Repository navigation
Handle channels with fewer CUDA versions in gen_quick_start_module.py - #2139
Merged
Merged
Conversation
`extract_arch_ver_map` fills `cuda.x`/`cuda.y`/`cuda.z` from the channel's CUDA versions and leaves the rest at the hard-coded defaults. Nightly now ships only 13.2 and 13.4, so the page advertised nightly as CUDA 13.2, 13.4 and 12.4 (or 13.2, 12.1, 12.4 while 13.4 was hidden from getting-started). - Drop CUDA labels the channel has no version for. The page already strikes out any option missing from `version_map`. - When the channel has a CUDA version but the OS has no build for it (13.4 on Windows), publish a "not available" note instead of keeping the previous command, which pointed at a different CUDA version. - A new stable entry is copied from `preview`, so recreate libtorch `versions` when a CUDA entry matches; otherwise a "not available" preview entry left the new release without its download link. Test plan: ran the update-quick-start-module workflow locally (six getting-started matrices from pytorch/test-infra, then `gen_quick_start_module.py --autogenerate`): - With test-infra advertising nightly 13.2 and 13.4: nightly map is `cuda.x=13.2, cuda.y=13.4`, Linux commands use `nightly/cu132` and `nightly/cu134`, Windows 13.4 shows "CUDA 13.4 is not available on Windows". - With test-infra still hiding 13.4: nightly map is `cuda.x=13.2` only. - Release (2.14.1) output is identical to the unpatched script.
✅ Deploy Preview for pytorch-dot-org-preview ready!
To edit notification comments on pull requests, go to your Netlify project configuration. |
atalman
had a problem deploying
to
pytorchbot-env
September 30, 2026 18:54 — with 
GitHub Actions
Failure
GitHub Actions
Failure
atalman
had a problem deploying
to
pytorchbot-env
September 30, 2026 18:55 — with
GitHub Actions
Failure
GitHub Actions
Failure
The page has three fixed CUDA boxes. For a channel with fewer CUDA versions (nightly now ships 13.2 and 13.4), `changeVersion()` only struck the unused box through: it stayed visible with a stale label, and stayed clickable. Hide CUDA boxes that are not in the channel's `version_map`, widen the visible ones so the CUDA group keeps its 6 of the row's 12 columns (two boxes at `col-md-3`, three at `col-md-2`), and move the selection to the first visible box if the selected one was hidden. ROCm and CPU keep the existing strike-through behaviour. Test plan: generated `assets/quick-start-module.js` from this template with the nightly/release getting-started matrices, then drove it in Node with a minimal DOM/jQuery stand-in (npm is unavailable on the devserver, so no jsdom or browser): - Stable: three boxes, CUDA 12.6 / 13.0 / 13.2, `col-md-2`. - Preview: two boxes, CUDA 13.2 / 13.4, `col-md-3`; third hidden. Commands use `nightly/cu132` and `nightly/cu134`. - Stable with CUDA 13.2 (`cuda.z`) selected, then Preview: selection moves to `cuda.x` and the command follows. With the previous template the struck-out box stayed selected. - Preview, Windows, CUDA 13.4: shows the "not available on Windows" note.
atalman
had a problem deploying
to
pytorchbot-env
September 30, 2026 19:00 — with
GitHub Actions
Failure
GitHub Actions
Failure
atalman
had a problem deploying
to
pytorchbot-env
September 30, 2026 19:00 — with
GitHub Actions
Failure
GitHub Actions
Failure
atalman
added a commit
that referenced
this pull request
Sep 30, 2026
After #2139 the page hides CUDA boxes the channel does not ship, but `update_versions` only skipped those labels, so `preview` in published_versions.json still carried `cuda.z` with a stale command on every OS and package. Add `sync_cuda_entries`, run before each channel's update: - nightly (preview): remove `cuda.x/y/z` entries the channel has no CUDA version for. - A new stable is created as a copy of preview, so re-create any label the release channel needs from a sibling CUDA entry, in cuda.x/y/z order. The existing update then fills in its command. Release entries are never dropped. Test plan: simulated the update-quick-start-module run (getting-started matrices from pytorch/test-infra main, published_versions.json from `site`): - preview: `cuda.z` removed for linux/windows/macos, pip and libtorch. - 2.14.1 (created from preview in this run): has cuda.x/y/z in order, and the whole file is identical to the current generator's output apart from the removed preview `cuda.z` entries. - Page logic (Node harness): nightly two boxes, stable three, selection moves off the hidden box, as before. - `sync_cuda_entries` edge cases: nightly back to three CUDA versions re-adds `cuda.z`; nightly with one drops `cuda.y`/`cuda.z`; release never drops entries. --------- Co-authored-by: Andrey Talman <atalman@users.noreply.github.com>
This branch had an error being deployed
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.
extract_arch_ver_mapfillscuda.x/cuda.y/cuda.zfrom the channel'sCUDA versions and leaves the rest at the hard-coded defaults. Nightly now
ships only 13.2 and 13.4, so the page advertised nightly as CUDA 13.2, 13.4
and 12.4 (or 13.2, 12.1, 12.4 while 13.4 was hidden from getting-started).
out any option missing from
version_map.Windows), publish a "not available" note instead of keeping the previous
command, which pointed at a different CUDA version.
preview, so recreate libtorchversionswhen a CUDA entry matches; otherwise a "not available" preview entry left the
new release without its download link.
Test plan: ran the update-quick-start-module workflow locally (six
getting-started matrices from pytorch/test-infra, then
gen_quick_start_module.py --autogenerate):cuda.x=13.2, cuda.y=13.4, Linux commands usenightly/cu132andnightly/cu134, Windows 13.4 shows "CUDA 13.4 is not available on Windows".cuda.x=13.2only.