We present OLMo 2, the next generation of our fully open language models. OLMo 2 includes dense autoregressive models with improved architecture and training recipe, pretraining data mixtures, and instruction tuning recipes. Our modified model architecture and training recipe achieve both better training stability and improved per-token efficiency. Our updated pretraining data mixture introduces a new, specialized data mix called Dolmino Mix 1124, which significantly improves model capabilities across many downstream task benchmarks when introduced via late-stage curriculum training (i.e. specialized data during the annealing phase of pretraining). Finally, we incorporate best practices from T\"ulu 3 to develop OLMo 2-Instruct, focusing on permissive data and extending our final-stage reinforcement learning with verifiable rewards (RLVR). Our OLMo 2 base models sit at the Pareto frontier of performance to compute, often matching or outperforming open-weight only models like Llama 3.1 and Qwen 2.5 while using fewer FLOPs and with fully transparent training data, code, and recipe. Our fully open OLMo 2-Instruct models are competitive with or surpassing open-weight only models of comparable size, including Qwen 2.5, Llama 3.1 and Gemma 2. We release all OLMo 2 artifacts openly -- models at 7B and 13B scales, both pretrained and post-trained, including their full training data, training code and recipes, training logs and thousands of intermediate checkpoints. The final instruction model is available on the Ai2 Playground as a free research demo.

This study introduces AGGA, a dataset comprising 80 academic guidelines for the use of Generative AIs (GAIs) and Large Language Models (LLMs) in academic settings, meticulously collected from official university websites. The dataset contains 188,674 words and serves as a valuable resource for natural language processing tasks commonly applied in requirements engineering, such as model synthesis, abstraction identification, and document structure assessment. Additionally, AGGA can be further annotated to function as a benchmark for various tasks, including ambiguity detection, requirements categorization, and the identification of equivalent requirements. Our methodologically rigorous approach ensured a thorough examination, with a selection of universities that represent a diverse range of global institutions, including top-ranked universities across six continents.

LLMs process text as sequences of tokens that roughly correspond to words, where less common words are represented by multiple tokens. However, individual tokens are often semantically unrelated to the meanings of the words/concepts they comprise. For example, Llama-2-7b's tokenizer splits the word "northeastern" into the tokens ['_n', 'ort', 'he', 'astern'], none of which correspond to semantically meaningful units like "north" or "east." Similarly, the overall meanings of named entities like "Neil Young" and multi-word expressions like "break a leg" cannot be directly inferred from their constituent tokens. Mechanistically, how do LLMs convert such arbitrary groups of tokens into useful higher-level representations? In this work, we find that last token representations of named entities and multi-token words exhibit a pronounced "erasure" effect, where information about previous and current tokens is rapidly forgotten in early layers. Using this observation, we propose a method to "read out" the implicit vocabulary of an autoregressive LLM by examining differences in token representations across layers, and present results of this method for Llama-2-7b and Llama-3-8B. To our knowledge, this is the first attempt to probe the implicit vocabulary of an LLM.

The integration of Generative Artificial Intelligence (GAI) and Large Language Models (LLMs) in academia has spurred a global discourse on their potential pedagogical benefits and ethical considerations. Positive reactions highlight some potential, such as collaborative creativity, increased access to education, and empowerment of trainers and trainees. However, negative reactions raise concerns about ethical complexities, balancing innovation and academic integrity, unequal access, and misinformation risks. Through a systematic survey and text-mining-based analysis of global and national directives, insights from independent research, and eighty university-level guidelines, this study provides a nuanced understanding of the opportunities and challenges posed by GAI and LLMs in education. It emphasizes the importance of balanced approaches that harness the benefits of these technologies while addressing ethical considerations and ensuring equitable access and educational outcomes. The paper concludes with recommendations for fostering responsible innovation and ethical practices to guide the integration of GAI and LLMs in academia.

We investigate the rate at which algorithms for pre-training language models have improved since the advent of deep learning. Using a dataset of over 200 language model evaluations on Wikitext and Penn Treebank spanning 2012-2023, we find that the compute required to reach a set performance threshold has halved approximately every 8 months, with a 95% confidence interval of around 5 to 14 months, substantially faster than hardware gains per Moore's Law. We estimate augmented scaling laws, which enable us to quantify algorithmic progress and determine the relative contributions of scaling models versus innovations in training algorithms. Despite the rapid pace of algorithmic progress and the development of new architectures such as the transformer, our analysis reveals that the increase in compute made an even larger contribution to overall performance improvements over this time period. Though limited by noisy benchmark data, our analysis quantifies the rapid progress in language modeling, shedding light on the relative contributions from compute and algorithms.

