Pretraining has been shown to scale well with compute, data size and data diversity. Multitask learning trains on a mixture of supervised datasets and produces improved performance compared to self-supervised pretraining. Until now, massively multitask learning required simultaneous access to all datasets in the mixture and heavy compute resources that are only available to well-resourced teams. In this paper, we propose ColD Fusion, a method that provides the benefits of multitask learning but leverages distributed computation and requires limited communication and no sharing of data. Consequentially, ColD Fusion can create a synergistic loop, where finetuned models can be recycled to continually improve the pretrained model they are based on. We show that ColD Fusion yields comparable benefits to multitask pretraining by producing a model that (a) attains strong performance on all of the datasets it was multitask trained on and (b) is a better starting point for finetuning on unseen datasets. We find ColD Fusion outperforms RoBERTa and even previous multitask models. Specifically, when training and testing on 35 diverse datasets, ColD Fusion-based model outperforms RoBERTa by 2.45 points in average without any changes to the architecture.

Previous studies observed that finetuned models may be better base models than the vanilla pretrained model. Such a model, finetuned on some source dataset, may provide a better starting point for a new finetuning process on a desired target dataset. Here, we perform a systematic analysis of this intertraining scheme, over a wide range of English classification tasks. Surprisingly, our analysis suggests that the potential intertraining gain can be analyzed independently for the target dataset under consideration, and for a base model being considered as a starting point. This is in contrast to current perception that the alignment between the target dataset and the source dataset used to generate the base model is a major factor in determining intertraining success. We analyze different aspects that contribute to each. Furthermore, we leverage our analysis to propose a practical and efficient approach to determine if and how to select a base model in real-world settings. Last, we release an updating ranking of best models in the HuggingFace hub per architecture https://ibm.github.io/model-recycling/.

人工智能的最新趋势是将验证的模型用于语言和视觉任务，这些模型已经实现了非凡的表现，但也令人困惑。因此，以各种方式探索这些模型的能力对该领域至关重要。在本文中，我们探讨了模型的可靠性，在其中我们将可靠的模型定义为一个不仅可以实现强大的预测性能，而且在许多涉及不确定性（例如选择性预测，开放式设置识别）的决策任务上，在许多决策任务上表现出色，而且表现良好。强大的概括（例如，准确性和适当的评分规则，例如在分布数据集中和分发数据集上的对数可能性）和适应性（例如，主动学习，几乎没有射击不确定性）。我们设计了40个数据集的10种任务类型，以评估视觉和语言域上可靠性的不同方面。为了提高可靠性，我们分别开发了VIT-PLEX和T5-PLEX，分别针对视觉和语言方式扩展了大型模型。 PLEX极大地改善了跨可靠性任务的最先进，并简化了传统协议，因为它可以改善开箱即用的性能，并且不需要设计分数或为每个任务调整模型。我们演示了高达1B参数的模型尺寸的缩放效果，并预处理数据集大小最多4B示例。我们还展示了PLEX在具有挑战性的任务上的功能，包括零射门的开放式识别，主动学习和对话语言理解中的不确定性。

现代的深度学习系统越来越多地部署在个性化和联合学习等情况下，需要支持i）学习少量数据，ii）沟通有效的分布式培训协议。在这项工作中，我们开发了胶片转移（FIT），该胶片在图像分类设置中满足了这些要求。 FIT使用自动配置的幼稚贝叶斯分类器在固定的主链上，该主链在大型图像数据集上仔细考虑。参数有效膜层用于调节主链，从而为下游任务塑造表示形式。该网络通过情节微调协议进行培训。该方法是参数效率的，这对于能够实现几次学习，廉价的个性化模型更新以及沟通有效的联合学习的关键。我们尝试适合各种下游数据集，并表明它可以比最先进的大型转移（位）算法在低射击和挑战性的VTAB-1K基准上获得更好的分类准确性，该算法的精度少于1％可更新参数。最后，我们证明了在分布式低弹药应用中拟合的参数效率，包括模型个性化和联合学习，其中模型更新大小是重要的性能指标。

对于大多数自然语言处理任务，主要的实践是使用较小的下游数据集对大型预验证变压器模型（例如BERT）。尽管这种方法取得了成功，但尚不清楚这些收益在多大程度上归因于用于预处理而不是训练预处理的目标本身所采用的大量背景语料库。本文介绍了一项大规模的自我预测研究，其中相同的（下游）训练数据都用于预训练和填充。在解决Electra和Roberta型号以及10个不同下游数据集的实验中，我们观察到在BookWiki语料库上进行自我预测的竞争对手标准预告片（尽管使用了$ 10 \ times $ $ -500 \ times $ -500 \ times $少的数据），在7美元上以7美元的价格优于$ 7 $和$ 5 $数据集。令人惊讶的是，这些特定于任务的预预性模型通常在其他任务（包括胶水基准）上表现良好。我们的结果表明，在许多情况下，可归因于预处理的绩效收益主要是由预处理目标本身驱动的，并不总是归因于大规模数据集的合并。考虑到网络规模预处理数据中对知识产权和进攻内容的担忧，这些发现尤其重要。

