The Allen Institute for AI (AI2) has launched *OLMo 2, the second generation of its **Open Language Model (OLMo)* series. Unlike many so-called “open” models, OLMo 2 aligns with the Open Source Initiative’s standards, offering full transparency in its development process. This includes publicly available training data, open-source code, reproducible training methods, and intermediate checkpoints.
OLMo 2 consists of two models: *OLMo 7B* (7 billion parameters) and *OLMo 13B* (13 billion parameters). Parameters indicate a model’s problem-solving capacity, and larger models generally excel. AI2 trained these models on 5 trillion tokens from high-quality sources, such as academic papers, Q&A forums, and curated websites.
Performance-wise, OLMo 2 surpasses similar open models, with the 7B model outperforming Meta’s Llama 3.1 (8B). Licensed under Apache 2.0, OLMo 2 is commercially viable, supporting diverse tasks like text summarization, coding, and question-answering.
Despite debates about the risks of open models, AI2 highlights their role in advancing ethical AI and democratizing access. OLMo 2 models are freely available on AI2’s website, reinforcing the organization’s commitment to open-source innovation.
 has launched *OLMo 2, the second generation of its **Open Language Model (OLMo)...){kind=link}