[YamonBot]

Remote API VLM pipelines cannot import or initialize in a slim environment without optional local ML dependencies. VlmPipeline imports the legacy HF and MLX providers before selecting an inference backend, so remote-only setup fails with ModuleNotFoundError.

Move each import into its existing local-provider branch. This uses the same provider-selection pattern as the VLM engine factory. Local provider arguments, remote-service permission checks, packaging, and conversion output remain unchanged.

Resolves #4966.

Checklist:

• Documentation reviewed; no public API or installation metadata changes.
• Examples reviewed; existing API options remain unchanged.
• Added four fresh-subprocess regressions for current and legacy API options, with remote services enabled and denied.

Review evidence:

• Base: 10054ec7bbdc9435ef7ca2ec6be90c782082ae52.
• Python 3.12.14, docling-slim 2.141.0 editable, docling-core 2.103.1, Pillow 12.3.0. Torch, Transformers, and MLX packages are physically absent from the isolated environment. No replacement modules or model weights used.
• Original source: all four new cases failed at the eager HF provider import. Corrected source: four passed. The assertions track provider module imports, so installed optional packages in other CI environments are not themselves treated as failures.
• PYTHONDONTWRITEBYTECODE=1 /tmp/docling-remote-provider-py312-dab32661/bin/python -m pytest -p no:tach -p no:cacheprovider tests/test_vlm_pipeline_optional_dependencies.py tests/test_api_vlm_engine.py tests/test_api_image_request.py tests/test_vlm_pipeline_status.py tests/test_vlm_pipeline_streaming.py tests/test_vlm_pipeline_force_backend_text.py tests/test_vlm_enrichment.py tests/test_vlm_convert_prediction_mapping.py tests/test_vlm_utils.py tests/test_vlm_prediction_token_coercion.py -q: 80 passed, 1 skipped after final review.
• UV_NO_SYNC=1 UV_PROJECT_ENVIRONMENT=/tmp/docling-remote-provider-py312-dab32661 make validate: passed with repository-pinned tools. The flag preserves the deliberately minimal environment.
• Native probes still select the HF and MLX branches and raise their missing-dependency errors when local inference is selected. Successful local model inference was not run.
• No reference data regenerated. No real remote inference or model download performed.

Assisted by Codex. Patch and executable evidence reviewed before submission. DCO sign-off included.

[github-actions]

✅ DCO Check Passed

Thanks @YamonBot, all your commits are properly signed off. 🎉

[mergify]

Merge Protections

🟢 Merge protection satisfied — ready to merge.

Show 1 satisfied protection

🟢 Enforce conventional commit

Make sure that we follow https://www.conventionalcommits.org/en/v1.0.0/

• title ~= ^(fix|feat|docs|style|refactor|perf|test|build|ci|chore|revert)(?:\(.+\))?(!)?:

[github-actions]

AI triage (advisory)

Checked commit 30891e1789. The analysis read the diff only. It did not run code from this pull request.

Warning

The AI analysis did not finish. Only the deterministic checks are shown.

Topic labels added: bug, vlm-pipeline.

CI safety: not assessed.

Possibly related pull requests

PR	State	Signal	Assessment
#4964	open	changes the same code	not assessed

_{The ai:* labels are advisory. They do not approve CI and they do not add tests:full. A new push runs this check again.}

[dolfim-ibm]

lgtm

[codecov]

Codecov Report

❌ Patch coverage is 0% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
docling/pipeline/vlm_pipeline.py	0.00%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!