OLMo 2 - a family of fully-open language models
10 months ago
- #AI
- #Language Models
- #Open Source
- OLMo 2 is a family of fully-open language models with open data, code, and evaluations.
- OLMo 2 32B is the largest model, outperforming GPT3.5-Turbo and GPT-4o mini on benchmarks.
- OLMo 2 7B and 13B are competitive with Meta and Mistral models on English benchmarks.
- OLMo 2 1B is the smallest model, enabling rapid iteration for researchers.
- OLMo is Ai2’s first Open Language Model framework, designed for open AI research.
- Ai2 emphasizes openness with fully accessible data, models, and code.
- OLMo 2 models and training data are freely available for scientific research.
- High-performance training code for OLMo 2 is open for extension and use.
- Evaluation code and data are openly available for reproduction and scrutiny.
- Technical details and updates on OLMo releases are provided in release notes.