[release/2.14] Pin validation to release/2.14 and bump CURRENT_CANDIDATE_VERSION - #8495
Merged
atalman merged 1 commit intoAug 11, 2026
Conversation
…ATE_VERSION Release-only follow-up to the 2.14 branch cut (79a17ec), porting the equivalents of pytorch#8171 / pytorch#8172 / pytorch#8216 from release/2.13. ## 1. `CURRENT_CANDIDATE_VERSION` -> 2.14.0 The branch cut moved `CURRENT_NIGHTLY_VERSION` to 2.14.0 but left `CURRENT_CANDIDATE_VERSION` at 2.13.0. For the test channel `initialize_globals()` sets `CURRENT_VERSION = CURRENT_CANDIDATE_VERSION`, which the generated matrix surfaces as `stable_version` and `smoke_test.py` then enforces. Since the test channel now serves the 2.14.0 release candidate, validation fails with: ``` RuntimeError: Torch version mismatch, expected 2.13.0 for channel test ``` Verified with `generate_binary_build_matrix.py --channel test`: `stable_version` goes 2.13.0 -> 2.14.0. `CURRENT_STABLE_VERSION` stays at 2.13.0 (the current GA release); it moves separately when 2.14 goes live. ## 2. `pytorch/pytorch` checkouts -> `ref: release/2.14` The validation jobs check out `repository: "pytorch/pytorch"` with `ref: main`, so release validation ran against main's `.ci` / `test/smoke_test` sources instead of the release branch. Pinned in all seven call sites across `validate-{linux,windows,macos-arm64,aarch64-linux}-binaries.yml` and `validate-docker-images.yml`, matching release/2.13's state. ## 3. Explicit `test-infra-ref: release/2.14` in matrix generation Belt-and-braces: the branch cut already rewrote the `test-infra-ref` default in `generate_binary_build_matrix.yml` to `release/2.14`, so this is not a live bug on 2.14 (it was on 2.13, where the default stayed `main`). Passing it explicitly keeps the matrix pinned to the release branch even if that default is ever re-synced from main. ## 4. Revert a branch-cut sed artifact The cut rewrote a comment referring to the *Poetry* project's own branch, `poetry@main` -> `poetry@release/2.14`, which is meaningless. Restored. Authored with the assistance of Claude Code.
|
@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
Bot
added
the
CLA Signed
atalman
added a commit
that referenced
this pull request
Aug 11, 2026
The 2.14 branch cut is done, so the test channel now serves 2.14.0 release candidates. This applies the standard post-branch-cut update to `main`, matching what `.github/workflows/release-pytorch-post-branch-cut.yml` does. ## 1. `CURRENT_CANDIDATE_VERSION` 2.13.0 -> 2.14.0 For the test channel `initialize_globals()` sets `CURRENT_VERSION = CURRENT_CANDIDATE_VERSION`, which the generated matrix surfaces as `stable_version` and `smoke_test.py` enforces. Left at 2.13.0, test-channel validation fails with: ``` RuntimeError: Torch version mismatch, expected 2.13.0 for channel test ``` `CURRENT_STABLE_VERSION` stays at 2.13.0 (the current GA release); it moves separately when 2.14 goes live. ## 2. Sync the test channel with nightly in the arch dicts - `CUDA_ARCHES_DICT`: test gains `13.4` - `ROCM_ARCHES_DICT`: test `7.1, 7.2` -> `7.2, 7.14` - `PYTHON_ARCHES_DICT`: already matched nightly, no change These match what `release/2.14` actually builds. Without them, 2.14 RC validation would skip the `cu134` and `rocm7.14` wheels and try to validate `rocm7.1` wheels that 2.14 does not produce. The release channel matrix is unchanged. ## Test plan - `python -m tools.tests.test_generate_binary_build_matrix` — 11 tests pass. - `python -m tools.tests.test_generate_binary_build_matrix --update-reference-files` — no asset changes. - Generated linux wheel matrix, before -> after: | channel | `stable_version` | cuda | rocm | | --- | --- | --- | --- | | test | 2.13.0 -> **2.14.0** | +`cu134` | `7.1` -> **`7.14`** | | release | unchanged (2.13.0) | unchanged | unchanged | | nightly | unchanged | unchanged | unchanged | Companion to #8495, which made the equivalent change on `release/2.14`. Authored with the assistance of Claude Code.
