The latest release of allenai/olmo-core is v3.0.0. This repository provides PyTorch building blocks for the OLMo model family — that is, reusable pieces needed for modeling and training — as a Python library. The README explains the official pre-training script for the public models and how to run inference from checkpoints using Hugging Face Transformers. Installation is available from the source code or from PyPI.
The center of this repository is not a single model but the publication of a training pipeline split into several pieces. To retrain a public language model, or to run inference from a checkpoint, you have to decide how the model architecture and the training steps should be combined and in what order; this library has organized that part into piece-by-piece code. The license is Apache-2.0, the primary language is Python, and it has 1,727 GitHub stars. The repository was created on March 23, 2024.
However, the collected material contains no changelog for v3.0.0, so what this release changed cannot be confirmed. The technology, quality, and novelty scores are empty, and only the momentum score reaches 2.2 on a 10-point scale. So the only thing that can be judged right now is the fact that the published training and inference code can be borrowed and used as-is. What v3.0.0 improved has to be seen by reading the release notes directly.