Skip to content

Update nightly version to 2.15.0 - #8509

Merged
atalman merged 1 commit into
pytorch:mainfrom
atalman:update_nightly_version_215
Aug 12, 2026
Merged

Update nightly version to 2.15.0#8509
atalman merged 1 commit into
pytorch:mainfrom
atalman:update_nightly_version_215

Conversation

@atalman

@atalman atalman commented Aug 12, 2026

Copy link
Copy Markdown
Contributor

The 2.14 branch has been cut and pytorch/pytorch main is now on 2.15.0a0, so trunk nightlies are 2.15.0.dev*. CURRENT_NIGHTLY_VERSION was left behind at 2.14.0.

This is the separate follow-up the go-live skill describes: the post-branch-cut automation advances CURRENT_CANDIDATE_VERSION (done in #8497) but deliberately does not touch CURRENT_NIGHTLY_VERSION.

After this the three constants describe what each channel actually serves:

CURRENT_NIGHTLY_VERSION   = "2.15.0"   # trunk nightlies
CURRENT_CANDIDATE_VERSION = "2.14.0"   # 2.14 RCs on the test channel
CURRENT_STABLE_VERSION    = "2.13.0"   # current GA

Behavioural effect

The only one is the nightly docker image tag in generate_docker_release_matrix.py, which is built from this constant:

before: ghcr.io/pytorch/pytorch-nightly:2.14.0.dev20260812-cuda12.6-cudnn9-runtime
after:  ghcr.io/pytorch/pytorch-nightly:2.15.0.dev20260812-cuda12.6-cudnn9-runtime

which now matches the nightly wheels actually being published.

The binary build matrix is unaffected: initialize_globals derives CURRENT_VERSION from CURRENT_CANDIDATE_VERSION on the test channel and CURRENT_STABLE_VERSION everywhere else, so CURRENT_NIGHTLY_VERSION never reaches it.

It also means the next branch cut starts from the right value, since release-pytorch-post-branch-cut.yml reads CURRENT_NIGHTLY_VERSION to set CURRENT_CANDIDATE_VERSION.

Test plan

  • python -m tools.tests.test_generate_binary_build_matrix — 14 tests pass.
  • --update-reference-files — no asset changes.
  • Binary build matrix byte-identical (md5) before and after for {linux, linux-aarch64, windows, macos-arm64} x {nightly, test, release}.
  • Docker release matrix nightly tag moves 2.14.0.dev -> 2.15.0.dev as above; test and release channels unchanged.

Authored with the assistance of Claude Code.

The 2.14 branch has been cut and `pytorch/pytorch` `main` is now on
`2.15.0a0`, so trunk nightlies are `2.15.0.dev*`. `CURRENT_NIGHTLY_VERSION`
was left behind at 2.14.0.

This is the separate follow-up the go-live skill describes: the post-branch-cut
automation advances `CURRENT_CANDIDATE_VERSION` (done in pytorch#8497) but
deliberately does not touch `CURRENT_NIGHTLY_VERSION`.

After this the three constants describe what each channel actually serves:

    CURRENT_NIGHTLY_VERSION   = "2.15.0"   # trunk nightlies
    CURRENT_CANDIDATE_VERSION = "2.14.0"   # 2.14 RCs on the test channel
    CURRENT_STABLE_VERSION    = "2.13.0"   # current GA

The only behavioural change is the nightly docker image tag in
`generate_docker_release_matrix.py`, which is built from this constant:

    before: ghcr.io/pytorch/pytorch-nightly:2.14.0.dev20260812-cuda12.6-cudnn9-runtime
    after:  ghcr.io/pytorch/pytorch-nightly:2.15.0.dev20260812-cuda12.6-cudnn9-runtime

which now matches the nightly wheels being published. The binary build matrix
is unaffected: `initialize_globals` derives `CURRENT_VERSION` from
`CURRENT_CANDIDATE_VERSION` on the test channel and `CURRENT_STABLE_VERSION`
everywhere else, so `CURRENT_NIGHTLY_VERSION` never reaches it.

It also means the next branch cut starts from the right value, since
`release-pytorch-post-branch-cut.yml` reads `CURRENT_NIGHTLY_VERSION` to set
`CURRENT_CANDIDATE_VERSION`.

Test plan:
- `python -m tools.tests.test_generate_binary_build_matrix` - 14 tests pass.
- `--update-reference-files` produces no asset changes.
- Binary build matrix is byte-identical (md5) before and after for
  {linux, linux-aarch64, windows, macos-arm64} x {nightly, test, release}.
- Docker release matrix nightly tag moves 2.14.0.dev -> 2.15.0.dev as above;
  the test and release channels are unchanged.

Authored with the assistance of Claude Code.
@vercel

vercel Bot commented Aug 12, 2026

Copy link
Copy Markdown

@atalman is attempting to deploy a commit to the Meta Open Source Team on Vercel.

A member of the Team first needs to authorize it.

@meta-cla meta-cla Bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Aug 12, 2026
@atalman
atalman merged commit e33a53f into pytorch:main Aug 12, 2026
88 of 95 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants