olmocr Releases

v0.4.3 adds support for finetuning with new configurations, improves the priority scheduler, and introduces benchmarking enhancements including support for Cirrascale and DeepInfra providers.

Oct 31, 2025

This release primarily consists of updates to tests and fixes for the continuous integration pipeline to ensure smooth deployment.

Oct 22, 2025

The release involves rebuilding the Docker image using a new runner to support larger jobs and improve build processes.

Oct 7, 2025

A minor maintenance release focused on code formatting and version bumps to maintain code quality standards.

Oct 6, 2025

This release fixes a release script issue where files were being added by default, ensuring cleaner and more controlled releases.

Sep 29, 2025

This version manages library dependencies (VLLM and Transformers) and adds baseline tests and benchmarking scripts to the project.

Aug 31, 2025

v0.3.3 includes pipeline fixes related to threading configuration (OMP_NUM_THREADS) and attempts to keep the queue loaded, alongside general linting.

Aug 15, 2025

This release focuses on optimizing resource usage by improving the pipeline to limit CPU consumption during page rendering, accompanied by general code cleanup.

Aug 14, 2025

Updates include better queue management, a bump in VLLM version, and improvements in tracking semaphore releases for larger jobs, along with internal AI2 budgeting updates.

Aug 14, 2025

A version focused on fixing critical document processing issues, specifically addressing problems with blank documents, and merging a significant pull request.

Aug 13, 2025

Introduces cumulative rotation support for document processing and saves extra metadata intended for future fine-tuning, alongside various fixes and formatting updates.

Aug 4, 2025

Focuses on memory optimization by lowering usage checks, fixing rotation logic, and improving error messages, alongside the addition of new tests.

Aug 4, 2025

A minor release involving pipeline cleanup, fixes for default model length handling, and improvements to table formatting (bolds) within documents.

Jul 23, 2025

A major version release introducing a new trainer implementation, VLLM integration, compression and calibration features, and a new default model with various configuration updates.

Jul 23, 2025

Updates documentation and adds helper scripts, specifically updating README files and Pareto plots for better benchmarking visualization.

Jun 23, 2025

A brief release consisting of a final cleanup commit to prepare for the next version iteration.

Jun 17, 2025

Optimizes CI/CD processes and increases worker limits to better suit high-performance H100 GPU configurations.

Jun 17, 2025

This release focuses on infrastructure stability and debugging improvements for the Beaker backend.

Jun 17, 2025

A major infrastructure update migrating the inference backend from SGLang to vLLM and updating the Docker environment.

Jun 17, 2025

Performance optimizations and benchmarking tooling improvements focused on model behavior and retry logic.

May 30, 2025