现代深度学习应用程序需要越来越多地计算培训最先进的模型。为了解决这一需求，大型企业和机构使用专用的高性能计算集群，其建筑和维护既昂贵又远远超出大多数组织的预算。结果，一些研究方向成为几个大型工业甚至更少的学术作用者的独家领域。为了减轻这种差异，较小的团体可以汇集他们的计算资源并运行有利于所有参与者的协作实验。这种范式称为网格或志愿者计算，在众多科学领域看到了成功的应用。然而，由于高延迟，不对称带宽以及志愿者计算独特的几个挑战，使用这种用于机器学习的方法是困难的。在这项工作中，我们仔细分析了这些约束，并提出了一种专门用于协作培训的新型算法框架。我们展示了我们在现实条件下的SWAV和Albert预先预价的方法的有效性，并在成本的一小部分中实现了与传统设置相当的性能。最后，我们提供了一份成功的协作语言模型预先追溯的详细报告，有40名参与者。

Transfer learning, where a model is first pre-trained on a data-rich task before being finetuned on a downstream task, has emerged as a powerful technique in natural language processing (NLP). The effectiveness of transfer learning has given rise to a diversity of approaches, methodology, and practice. In this paper, we explore the landscape of transfer learning techniques for NLP by introducing a unified framework that converts all text-based language problems into a text-to-text format. Our systematic study compares pre-training objectives, architectures, unlabeled data sets, transfer approaches, and other factors on dozens of language understanding tasks. By combining the insights from our exploration with scale and our new "Colossal Clean Crawled Corpus", we achieve state-of-the-art results on many benchmarks covering summarization, question answering, text classification, and more. To facilitate future work on transfer learning for NLP, we release our data set, pre-trained models, and code.

少量学习时，基于及时的方法很强劲。然而，Perez等人。 （2021年）最近对他们的表现产生了疑问，因为它们难以在“真实”的几次拍摄设置中获得良好的结果，其中提示和超级参数无法在DEV集上调整。鉴于此，我们对PET进行了广泛的研究，该方法将文本指令与基于示例的FENETUNING结合起来。我们表明，如果正确配置，宠物在真正的几次拍摄设置中强烈执行，即，没有开发装置。这对这种强大的表现至关重要是宠物智能处理多个提示的能力。然后，我们通过在RAFT上运行PET来将我们的调查结果置于真实世界的测试中，直接从现实的NLP应用程序采取的任务的基准，没有标记的开发或测试集。宠物在筏上实现了新的艺术状态，并且在11个任务中靠近非专家人类进行了近距离进行。这些结果表明，基于及时的学习者像宠物Excel这样的真正的几次拍摄学习和支持我们的信念，即从指示中学习的信念将在人类少量学习能力的路径上发挥重要作用。

在基于典型的深度神经网络训练期间，所有模型的参数都在每次迭代时更新。最近的工作表明，在训练期间只能更新模型参数的小型子集，这可以减轻存储和通信要求。在本文中，我们表明，可以在模型的参数上诱导一个固定的稀疏掩码，该屏蔽选择要在许多迭代中更新的子集。我们的方法用最大的Fisher信息构造出k $参数的掩码，作为一个简单的近似，与手头的任务最重要的近似值。在参数高效转移学习和分布式培训的实验中，我们表明我们的方法与其他方法的性能相匹配或超出稀疏更新的其他方法的性能，同时在内存使用和通信成本方面更有效。我们公开发布我们的代码，以促进我们的方法的进一步应用。

Fine-tuning pre-trained language models has become the prevalent paradigm for building downstream NLP models. Oftentimes fine-tuned models are readily available but their training data is not, due to data privacy or intellectual property concerns. This creates a barrier to fusing knowledge across individual models to yield a better single model. In this paper, we study the problem of merging individual models built on different training data sets to obtain a single model that performs well both across all data set domains and can generalize on out-of-domain data. We propose a dataless knowledge fusion method that merges models in their parameter space, guided by weights that minimize prediction differences between the merged model and the individual models. Over a battery of evaluation settings, we show that the proposed method significantly outperforms baselines such as Fisher-weighted averaging or model ensembling. Further, we find that our method is a promising alternative to multi-task learning that can preserve or sometimes improve over the individual models without access to the training data. Finally, model merging is more efficient than training a multi-task model, thus making it applicable to a wider set of scenarios.

