OCR v2 defaults, captioning link, B200 nemotron-parse (NVBug 6204537)

[kheiss-uwzoo]

Summary

Doc-only updates for extraction documentation on main.

• NVBug 6204537: Correct the support matrix so B200 shows nemotron-parse as deployable (1 GPU, ~16GB disk), matching successful Helm deployment of nemotron-parse-v1.2 (NIMCache Ready, NIMService/Pod Running/Ready). RTX Pro 6000 and H200 NVL remain Not supported; footnote ² still applies only to 32GB (RTX PRO 4500). Documentation correctness only; separate from end-to-end SDK workflow issues in NVBug 6198661.
• OCR v2 multilingual defaults: Adds Nemotron OCR v2 language mode—local Hugging Face inference defaults to multilingual (multi), with --ocr-lang english / --ocr-version v1 documented—and a cross-link from OCR and scanned documents.
• Helm / NIM: Points to nimOperator.ocr; when the chart targets nemotron-ocr-v2, the deployed NIM also defaults to multilingual (confirm repository / tag before upgrade).
• Captioning Related link: Trims redundant hardware prose from the image-captioning cross-link in multimodal-extraction.md.

Test plan

• MkDocs build for extraction docs
• Verify anchor #nemotron-ocr-v2-language-mode resolves
• Confirm nemotron-parse B200 cells render as 1 / ~16GB in the support matrix table
• Spot-check OCR section and captioning Related link in rendered HTML

[greptile-apps]

Greptile Summary

This documentation-only PR updates the extraction docs on main with three changes: it adds a new "Nemotron OCR v2 language mode" section to the support matrix explaining multilingual defaults for local HuggingFace inference and Helm/NIM deployments, adds a corresponding cross-link paragraph in the OCR section of multimodal-extraction.md, and updates the nemotron-parse hardware matrix to mark B200 as supported (1 GPU, ~16GB disk).

• OCR v2 defaults: New #nemotron-ocr-v2-language-mode subsection documents multilingual default behavior for local HF inference and points to nimOperator.ocr in the Helm chart; a matching anchor link is added in multimodal-extraction.md.
• B200 support matrix correction: nemotron-parse rows now correctly show 1 GPU / ~16GB disk for B200, consistent with a verified Helm deployment of nemotron-parse-v1.2; RTX Pro 6000 and H200 NVL remain "Not supported".
• Captioning related link: Trailing descriptor text trimmed from the image-captioning cross-link.

Confidence Score: 5/5

Documentation-only change; no code paths altered. Safe to merge.

All changes are confined to two Markdown files. The new #nemotron-ocr-v2-language-mode anchor matches its cross-reference exactly, the B200 table cells are updated consistently across both nemotron-parse rows, and the trimmed captioning link remains valid.

No files require special attention.

Important Files Changed

Filename	Overview
docs/docs/extraction/prerequisites-support-matrix.md	Adds 'Nemotron OCR v2 language mode' subsection with MkDocs anchor and a note covering both local HF and Helm/NIM defaults; corrects nemotron-parse B200 cells from 'Not supported' to 1 GPU / ~16GB disk.
docs/docs/extraction/multimodal-extraction.md	Adds a one-paragraph OCR v2 multilingual-default explanation with a cross-link to the new anchor in the support matrix; trims redundant descriptor from the image-captioning related link. Anchor reference matches the newly added section.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[User runs OCR extraction] --> B{Deployment type?}
    B -->|Local HuggingFace| C[Nemotron OCR v2\ndefault: multilingual 'multi']
    B -->|Helm / NIM| D[nimOperator.ocr block\nin values.yaml]
    B -->|Remote NIM endpoint| E[NIM own model & language\nbehavior — CLI selectors\nnot forwarded]
    C --> F{Language override needed?}
    F -->|English only v2| G[--ocr-lang english\nocr_lang API param]
    F -->|Legacy engine| H[--ocr-version v1]
    F -->|Keep multilingual| I[No flag needed]
    D --> J{Chart targets nemotron-ocr-v2?}
    J -->|Yes| K[NIM runs in\nmultilingual mode by default]
    J -->|No / other tag| L[Confirm repository & tag\nbefore upgrading]

Loading

_{Reviews (4): Last reviewed commit: "Merge branch 'main' into kheiss/docs-ocr..." | Re-trigger Greptile}

[greptile-apps]

Hard-coded line anchor points to the v1 block, not v2

The link values.yaml#L817-L852 lands on the image: sub-key of the ocr: block — which on main is configured with repository: nvcr.io/nim/nvidia/nemotron-ocr-v1 (comment on line 814 explicitly says "Nemotron OCR v1"). The ocr: key itself starts at line 815, so the anchor is also two lines low. A reader following this link will see a v1 configuration, directly contradicting the surrounding note about v2 multilingual defaults. The conditional hedge ("When that block targets nemotron-ocr-v2 for your release") does not prevent the confusion — it just shifts blame. Either update values.yaml on main to reference v2 first, or remove the fragile #L817-L852 fragment and link to the nimOperator.ocr section of the Helm README instead.

Prompt To Fix With AI

This is a comment left during a code review.
Path: docs/docs/extraction/prerequisites-support-matrix.md
Line: 79

Comment:
**Hard-coded line anchor points to the v1 block, not v2**

The link `values.yaml#L817-L852` lands on the `image:` sub-key of the `ocr:` block — which on `main` is configured with `repository: nvcr.io/nim/nvidia/nemotron-ocr-v1` (comment on line 814 explicitly says "Nemotron OCR v1"). The `ocr:` key itself starts at line 815, so the anchor is also two lines low. A reader following this link will see a v1 configuration, directly contradicting the surrounding note about v2 multilingual defaults. The conditional hedge ("When that block targets nemotron-ocr-v2 for your release") does not prevent the confusion — it just shifts blame. Either update `values.yaml` on `main` to reference v2 first, or remove the fragile `#L817-L852` fragment and link to the `nimOperator.ocr` section of the Helm README instead.

How can I resolve this? If you propose a fix, please make it concise.