atalman
added a commit
that referenced
this pull request
Aug 11, 2026
…ds (#8498) Follow-up to #8495, which bumped `CURRENT_CANDIDATE_VERSION` to 2.14.0 on this branch but left the arch dicts on their 2.13-era values. The test channel now serves 2.14.0 release candidates, so its arch lists must match what `pytorch/pytorch` `release/2.14` actually builds ([`.github/scripts/generate_binary_build_matrix.py`](https://github.com/pytorch/pytorch/blob/release/2.14/.github/scripts/generate_binary_build_matrix.py)): ```python CUDA_ARCHES = ["12.6", "13.0", "13.2", "13.4"] ROCM_ARCHES = ["7.2", "7.14"] ``` ## Changes | dict | before | after | why | | --- | --- | --- | --- | | `CUDA_ARCHES_DICT["test"]` | `12.6, 13.0, 13.2` | `12.6, 13.0, 13.2, 13.4` | cu134 RC wheels were never validated | | `ROCM_ARCHES_DICT["test"]` | `7.1, 7.2` | `7.2, 7.14` | `7.1` is not built for 2.14 — validation looked for wheels that don't exist and skipped rocm7.14 | | `PYTHON_ARCHES_DICT["test"]` | — | unchanged | already matches nightly | The `release` channel entries are deliberately left alone — they still describe GA 2.13 and move as part of the 2.14 go-live step. ## Test plan - `python -m tools.tests.test_generate_binary_build_matrix` — 11 tests pass. - `python -m tools.tests.test_generate_binary_build_matrix --update-reference-files` — no asset changes. - Generated linux wheel matrix, before -> after: | channel | `stable_version` | cuda | rocm | | --- | --- | --- | --- | | test | 2.14.0 (unchanged) | +`cu134` | `7.1` -> **`7.14`** | | release | unchanged (2.13.0) | unchanged | unchanged | | nightly | unchanged | unchanged | unchanged | #8497 makes the equivalent arch sync on `main`. Authored with the assistance of Claude Code.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Release-only follow-up to the 2.14 branch cut (79a17ec), porting the equivalents of #8171 / #8172 / #8216 from release/2.13.
1.
CURRENT_CANDIDATE_VERSION-> 2.14.0The branch cut moved
CURRENT_NIGHTLY_VERSIONto 2.14.0 but leftCURRENT_CANDIDATE_VERSIONat 2.13.0. For the test channelinitialize_globals()setsCURRENT_VERSION = CURRENT_CANDIDATE_VERSION, which the generated matrix surfaces asstable_versionandsmoke_test.pythen enforces. Since the test channel now serves the 2.14.0 release candidate, validation fails with:Verified with
generate_binary_build_matrix.py --channel test:stable_versiongoes 2.13.0 -> 2.14.0.CURRENT_STABLE_VERSIONstays at 2.13.0 (the current GA release); it moves separately when 2.14 goes live.2.
pytorch/pytorchcheckouts ->ref: release/2.14The validation jobs check out
repository: "pytorch/pytorch"withref: main, so release validation ran against main's.ci/test/smoke_testsources instead of the release branch. Pinned in all seven call sites acrossvalidate-{linux,windows,macos-arm64,aarch64-linux}-binaries.ymlandvalidate-docker-images.yml, matching release/2.13's state.3. Explicit
test-infra-ref: release/2.14in matrix generationBelt-and-braces: the branch cut already rewrote the
test-infra-refdefault ingenerate_binary_build_matrix.ymltorelease/2.14, so this is not a live bug on 2.14 (it was on 2.13, where the default stayedmain). Passing it explicitly keeps the matrix pinned to the release branch even if that default is ever re-synced from main.4. Revert a branch-cut sed artifact
The cut rewrote a comment referring to the Poetry project's own branch,
poetry@main->poetry@release/2.14, which is meaningless. Restored.Authored with the assistance of Claude Code.