MiniCPM-OCR

Indra OCR is a CPU-oriented document OCR and parsing tool developed by Datalab. It converts PDFs and images into Markdown, plain text, or JSON containing recognized lines, bounding boxes, layout blocks, labels, reading order, confidences, rotation, and skew information. It can use a digital PDF's text layer where it passes checks and OCR only missing content, or OCR every page. The pipeline uses ONNX Runtime models without PyTorch to detect lines, recognize text, classify blocks such as headings, lists, tables, captions, footnotes, headers, and forms, and order them for reading. Its local viewer displays page overlays, Markdown, source lines, JSON, confidences, and per-stage timings. An optional math mode recognizes mathematical lines and emits inline LaTeX. The Python package and CLI are distributed under Apache 2.0, while the model weights use a modified OpenRAIL-M license with additional commercial-use conditions. The models are downloaded from Hugging Face on first use; the repository reports a model size of about 18 MB. OCR support covers Latin, Cyrillic, Greek, Devanagari, Bengali, and Chinese scripts, while other scripts can be processed through digital PDF text layers. The project documents limitations for handwriting, degraded scans, complex tables, vertical text, and some scripts in scanned images.

OXR (Optical Everything Recognition) is an open-source AI document-parsing model and service from the SoMark team. It converts PDFs and document images into agent-oriented Markdown or JSON, extracting text, tables, formulas, code blocks, chemical equations, chemical structures, and other document elements with layout bounding-box coordinates. Its architecture separates layout analysis from the OXR recognition model, while integrating reading-order handling into recognition. The layout-analysis component is independently replaceable, allowing a custom layout model to be used while retaining OXR's content recognition and reading-order capabilities. During post-training, it uses a recursive self-improvement loop that targets samples the model handles incorrectly or weakly without human annotation. OXR can run locally on Linux with a supported NVIDIA GPU and is distributed with Docker and source-based deployment options. It also provides a web demo and parsing API, with compatibility for SoMark's Python and JavaScript SDKs and CLI tools. The repository is released under the custom OXR Open Source License, based on Apache 2.0 with additional terms.

Umi-OCR is free, open-source, offline OCR software for Windows 7 x64 and later. It recognizes text from screenshots, pasted images, batches of local images, and PDF documents, and can save batch results as TXT, JSONL, Markdown, or CSV files. Its OCR post-processing can merge lines into natural paragraphs, preserve code-block indentation, handle vertical text, and ignore configured rectangular regions such as watermarks or logos. It also includes screenshot, image, and drag-and-drop scanning for QR codes and barcodes, plus QR-code generation. The application can switch between OCR engines through plugins and provides command-line and HTTP API interfaces. The release is distributed as a portable archive or self-extracting archive, requiring no installation or network connection to run. The project is hosted on GitHub and supports multiple interface languages and themes.

doc7 is an open-source document-to-Markdown tool that converts PDFs, Office files, scans, screenshots, charts, formulas, and diagrams into AI-ready Markdown using an OpenAI-compatible multimodal vision model. It renders and understands complete pages rather than relying on character extraction, preserving text, tables, formulas, chart and diagram relationships, image meaning, and visible UI state; the video also describes page-level retry and resume. The model can run locally through LM Studio or Ollama or through a remote endpoint, with no required OCR stack or document-processing service. The same binary provides an interactive CLI, batch processing, model checks, MCP support, a Go SDK, and an asynchronous HTTP service, with installers for macOS, Linux, and Windows.

MiniMind is an open-source tiny large-language-model project and end-to-end training tutorial developed by Jingyao Gong. Its main dense model is approximately 64M parameters, and the repository provides the model architecture, tokenizer, datasets, inference code, and training pipeline. The pipeline covers pretraining, supervised fine-tuning, hand-written LoRA, DPO, PPO, GRPO, CISPO, model distillation, tool calling, adaptive thinking, and agentic reinforcement learning. Core algorithms are implemented directly with native PyTorch rather than relying on high-level abstractions from third-party training libraries. Its agentic reinforcement-learning path performs multi-turn rollouts, executes generated tool calls, appends tool observations to the context, calculates trajectory-level rewards, and updates the policy; rollout can use local PyTorch generation or an SGLang server. MiniMind supports dense and mixture-of-experts variants, single- and multi-GPU training, YaRN-based RoPE length extrapolation, Transformers-format models, and inference through llama.cpp, vLLM, or Ollama. The repository also includes a Streamlit chat interface and a lightweight OpenAI-compatible API server with tool-call and reasoning fields. It is released under the Apache License 2.0.