# aiptimizer/TurboOCR

TurboOCR, >200 img/s OmnidocBench. TensorRT FP16, PP-OCRv6, HTTP + gRPC

Repository: https://github.com/aiptimizer/TurboOCR
Canonical: https://ross.abutalabs.com/products/turboocr
Homepage: https://turboocr.com
Language: C++
License: MIT
License Family: permissive
Topics: document-parsing, fastapi, fp16, grpc, ocr, paddleocr, pdf-extraction, rag, tensorrt, text-detection, text-recognition, gpu-ocr, document-ai, easyocr, inference-server, nvidia, qwen-vl
Last push: 2026-08-31T22:28:32+00:00

## Health v2 (maintenance only)
Score: 82/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 100, release rhythm 98, longevity 11
- inputs: {"age_days": 166, "days_push": 2, "days_rel": 15, "gap_med": 3.5, "n_releases_24m": 19}
- flags: young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1043, forks 100 (observed 2026-09-01T02:13:58.538529+00:00)

## What it is
TurboOCR is an extremely fast GPU-accelerated document parser written in C++ that combines OCR, layout analysis, table extraction, and formula recognition to output Markdown. It uses TensorRT FP16 inference with PP-OCRv6 models and serves results over HTTP and gRPC APIs.

## Use cases
- parse scanned documents to markdown at high throughput
- run OCR on a GPU server with an HTTP or gRPC API
- extract text, tables, and formulas from PDFs for RAG pipelines
- batch process millions of document images quickly
- self-host a fast OCR inference server on NVIDIA GPUs
- convert document images to structured markdown

## When to choose
- you need maximum OCR throughput (hundreds of images per second) on NVIDIA hardware
- you want a self-hosted document parsing service with HTTP and gRPC APIs
- you need OCR plus layout, table, and formula extraction feeding a RAG pipeline
- you can deploy on Linux with CUDA/TensorRT

## When to avoid
- you need CPU-only or non-NVIDIA GPU support (Metal/OpenVINO/ROCm backends are not shipped yet)
- you need a lightweight embedded OCR library rather than a server
- you require Windows or macOS deployment
- you need a pure Python OCR solution

## Facets
- artifact type: service
- maturity: active
- function: ocr, pdf, machine-learning, llm-inference, http-server, rpc, rag
- domain: computer-vision, artificial-intelligence, pdf, gpu-computing
- platform: cpp, self-hosted
- tags: tensorrt, pp-ocrv6, paddleocr, fp16, grpc, fastapi, document-parsing, inference-server, nvidia, text-recognition, layout-analysis, table-extraction, formula-recognition, markdown-output, documents, linux, docker, gpu

## Member repositories
- aiptimizer/TurboOCR (main) score 82

## Provenance
- Observed fields: from GitHub, fetched 2026-09-01T02:13:58.538529+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T07:08:02.117614+00:00, confidence not recorded.
  - readme: https://github.com/aiptimizer/TurboOCR (fetched 2026-09-01T02:13:58.538529+00:00, sha 24ba03c1f995)
  - homepage: https://turboocr.com (fetched 2026-08-29T13:08:14.804183+00:00, sha d065282287a5)
- Data as of 2026-08-30T08:39:29.467469+00:00.