[Auto] Fix DPMSolverMultistepScheduler duplicate timesteps (cluster-12771-4): merged 1 of 3 PRs

[evalstate]

Cluster: cluster-12771-4
Base: origin/main
Branch: merge-cluster-cluster-12771-4-20260427125508

Merged PRs:

• Fix duplicate timesteps in DPMSolverMultistepScheduler with sigma conversion methods huggingface/diffusers#12788: Canonical open PR; merged locally with conflict resolution, adding timestep deduplication for sigma conversion paths and regression coverage.

Skipped PRs:

• fix: deduplicate timesteps in DPMSolverMultistep for squaredcos_cap_v2 + karras/lu sigmas huggingface/diffusers#13240: Open but narrower duplicate of Fix duplicate timesteps in DPMSolverMultistepScheduler with sigma conversion methods huggingface/diffusers#12788; conflicted after Fix duplicate timesteps in DPMSolverMultistepScheduler with sigma conversion methods huggingface/diffusers#12788 in the same scheduler Karras/Lu timestep conversion blocks, so merge was aborted.
• fix: deduplicate timesteps in DPMSolverMultistepScheduler to prevent IndexError huggingface/diffusers#13239: Closed unmerged duplicate/superseded alternative; conflicted after Fix duplicate timesteps in DPMSolverMultistepScheduler with sigma conversion methods huggingface/diffusers#12788 in the same scheduler Karras/Lu timestep conversion blocks, so merge was aborted.

Failed PRs:

• None.

Notes:

• Cluster issue DPMSolverMultistepScheduler out-of-bounds error when using beta_schedule="squaredcos_cap_v2" + use_karras_sigmas=True huggingface/diffusers#12771 remains open and concerns DPMSolverMultistepScheduler out-of-bounds errors with squaredcos_cap_v2 and sigma conversion methods.
• Final diff changes src/diffusers/schedulers/scheduling_dpmsolver_multistep.py and tests/schedulers/test_scheduler_dpm_multi.py.
• py_compile passed for the changed scheduler and test files.
• Targeted pytest and ruff/style were not run because pytest/ruff are unavailable in this environment.

Next steps:

• Review the Fix duplicate timesteps in DPMSolverMultistepScheduler with sigma conversion methods huggingface/diffusers#12788 merge resolution, especially warning behavior when requested inference steps collapse to fewer unique timesteps.
• Run targeted scheduler tests and make style in a fully provisioned environment.
• Treat fix: deduplicate timesteps in DPMSolverMultistepScheduler to prevent IndexError huggingface/diffusers#13239 and fix: deduplicate timesteps in DPMSolverMultistep for squaredcos_cap_v2 + karras/lu sigmas huggingface/diffusers#13240 as duplicate alternatives unless maintainers prefer their narrower Karras/Lu-only scope.

[evalstate]

Trace for this mergeability run: https://huggingface.co/datasets/evalstate/transformers-merge-experiments/blob/main/2604271459-AYLeQw__dev__codex.jsonl