Large language models, which are often trained for hundreds of thousands of compute days, have shown remarkable capabilities for zero-and few-shot learning. Given their computational cost, these models are difficult to replicate without significant capital. For the few that are available through APIs, no access is granted to the full model weights, making them difficult to study. We present Open Pre-trained Transformers (OPT), a suite of decoder-only pre-trained transformers ranging from 125M to 175B parameters, which we aim to fully and responsibly share with interested researchers. We show that OPT-175B is comparable to GPT-3, 1 while requiring only 1/7th the carbon footprint to develop. We are also releasing our logbook detailing the infrastructure challenges we faced, along with code for experimenting with all of the released models. * Equal contribution. † Work done while at Meta AI. 1 Following Brown et al. (2020), we use GPT-3 to refer to both the 175B model and the smaller scale models as well.2 Exceptions include work by EleutherAI, who released dense models up to 20B in size (Black et al., 2022), Salesforce (Nijkamp et al., 2022), and Meta AI, who released dense models up to 13B and sparse models up to 1. 1T (Artetxe et al., 2021). There is also ongoing work from the BigScience workshop (https://bigscience. huggingface.co/), which aims to open source very large multilingual language models and datasets.

Mixture of Experts layers (MoEs) enable efficient scaling of language models through conditional computation. This paper presents a detailed empirical study of how autoregressive MoE language models scale in comparison with dense models in a wide range of settings: in-and out-of-domain language modeling, zero-and few-shot priming, and full finetuning. With the exception of fine-tuning, we find MoEs to be substantially more compute efficient. At more modest training budgets, MoEs can match the performance of dense models using ∼4 times less compute. This gap narrows at scale, but our largest MoE model (1.1T parameters) consistently outperforms a compute-equivalent dense model (6.7B parameters). Overall, this performance gap varies greatly across tasks and domains, suggesting that MoE and dense models generalize differently in ways that are worthy of future study. We make our code and models publicly available for research use. 1 * Equal contribution. Authors listed alphabetically.

No abstract

No abstract

No abstract