[generation] Fix misleading synced_gpus warning in continuous batching

[Partha-Shankar]

What does this PR do?

This PR fixes a misleading user-facing warning message when calling generate() with synced_gpus=True in continuous batching mode (e.g. cache_implementation="paged").
Closes #47276

Summary

The warning message previously stated:
synced_gpus is not ignored for continuous batching. Got synced_gpus = True
This double negative incorrectly implies that synced_gpus is active or supported. In fact, synced_gpus is discarded during continuous batching execution. The comment directly above the warning also explicitly reads # others are ignored.
This PR corrects the warning to state:
synced_gpus is ignored for continuous batching. Got synced_gpus = True

Root Cause

A simple negation typo in the warning string format.

Solution

Change "is not ignored" to "is ignored" in the warning message in src/transformers/generation/utils.py.

Why This Is a Standalone Fix

This PR addresses a user-facing negation typo in the warning message logged
when synced_gpus is passed during continuous batching. Because continuous
batching ignores synced_gpus, the warning mistakenly read "synced_gpus is
not ignored...", which can cause significant confusion for users debugging
distributed setups who believe the parameter is active. As this is a highly
visible, misleading warning specifically affecting the continuous batching
entry point, and a search of the generation codebase confirmed no other
stale warning negation bugs nearby, it is submitted as a targeted, standalone
fix rather than bundled into a broader cleanup.

Tests Run

• Verified the fix manually and confirmed no regressions via the full existing test suite: pytest tests/generation/test_continuous_batching.py -v → 49 passed, 71 skipped, 0 failed.
• Checked code style and formatting via ruff check and ruff format locally.

AI-Assisted Tooling & Guidelines Compliance

• AI Tooling Used: Google Antigravity was used to review the codebase and
  assist in understanding the continuous batching parameter validation flow.
  The fix was authored directly by the developer.
• Coordination Link: Misleading synced_gpus warning message in continuous batching #47276
• Duplicate-PR Check: Searched open and recently-closed pull requests on
  huggingface/transformers for PRs touching the same warning string or file
  (src/transformers/generation/utils.py). No duplicates or similar
  sweeping cleanup PRs were found.
• Verification & Test Output:

  $ ruff check src/transformers/generation/utils.py
  All checks passed!

  $ ruff format --check src/transformers/generation/utils.py
  1 file already formatted

  $ pytest tests/generation/test_continuous_batching.py -v
  ======================= 49 passed, 71 skipped in 12.14s =======================

[remi-or]

Thanks for fixing the nit, please remove the test and I will merge.

[Partha-Shankar]

Done — removed the test, only the warning-message fix remains. Thanks for the review!

[remi-or]

LGTM.

[github-actions]

---

CI recap

Dashboard: View test results in Grafana
Latest run: 29225200277:2
Result: failure | Jobs: 15 | Tests: 170,777 | Failures: 0 | Duration: 15h 26m

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.