随着时间的推移，保持语言技术的性能是很好的实际兴趣。在这里，我们在涉及系统性能的时间效果，建立更细微的术语，用于讨论该主题和适当的实验设计，以支持有关观察到的现象的效果的调查。我们提出了一系列与由大型神经预磨削表示的系统进行用于英语的系统，证明{\ EM时间模型恶化}并不像较大的关注，有一些模型实际上在从稍后的时间段绘制的数据上进行测试时改善。然而，{\ EM时间域自适应}是有益的，当系统在时间上训练时，可以更好地进行给定时间段的性能更好。我们的实验表明，在预磨削表示时，时间模型劣化和时间域适应之间的区别变得突出。最后，我们研究了两种方法对时间域适应的效果，没有人为的新数据的注释，自我标签证明是优于持续的预训练。值得注意的是，对于命名实体识别，自我标签导致比人类注释更好的时间适应。

转移学习提供了一种在学习另一个任务时从一个任务中利用知识的方式。执行转移学习通常涉及通过训练数据集上的梯度下降来迭代地更新模型的参数。在本文中，我们介绍了一种基本上不同的方法，用于将知识转移到跨模型，这些方法将多个模型“合并”成一个。我们的方法有效地涉及计算模型参数的加权平均值。我们表明，该平均值相当于从模型权重的后部的大致抽样。在某些情况下使用各向同性高斯近似时，我们还通过Fisher信息近似于精确矩阵来证明优势。总之，我们的方法使得与基于标准梯度的培训相比，可以以极低的计算成本将多种模型中的“知识”组合。我们展示了模型合并在中间任务培训和域适应问题上实现了基于梯度下降的转移学习的可比性。我们还表明，我们的合并程序使得可以以先前未开发的方式结合模型。为了测量我们方法的稳健性，我们对我们算法的设计进行了广泛的消融。

Training large, deep neural networks to convergence can be prohibitively expensive. As a result, often only a small selection of popular, dense models are reused across different contexts and tasks. Increasingly, sparsely activated models, which seek to decouple model size from computation costs, are becoming an attractive alternative to dense models. Although more efficient in terms of quality and computation cost, sparse models remain data-hungry and costly to train from scratch in the large scale regime. In this work, we propose sparse upcycling -- a simple way to reuse sunk training costs by initializing a sparsely activated Mixture-of-Experts model from a dense checkpoint. We show that sparsely upcycled T5 Base, Large, and XL language models and Vision Transformer Base and Large models, respectively, significantly outperform their dense counterparts on SuperGLUE and ImageNet, using only ~50% of the initial dense pretraining sunk cost. The upcycled models also outperform sparse models trained from scratch on 100% of the initial dense pretraining computation budget.

大型审慎的语言模型（PLM）通常是通过微调或提示来适应域或任务的。填充需要修改所有参数，并具有足够的数据以避免过度拟合，同时提示不需要培训，也不需要示例，而是限制性能。取而代之的是，我们通过学习学习一般和适应性PLM之间的差异来为数据和参数有效适应。通过我们提出的动态低级别重新聚体和学识渊博的体系结构控制器，通过模型权重和子层结构来表示这种差异。实验对话完成，低资源抽象摘要以及多域语言建模的实验显示了通过域自适应预处理进行适应时间和性能的改善。消融表明我们的任务自适应重新聚体化（TARP）和模型搜索（TAMS）组件分别改进了其他参数效率转移（如适配器和结构学习方法），例如学习的稀疏。

Automated Program Repair (APR) is defined as the process of fixing a bug/defect in the source code, by an automated tool. APR tools have recently experienced promising results by leveraging state-of-the-art Neural Language Processing (NLP) techniques. APR tools such as TFix and CodeXGLUE combine text-to-text transformers with software-specific techniques are outperforming alternatives, these days. However, in most APR studies the train and test sets are chosen from the same set of projects. In reality, however, APR models are meant to be generalizable to new and different projects. Therefore, there is a potential threat that reported APR models with high effectiveness perform poorly when the characteristics of the new project or its bugs are different than the training set's(Domain Shift). In this study, we first define and measure the domain shift problem in automated program repair. Then, we then propose a domain adaptation framework that can adapt an APR model for a given target project. We conduct an empirical study with three domain adaptation methods FullFineTuning, TuningWithLightWeightAdapterLayers, and CurriculumLearning using two state-of-the-art domain adaptation tools (TFix and CodeXGLUE) and two APR models on 611 bugs from 19 projects. The results show that our proposed framework can improve the effectiveness of TFix by 13.05% and CodeXGLUE by 23.4%. Another contribution of this study is the proposal of a data synthesis method to address the lack of labelled data in APR. We leverage transformers to create a bug generator model. We use the generated synthetic data to domain adapt TFix and CodeXGLUE on the projects with no data (Zero-shot learning), which results in an average improvement of 5.76% and 24.42% for TFix and CodeXGLUE, respectively.

