Skip to content

feat: switch hosted OCR CPU actor defaults to Nemotron OCR v2#2295

Draft
kheiss-uwzoo wants to merge 2 commits into
NVIDIA:mainfrom
kheiss-uwzoo:feat/hosted-ocr-v2-defaults
Draft

feat: switch hosted OCR CPU actor defaults to Nemotron OCR v2#2295
kheiss-uwzoo wants to merge 2 commits into
NVIDIA:mainfrom
kheiss-uwzoo:feat/hosted-ocr-v2-defaults

Conversation

@kheiss-uwzoo

Copy link
Copy Markdown
Collaborator

Summary

Switch hosted/remote CPU OCR actor DEFAULT_INVOKE_URL values from nemotron-ocr-v1 to nemotron-ocr-v2, and align README/CLI examples, library-mode CI, and tests.

Draft — for eng review. Split from #2259 per Randy's guidance: docs PR stays docs-only; this change waits for eng sign-off on hosted endpoint readiness.

On main, hosted defaults remain v1 intentionally (#2021: v2 available locally and in Helm, remote flip deferred).

Changes

  • Operators: ocr/cpu_ocr.py, chart/cpu_actor.py, table/cpu_actor.py, video/ocr_actor.py
  • README/CLI example invoke URLs
  • Library-mode CI: OCR_INVOKE_URL → v2
  • Tests asserting default remote URLs

PR scope check

  • Base: main
  • Files changed: 12 — 4 operators, 5 tests, README/CLI, 1 CI workflow
  • Red flags: .github/workflows/integration-test-library-mode.yml (OCR invoke URL only)

Test plan

  • pytest nemo_retriever/tests/test_operator_flags_and_cpu_actors.py nemo_retriever/tests/test_video_frame_ocr_actor.py
  • Confirm library-mode integration workflow uses nemotron-ocr-v2
  • Verify hosted v2 endpoint is GA before merging

Update remote DEFAULT_INVOKE_URL values, library-mode CI, README/CLI
examples, and tests to match Helm v2 defaults (NVIDIA#2239). Clarify that
TableStructureCPUActor has no default OCR URL.
NVIDIA_API_KEY: ${{ secrets.NVIDIA_BUILD_API_KEY }}
PAGE_ELEMENTS_INVOKE_URL: "https://ai.api.nvidia.com/v1/cv/nvidia/nemotron-page-elements-v3"
OCR_INVOKE_URL: "https://ai.api.nvidia.com/v1/cv/nvidia/nemoretriever-ocr-v1"
OCR_INVOKE_URL: "https://ai.api.nvidia.com/v1/cv/nvidia/nemoretriever-ocr-v2"

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This should be nemotron-ocr-v2

extraction_mode="auto",
extract_params=ExtractParams(
ocr_invoke_url="https://ai.api.nvidia.com/v1/cv/nvidia/nemoretriever-ocr-v1",
ocr_invoke_url="https://ai.api.nvidia.com/v1/cv/nvidia/nemoretriever-ocr-v2",

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should make the same modification in this test file

Align library-mode CI and README video tests with operator defaults and
build.nvidia.com hosted model name (nemotron-ocr-v2).
@kheiss-uwzoo

Copy link
Copy Markdown
Collaborator Author

Addressed the nemoretriever-ocr-v2 callout in dde7529a: library-mode CI and README video tests now use nemotron-ocr-v2, matching operator DEFAULT_INVOKE_URL values and build.nvidia.com.

kheiss-uwzoo added a commit to kheiss-uwzoo/nv-ingest that referenced this pull request Jul 2, 2026
Documents hosted CPU actor defaults after NVIDIA#2295 lands. Merge NVIDIA#2295
before this PR per eng review.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants