Skip to content

Expand OCR provider support and refresh benchmarking/docs - #20

Merged
s-jse merged 58 commits into
mainfrom
improve-docs
Apr 15, 2026
Merged

Expand OCR provider support and refresh benchmarking/docs#20
s-jse merged 58 commits into
mainfrom
improve-docs

Conversation

@s-jse

@s-jse s-jse commented Apr 15, 2026

Copy link
Copy Markdown
Member

Summary

This PR expands Churro's built-in OCR coverage, improves leaderboard and benchmarking ergonomics, and refreshes the docs/onboarding flow.

What changed

  • add built-in OCR profiles, templates, and docs for FireRed-OCR, GLM-OCR, Infinity-Parser-7B, MinerU2.5, Nanonets-OCR2-3B, and Qianfan-OCR
  • improve the benchmark leaderboard with expandable per-language printed/handwritten scores, updated icons/results, and simplified score formatting
  • add reasoning_effort support for litellm and openai-compatible benchmarking, handle timed-out OCR pages more gracefully, close cached LiteLLM async clients, and tighten retry timeout behavior
  • refactor shared provider/page-detection helpers and tighten Ruff/ty typing and complexity checks across the library, tooling, and tests
  • reorganize the README, getting started guide, CLI docs, and API docs around a clearer CLI-first onboarding path

Breaking changes

  • stop re-exporting template types from churro_ocr; import HFChatTemplate and OCRPromptTemplate from churro_ocr.templates instead

s-jse added 30 commits April 12, 2026 16:41
Keep the top-level package focused on the main OCR API surface.

BREAKING CHANGE: import HFChatTemplate and OCRPromptTemplate from churro_ocr.templates instead of churro_ocr.
Standardize the primary install flow on  plus direct  commands.

Also document PDF/all runtime targets and refresh the provider guide for the newer template and prompt helper exports.
Align the docs home, getting-started flow, CLI guide, and task pages around one local HF quick-start.

Split advanced provider customization into a dedicated guide and keep providers focused on runtime and backend choice.
Document the Python 3.12+ version requirement and require docs review plus relevant Pixi checks after changes.
Replace the earlier MinerU integration with the MinerU2.5 two-step pipeline for both Hugging Face and openai-compatible/vLLM providers.

Add the supporting prompts, templates, CLI wiring, evaluation normalization, documentation, and regression coverage for the new backend flow.
s-jse added 28 commits April 13, 2026 18:10
@s-jse
s-jse merged commit 372671e into main Apr 15, 2026
2 checks passed
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.

1 participant