Recent NLP models have the great ability to generalise `zero-shot' to new tasks using only an instruction as guidance. However, these approaches usually repeat their instructions with every input, requiring costly reprocessing of lengthy instructions for every inference example. To alleviate this, we introduce Hypernetworks for INstruction Tuning (HINT), which convert task instructions and examples using a pretrained text encoder into parameter-efficient modules inserted into an underlying model, eliminating the need to include instructions in the model input. Compared to prior approaches that concatenate instructions with every input instance, we find that HINT models are significantly more compute-efficient and consistently outperform these approaches for a given inference budget.

In this work, we explore "prompt tuning," a simple yet effective mechanism for learning "soft prompts" to condition frozen language models to perform specific downstream tasks. Unlike the discrete text prompts used by GPT-3, soft prompts are learned through backpropagation and can be tuned to incorporate signals from any number of labeled examples. Our end-to-end learned approach outperforms GPT-3's few-shot learning by a large margin. More remarkably, through ablations on model size using T5, we show that prompt tuning becomes more competitive with scale: as models exceed billions of parameters, our method "closes the gap" and matches the strong performance of model tuning (where all model weights are tuned). This finding is especially relevant because large models are costly to share and serve and the ability to reuse one frozen model for multiple downstream tasks can ease this burden. Our method can be seen as a simplification of the recently proposed "prefix tuning" of Li and Liang (2021) and we provide a comparison to this and other similar approaches. Finally, we show that conditioning a frozen model with soft prompts confers benefits in robustness to domain transfer and enables efficient "prompt ensembling." * Work done as a Google AI Resident.

Prior work has shown that it is possible to expand pretrained Masked Language Models (MLMs) to new languages by learning a new set of embeddings, while keeping the transformer body frozen. Despite learning a small subset of parameters, this approach is not compute-efficient, as training the new embeddings requires a full forward and backward pass over the entire model. In this work, we propose mini-model adaptation, a compute-efficient alternative that builds a shallow mini-model from a fraction of a large model's parameters. New language-specific embeddings can then be efficiently trained over the mini-model, and plugged into the aligned large model for rapid cross-lingual transfer. We explore two approaches to learn mini-models: MiniJoint, which jointly pretrains the primary model and the mini-model using a single transformer with a secondary MLM head at a middle layer; and MiniPost, where we start from a regular pretrained model and build a mini-model by extracting and freezing a few layers and learning a small number of parameters on top. Experiments on XNLI, MLQA and PAWS-X show that mini-model adaptation matches the performance of the standard approach using up to 2.4x less compute.

由于表现强劲，预用的语言模型已成为许多NLP任务的标准方法，但他们培训价格昂贵。我们提出了一个简单高效的学习框架TLM，不依赖于大规模预制。给定一些标记的任务数据和大型常规语料库，TLM使用任务数据作为查询来检索一般语料库的微小子集，并联合优化任务目标和从头开始的语言建模目标。在四个域中的八个分类数据集上，TLM实现了比预用语言模型（例如Roberta-Light）更好地或类似的结果，同时减少了两个数量级的训练拖鞋。高精度和效率，我们希望TLM将有助于民主化NLP并加快发展。

Recent trends in language modeling have focused on increasing performance through scaling, and have resulted in an environment where training language models is out of reach for most researchers and practitioners. While most in the community are asking how to push the limits of extreme computation, we ask the opposite question: How far can we get with a single GPU in just one day? We investigate the downstream performance achievable with a transformer-based language model trained completely from scratch with masked language modeling for a single day on a single consumer GPU. Aside from re-analyzing nearly all components of the pretraining pipeline for this scenario and providing a modified pipeline with performance close to BERT, we investigate why scaling down is hard, and which modifications actually improve performance in this scenario. We provide evidence that even in this constrained setting, performance closely follows scaling laws observed in large-compute settings. Through the lens of scaling laws, we categorize a range of recent improvements to training and architecture and discuss their merit and practical applicability (or lack thereof) for the limited compute setting.