# OLMo

Modeling, training, eval, and inference code for OLMo

Repository: https://github.com/allenai/OLMo
Canonical: https://ross.abutalabs.com/products/olmo
Homepage: https://allenai.org/olmo
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2025-11-24T19:13:15+00:00
Link (homepage): https://allenai.org/olmo
Link (site_page): https://allenai.org/about
Link (site_page): https://docs.allenai.org/
Link (site_page): https://allenai.org/blog

## Health v2 (maintenance only)
Score: 52/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 53, release rhythm 28, longevity 92
- inputs: {"age_days": 1290, "days_push": 282, "days_rel": 622, "gap_med": 62, "n_releases_24m": 2}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 6648, forks 794 (observed 2026-08-28T04:09:47.267096+00:00)

## What it is
olmOCR is an open toolkit from Ai2 that converts PDFs and image-based documents into clean, reading-order Markdown using a fine-tuned 7B vision-language model, designed for building LLM training datasets. It is part of the broader fully open OLMo ecosystem, which includes open language models, training code, and the Dolma pretraining corpus.

## Use cases
- convert pdfs to markdown for llm training data
- ocr scanned documents with a vision language model
- extract text from multi-column pdfs in reading order
- linearize pdfs into plain text for model pretraining
- process millions of documents cheaply with gpu ocr
- benchmark ocr quality on pdfs with tables and equations
- fine-tune my own document parsing model

## When to choose
- you need high-quality, reading-order text extraction from PDFs or scanned images at scale
- you are building open LLM training corpora and need fully open tooling and models
- you have GPU capacity and want low per-page conversion costs
- you need handling of equations, tables, handwriting, and complex layouts

## When to avoid
- you need OCR on CPU-only machines or without a GPU
- you need a lightweight, non-ML OCR solution for simple text PDFs
- you need a hosted SaaS with a GUI rather than a self-run pipeline
- your documents are natively digital PDFs where simple text extraction suffices

## Facets
- artifact type: library
- maturity: active
- function: ocr, pdf, machine-learning, llm-training, data-generation, image-processing
- domain: artificial-intelligence, large-language-models, pdf, deep-learning
- platform: python, cli
- tags: pdf-to-markdown, vision-language-model, document-parsing, llm-datasets, open-models, allenai, olmo-family, natural-language-processing, data-engineering, linux, gpu, docker

## Member repositories
- allenai/OLMo (main) score 52
- allenai/olmocr (infra) score 69
- allenai/open-instruct (backend) score 88
- allenai/dolma (infra) score 75
- allenai/OLMo-core (backend) score 91
- allenai/OLMoE (plugin) score 43

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:47.267096+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 2026-08-29T17:00:07.492402+00:00, confidence not recorded.
  - readme: https://github.com/allenai/OLMo (fetched 2026-08-28T04:09:47.267096+00:00, sha 9f0071ee37a8)
  - homepage: https://allenai.org/olmo (fetched 2026-08-29T07:58:48.351588+00:00, sha c28bb7b65913)
  - site_page: https://allenai.org/about (fetched 2026-08-29T07:58:48.360961+00:00, sha 3ba69c8db9fc)
  - site_page: https://docs.allenai.org/ (fetched 2026-08-29T07:58:48.363002+00:00, sha 82d8481d566c)
  - registry_pypi: https://pypi.org/pypi/olmocr/json (fetched 2026-08-29T07:58:48.366199+00:00, sha 40e77c548036)
  - site_page: https://allenai.org/blog (fetched 2026-08-29T07:58:48.364601+00:00, sha 94d5da0018fc)
- Data as of 2026-08-30T08:39:29.467469+00:00.