艾伦人工智能研究所发布首个真正开放的大语言模型 OLMo
西雅图——艾伦人工智能研究所（AI2）近日推出了一个突破性的大语言模型（LLM）OLMo，该模型完全开源，为其内部运作提供了前所未有的透明度。
OLMo 被 AI2 称为“第一个真正开放的 LLM”，因为它提供的不仅仅是模型代码和权重。该研究所还发布了用于开发 OLMo 的完整训练数据、训练代码、评估基准和工具包。
这种开放性标志着开源人工智能生态系统的一个重大进步。研究人员和开发人员现在可以完全访问 OLMo 的所有组件，从而能够对其进行定制、改进和探索其潜力。
OLMo 是一个多模态模型，这意味着它可以执行广泛的任务，包括自然语言处理、计算机视觉和语音识别。它在各种基准测试中表现出色，包括 GLUE、SuperGLUE 和 OpenBookQA。
AI2 首席执行官 Oren Etzioni 表示：“OLMo 的发布是开源人工智能的一个转折点。它将使研究人员和开发人员能够以前所未有的方式探索和利用 LLM 的力量。”
OLMo 的开源性质预计将加速 LLM 的研究和开发。研究人员现在可以轻松地复制和改进 AI2 的工作，而开发人员可以将 OLMo 集成到他们的应用程序和产品中。
除了其开放性之外，OLMo 还具有以下特点：
* 大规模：OLMo 在一个包含 1.2 万亿个单词的语料库上进行训练，使其成为最大的开源 LLM 之一。
* 高效：OLMo 经过优化，可以在各种硬件上高效运行，包括云计算平台和边缘设备。
* 可扩展：OLMo 的模块化设计允许轻松扩展和定制，以满足特定的需求。
OLMo 的发布受到了人工智能社区的广泛赞誉。卡内基梅隆大学机器学习教授 Tom Mitchell 表示：“OLMo 的开放性是一个巨大的进步。它将使研究人员和开发人员能够以前所未有的方式推进 LLM 的研究和应用。”
OLMo 可从 AI2 的网站免费下载。
英语如下：
**Headline:** Allen AI Releases First Truly Open-Source Large Language Model
**Keywords:** Open-source model, transparency, comprehensive toolkit
**Body:**
The Allen Institute for Artificial Intelligence (AI2) has released OLMo, thefirst truly open large language model (LLM), offering unprecedented transparency into its inner workings.
OLMo is what AI2 calls “the first truly open LLM” because it provides more than just the model code and weights. The institute has also released the complete training data, training code, evaluation benchmarks, and toolkits used to develop OLMo.
This openness marks a major advancement for the open-source AI ecosystem. Researchers and developers now have full access to all of OLMo’s components, enabling them to customize it, improve it, and explore its potential.
OLMo is a multimodal model, meaning it can perform a wide range of tasks, including natural language processing, computer vision, and speech recognition. It has demonstrated strong performance on a variety of benchmarks, including GLUE, SuperGLUE, and OpenBookQA.
“The release of OLMo is a watershed moment for open-source AI,” said Oren Etzioni,CEO of AI2. “It will empower researchers and developers to explore and harness the power of LLMs in ways that were previously impossible.”
The open-source nature of OLMo is expected to accelerate research and development of LLMs. Researchers can now easily replicate and improve upon AI2’s work, while developers can integrate OLMo into their applications and products.
In addition to its openness, OLMo features:
* **Scale:** OLMo was trained on a dataset of 1.2 trillion words, making it one of the largest open-source LLMs available.
* **Efficiency:** OLMo has been optimized to run efficiently on a variety of hardware, including cloud computing platforms and edge devices.
* **Extensibility:** OLMo’s modular design allows it to be easily extended and customized to meet specific needs.
The release of OLMo has been met with widespread praise from the AI community. “The openness of OLMo is a huge step forward,” said Tom Mitchell, professor of machine learning at Carnegie Mellon University. “It will enable researchers and developers to advance LLM research and applications in ways that were previously impossible.”
OLMo is available for free download from AI2’s website.
【来源】https://www.maginative.com/article/ai2-unveils-olmo-the-first-truly-open-large-language-model/
Views: 1