Ai2 Unveils OLMo 2, Truly Open AI Models Rivaling Meta's Llama
Ai2 releases OLMo 2, a new family of fully open-source language models that meet strict open source definitions while competing with Meta's Llama 3.1 in performance.
Nonprofit research organization Ai2 releases OLMo 2, a new family of open language models that meets the Open Source Initiative's strict definition of open source AI. Unlike many supposedly open models, OLMo 2 provides public access to the exact training data, training code, and reproducible recipes required to build the system from scratch.
The OLMo 2 family includes two models with 7 billion and 13 billion parameters, which Ai2 trains using a massive dataset of 5 trillion tokens gathered from high-quality websites, academic papers, and math workbooks. These models handle a variety of text-based tasks such as answering questions, summarizing documents, and writing code.
Ai2 claims that OLMo 2 represents a dramatic improvement over its previous generation and states that the 7B model outperforms Meta's Llama 3.1 8B. Available for download under the permissive Apache 2.0 license, these models allow developers to use and commercially deploy the technology without restrictions.