a look on where we are now and make a call.
There was a problem hiding this comment.
Apologies I missed this convo-- Let me refresh my current state on the CI upgrade, and get back by EOD. I believe upgrading the other CI images should not be an extreme lift
There was a problem hiding this comment.
Appreciate folks patience on this. Iterating on the CI upgrade on #61533 , and I'll update this thread with progress as it is available
There was a problem hiding this comment.
CI upgrade has landed #61533
You should be good to rebase off latest master to pull these changes 🙏
There was a problem hiding this comment.
Thanks and rebased!
Signed-off-by: Lehui Liu <lehui@anyscale.com>
Signed-off-by: Lehui Liu <lehui@anyscale.com>
Signed-off-by: Lehui Liu <lehui@anyscale.com>
Signed-off-by: Lehui Liu <lehui@anyscale.com>
Signed-off-by: Lehui Liu <lehui@anyscale.com>
Signed-off-by: Lehui Liu <lehui@anyscale.com>
Signed-off-by: Lehui Liu <lehui@anyscale.com>
Signed-off-by: Lehui Liu <lehui@anyscale.com>
Signed-off-by: Lehui Liu <lehui@anyscale.com>
…ect#61328) 1. upgrading torch/torchvision to `torch==2.7.0+cu128` and `torchvision==0.22.0` 2. skip one torchtext test as torchtext is discontinued 3. change the weight_only=False default for torch.load since ``` [2026-02-26T00:46:49Z] (1) In PyTorch 2.6, we changed the default value of the `weights_only` argument in `torch.load` from `False` to `True`. Re-running `torch.load` with `weights_only` set to `False` will likely succeed, but it can result in arbitrary code execution. Do it only if you got the file from a trusted source. [2026-02-26T00:46:49Z] (2) Alternatively, to load with `weights_only=True` please check the recommended steps in the following error message. ``` 4. update the base gpu ubuntu version to 22.04 to make it compatible with glibc Signed-off-by: Lehui Liu <lehui@anyscale.com> Signed-off-by: Pedro Jeronimo <pedro.jeronimo@tecnico.ulisboa.pt>
…ect#61328) 1. upgrading torch/torchvision to `torch==2.7.0+cu128` and `torchvision==0.22.0` 2. skip one torchtext test as torchtext is discontinued 3. change the weight_only=False default for torch.load since ``` [2026-02-26T00:46:49Z] (1) In PyTorch 2.6, we changed the default value of the `weights_only` argument in `torch.load` from `False` to `True`. Re-running `torch.load` with `weights_only` set to `False` will likely succeed, but it can result in arbitrary code execution. Do it only if you got the file from a trusted source. [2026-02-26T00:46:49Z] (2) Alternatively, to load with `weights_only=True` please check the recommended steps in the following error message. ``` 4. update the base gpu ubuntu version to 22.04 to make it compatible with glibc Signed-off-by: Lehui Liu <lehui@anyscale.com>
6 participants
Description
torch==2.7.0+cu128andtorchvision==0.22.0Additional information
test with premerge