Language models (LMs) have become ubiquitous in both NLP research and in commercial product offerings. As their commercial importance has surged, the most powerful models have become closed off, gated behind proprietary interfaces, with important details of their training data, architectures, and development undisclosed. Given the importance of these details in scientifically studying these models, including their biases and potential risks, we believe it is essential for the research community to have access to powerful, truly open LMs. To this end, this technical report details the first release of OLMo, a state-of-the-art, truly Open Language Model and its framework to build and study the science of language modeling. Unlike most prior efforts that have only released model weights and inference code, we release OLMo and the whole framework, including training data and training and evaluation code. We hope this release will empower and strengthen the open research community and inspire a new wave of innovation.

Language models have become a critical technology to tackling a wide range of natural language processing tasks, yet many details about how the best-performing language models were developed are not reported. In particular, information about their pretraining corpora is seldom discussed: commercial language models rarely provide any information about their data; even open models rarely release datasets they are trained on, or an exact recipe to reproduce them. As a result, it is challenging to conduct certain threads of language modeling research, such as understanding how training data impacts model capabilities and shapes their limitations. To facilitate open research on language model pretraining, we release Dolma, a three trillion tokens English corpus, built from a diverse mixture of web content, scientific papers, code, public-domain books, social media, and encyclopedic materials. In addition, we open source our data curation toolkit to enable further experimentation and reproduction of our work. In this report, we document Dolma, including its design principles, details about its construction, and a summary of its contents. We interleave this report with analyses and experimental results from training language models on intermediate states of Dolma to share what we have learned about important data curation practices, including the role of content or quality filters, deduplication, and multi-source mixing. Dolma has been used to train OLMo, a state-of-the-art, open language model and framework designed to build and study the science of language modeling.

A prerequisite for safe autonomy-in-the-wild is safe testing-in-the-wild. Yet real-world autonomous tests face several unique safety challenges, both due to the possibility of causing harm during a test, as well as the risk of encountering new unsafe agent behavior through interactions with real-world and potentially malicious actors. We propose a framework for conducting safe autonomous agent tests on the open internet: agent actions are audited by a context-sensitive monitor that enforces a stringent safety boundary to stop an unsafe test, with suspect behavior ranked and logged to be examined by humans. We design a basic safety monitor (AgentMonitor) that is flexible enough to monitor existing LLM agents, and, using an adversarial simulated agent, we measure its ability to identify and stop unsafe situations. Then we apply the AgentMonitor on a battery of real-world tests of AutoGPT, and we identify several limitations and challenges that will face the creation of safe in-the-wild tests as autonomous agents grow more capable.

Purpose: Demonstrating and assessing self-supervised machine learning fitting of the VERDICT (Vascular, Extracellular and Restricted DIffusion for Cytometry in Tumours) model for prostate. Methods: We derive a self-supervised neural network for fitting VERDICT (ssVERDICT) that estimates parameter maps without training data. We compare the performance of ssVERDICT to two established baseline methods for fitting diffusion MRI models: conventional nonlinear least squares (NLLS) and supervised deep learning. We do this quantitatively on simulated data, by comparing the Pearson's correlation coefficient, mean-squared error (MSE), bias, and variance with respect to the simulated ground truth. We also calculate in vivo parameter maps on a cohort of 20 prostate cancer patients and compare the methods' performance in discriminating benign from cancerous tissue via Wilcoxon's signed-rank test. Results: In simulations, ssVERDICT outperforms the baseline methods (NLLS and supervised DL) in estimating all the parameters from the VERDICT prostate model in terms of Pearson's correlation coefficient, bias, and MSE. In vivo, ssVERDICT shows stronger lesion conspicuity across all parameter maps, and improves discrimination between benign and cancerous tissue over the baseline methods. Conclusion: ssVERDICT significantly outperforms state-of-the-art methods for VERDICT model fitting, and shows for the first time, fitting of a complex three-compartment biophysical model with machine learning without the requirement of explicit training labels.

Attenuation correction is an essential requirement of positron emission tomography (PET) image reconstruction to allow for accurate quantification. However, attenuation correction is particularly challenging for PET-MRI as neither PET nor magnetic resonance imaging (MRI) can directly image tissue attenuation properties. MRI-based computed tomography (CT) synthesis has been proposed as an alternative to physics based and segmentation-based approaches that assign a population-based tissue density value in order to generate an attenuation map. We propose a novel deep fully convolutional neural network that generates synthetic CTs in a recursive manner by gradually reducing the residuals of the previous network, increasing the overall accuracy and generalisability, while keeping the number of trainable parameters within reasonable limits. The model is trained on a database of 20 pre-acquired MRI/CT pairs and a four-fold random bootstrapped validation with a 80:20 split is performed. Quantitative results show that the proposed framework outperforms a state-of-the-art atlas-based approach decreasing the Mean Absolute Error (MAE) from 131HU to 68HU for the synthetic CTs and reducing the PET reconstruction error from 14.3% to 7.2%.