feat: adding enrichments after the VLM pipeline

[PeterStaar-IBM]

Summary

• Run picture classification, description, and chart extraction after the VLM pipeline. Classification runs before chart extraction.
• Skip metadata already supplied by the VLM and request only missing chart outputs. Keep picture crops available until enrichment finishes.
• Restore the older CSV-only Granite Vision Chart2CSV preset and expose chart preset selection in the CLI.
• Select MLX automatically for Granite Vision 4.1 on compatible Apple Silicon systems, with Transformers fallback. Add an explicit MLX preset and raise the mlx-vlm minimum to 0.7.0.
• Update model download support, documentation, and regression tests.

Checklist:

• Documentation has been updated, if necessary.
• Examples have been added, if necessary.
• Tests have been added, if necessary.

[github-actions]

✅ DCO Check Passed

Thanks @PeterStaar-IBM, 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)(?:\(.+\))?(!)?:

[dosubot]

📄 Knowledge review

✏️ Suggested updates

1 page suggestion needs review.

Page	Library	Status
skill.md	Docling	🟡 Review

📝 skill.md

@@ -395,6 +395,7 @@
 | `--enrich-picture-classes` | Enable picture classification |
 | `--enrich-picture-description` | Enable picture description |
 | `--enrich-chart-extraction` | Enable chart data extraction |
+| `--chart-extraction-preset PRESET` | Chart extraction preset (granite_vision_v4, granite_vision) |
 | `--image-export-mode MODE` | Image export mode for HTML/MD/JSON outputs |
 | `--num-threads N` | Number of threads |
 | `--device DEVICE` | Accelerator device (cpu, cuda, mps) |
@@ -1114,11 +1115,41 @@
 from docling.datamodel.pipeline_options import (
     PdfPipelineOptions, ChartExtractionModelKind
 )
-
+from docling.datamodel.chart_extraction_options import ChartExtractionVlmEngineOptions
+
+# Using PdfPipelineOptions (for PDF pipeline)
 opts = PdfPipelineOptions(
     do_picture_classification=True,
-    chart_extraction_model=ChartExtractionModelKind.GRANITE_VISION,
-    # or: ChartExtractionModelKind.GRANITE_VISION_V4 (runs chart2csv, chart2code, chart2summary)
+    do_chart_extraction=True,
+    # Default: from_preset('granite_vision_v4') — chart2csv, chart2code, chart2summary
+    chart_extraction_options=ChartExtractionVlmEngineOptions.from_preset('granite_vision_v4'),
+)
+
+# CSV-only model (older Chart2CSV model)
+opts = PdfPipelineOptions(
+    do_picture_classification=True,
+    do_chart_extraction=True,
+    chart_extraction_options=ChartExtractionVlmEngineOptions.from_preset('granite_vision'),
+)
+
+# Backwards-compatible enum (deprecated; use ChartExtractionVlmEngineOptions instead)
+opts = PdfPipelineOptions(
+    do_picture_classification=True,
+    chart_extraction_model=ChartExtractionModelKind.GRANITE_VISION,  # CSV-only model
+    # or: ChartExtractionModelKind.GRANITE_VISION_V4 (chart2csv, chart2code, chart2summary)
+)
+```
+
+Chart extraction enrichments can also be configured for `VlmPipelineOptions` (VLM pipeline):
+
+```python
+from docling.datamodel.pipeline_options import VlmPipelineOptions
+from docling.datamodel.chart_extraction_options import ChartExtractionVlmEngineOptions
+
+# Enable chart extraction in VLM pipeline
+vlm_opts = VlmPipelineOptions(
+    do_chart_extraction=True,
+    chart_extraction_options=ChartExtractionVlmEngineOptions.from_preset('granite_vision_v4'),
 )
 ```
 
@@ -1331,6 +1362,7 @@
 from docling.datamodel.pipeline_options import (
     PdfPipelineOptions,
     PipelineOptions,
+    VlmPipelineOptions,
     # OCR engines
     EasyOcrOptions,
     TesseractCliOcrOptions,
@@ -1348,6 +1380,7 @@
     CodeFormulaVlmOptions,
     ChartExtractionModelKind,
 )
+from docling.datamodel.chart_extraction_options import ChartExtractionVlmEngineOptions
 from docling.datamodel.vlm_engine_options import (
     TransformersVlmEngineOptions,
     MlxVlmEngineOptions,

Accept · Edit · Decline

---

[codecov]

Codecov Report

❌ Patch coverage is 72.28916% with 23 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...g/models/stages/chart_extraction/granite_vision.py	64.86%	9 Missing and 4 partials ⚠️
docling/pipeline/vlm_pipeline.py	63.63%	6 Missing and 2 partials ⚠️
docling/cli/main.py	75.00%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!