Pass merge levels to remote OCR

[charlesbluca]

Description

Remote HTTP OCR now passes explicit merge_levels into NIM image-inference requests so it matches the local OCR path:

• Full-image/video OCR repeats the actor's configured merge level for every valid image in the batch.
• Page-element OCR sends word for table crops and paragraph for charts, infographics, and text/title/header-footer crops.
• Graphic-elements chart OCR sends word, matching its local OCR behavior.

Root cause: NIMClient already supported per-image merge_levels, but several graph paths built remote OCR requests without populating that field, so the endpoint default was used instead of modality-specific local behavior.

Validation:

• uv run --extra dev pytest tests/test_video_frame_ocr_actor.py tests/test_table_structure.py tests/test_chart_graphic_elements.py
• git diff --check

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.
• Not applicable: no docker-compose.yaml environment variables changed.

[greptile-apps]

Greptile Summary

This PR fixes a behavioral gap where remote NIM OCR calls in several graph paths were not sending explicit merge_levels, causing the endpoint to apply its default instead of the modality-specific levels used by the local OCR path.

• ocr/shared.py: Introduces _MERGE_LEVEL_BY_LABEL (a module-level mapping from label name to merge level) and a helper _merge_level_for_ocr_label() that raises ValueError on unknown labels. Both the ocr_b64_to_text (full-image/video path) and ocr_page_elements (page-element path) remote code paths are updated to populate merge_levels per-image before calling invoke_image_inference_batches.
• chart/shared.py: The graphic_elements_ocr_page_elements remote path is updated to send \"word\" merge level for every chart crop, matching its local OCR counterpart.
• Tests (test_video_frame_ocr_actor.py, test_table_structure.py, test_chart_graphic_elements.py): New unit tests assert that remote OCR calls carry the correct merge_levels values for each modality, and that the mapping rejects unknown labels.

Confidence Score: 5/5

Safe to merge — the changes are narrowly scoped to adding explicit merge_level forwarding on remote paths that previously relied on the endpoint default.

All three remote OCR paths now populate merge_levels in a way that is consistent with their local counterparts, and the new _merge_level_for_ocr_label helper raises an explicit ValueError for unrecognized labels rather than silently falling through. The label set passed to the helper is fully covered by _MERGE_LEVEL_BY_LABEL under the current call sites. New tests cover the happy path and the label-rejection case, and the existing test suite has been extended to assert on the new merge_levels argument.

No files require special attention.

Important Files Changed

Filename	Overview
nemo_retriever/src/nemo_retriever/ocr/shared.py	Adds _MERGE_LEVEL_BY_LABEL + helper to consistently derive per-label merge levels; updates remote paths in ocr_b64_to_text and ocr_page_elements to pass explicit merge_levels; refactors local-path local_jobs init to derive keys dynamically from the same mapping.
nemo_retriever/src/nemo_retriever/chart/shared.py	One-line fix to pass merge_levels=["word"] * len(flat_crop_b64s) in the graphic-elements remote OCR call, matching the local path.
nemo_retriever/tests/test_table_structure.py	New tests verify: _merge_level_for_ocr_label mapping correctness + rejection of unknown labels; local path respects monkeypatched mapping; remote path sends per-modality merge levels for mixed-label pages.
nemo_retriever/tests/test_chart_graphic_elements.py	Adds a test asserting the remote graphic-elements OCR call passes merge_levels=["word"].
nemo_retriever/tests/test_video_frame_ocr_actor.py	Extends existing batched-call test to assert merge_levels and adds a new test verifying that a non-default merge_level configured on the actor is forwarded to the remote call.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[OCR Request] --> B{Remote or Local?}
    B -- Remote --> C{Which path?}
    B -- Local --> D{Which path?}

    C -- "ocr_b64_to_text\n(video/full-image)" --> E["merge_levels = [merge_level] * len(valid_b64)"]
    C -- "ocr_page_elements\n(page elements)" --> F["merge_levels = [_merge_level_for_ocr_label(label)\nfor each crop]"]
    C -- "graphic_elements_ocr_page_elements\n(chart GE crops)" --> G["merge_levels = ['word'] * len(flat_crop_b64s)"]

    E --> H[invoke_image_inference_batches]
    F --> H
    G --> H

    D -- "ocr_page_elements\n(local jobs)" --> I["local_jobs = {ml: [] for ml in _MERGE_LEVEL_BY_LABEL.values()}"]
    I --> J["_merge_level_for_ocr_label(label_name) for each crop"]
    J --> K[model.invoke per merge-level batch]

    subgraph Mapping ["_MERGE_LEVEL_BY_LABEL"]
        M1["table → 'word'"]
        M2["chart → 'paragraph'"]
        M3["infographic → 'paragraph'"]
        M4["text / title / header_footer → 'paragraph'"]
    end

    F -. uses .-> Mapping
    J -. uses .-> Mapping

Loading

_{Reviews (5): Last reviewed commit: "Merge branch 'main' into codex/remote-oc..." | Re-trigger Greptile}