Large pretrained language models have shown surprising In-Context Learning (ICL) ability. With a few demonstration input-label pairs, they can predict the label for an unseen input without additional parameter updates. Despite the great success in performance, the working mechanism of ICL still remains an open problem. In order to better understand how ICL works, this paper explains language models as meta-optimizers and understands ICL as a kind of implicit finetuning. Theoretically, we figure out that the Transformer attention has a dual form of gradient descent based optimization. On top of it, we understand ICL as follows: GPT first produces meta-gradients according to the demonstration examples, and then these meta-gradients are applied to the original GPT to build an ICL model. Experimentally, we comprehensively compare the behavior of ICL and explicit finetuning based on real tasks to provide empirical evidence that supports our understanding. The results prove that ICL behaves similarly to explicit finetuning at the prediction level, the representation level, and the attention behavior level. Further, inspired by our understanding of meta-optimization, we design a momentum-based attention by analogy with the momentum-based gradient descent algorithm. Its consistently better performance over vanilla attention supports our understanding again from another aspect, and more importantly, it shows the potential to utilize our understanding for future model designing.

With the increasing ability of large language models (LLMs), in-context learning (ICL) has become a new paradigm for natural language processing (NLP), where LLMs make predictions only based on contexts augmented with a few training examples. It has been a new trend exploring ICL to evaluate and extrapolate the ability of LLMs. In this paper, we aim to survey and summarize the progress, challenges, and future work in ICL. We first present a formal definition of ICL and clarify its correlation to related studies. Then, we organize and discuss advanced techniques of ICL, including training strategies, prompting strategies, and so on. Finally, we present the challenges of ICL and provide potential directions for further research. We hope our work can encourage more research on uncovering how ICL works and improving ICL in future work.

基础模型由于在广泛的下游应用中的有效性而受到了很多关注。尽管在体系结构方面存在很大的融合，但大多数审慎的模型通常仍用于特定任务或模式。在这项工作中，我们建议将语言模型用作各种基础模型的通用接口。一系列预处理的编码者感知到了多种方式（例如视觉和语言），并与扮演通用任务层角色的语言模型对接。我们提出了一个半伴侣的语言建模目标，以共同确定界面和模块化编码器。我们从因果关系和非因果建模中涵盖了优势和能力，从而结合了两个世界的最佳状态。具体而言，所提出的方法不仅从因果语言建模中继承了内在学习和开放式生成的能力，而且由于双向编码器而有利于填补。更重要的是，我们的方法无缝地解锁了上述功能的组合，例如，通过填充编码器启用了文本学习或指导。各种仅语言和视觉语言基准的实验结果表明，我们的模型表现优于或与鉴定，零弹性概括和几乎没有的学习的专业模型竞争。

Large language models have exhibited intriguing in-context learning capability, achieving promising zero- and few-shot performance without updating the parameters. However, conventional in-context learning is usually restricted by length constraints, rendering it ineffective to absorb supervision from a large number of examples. In order to go beyond few shots, we introduce structured prompting that breaks the length limit and scales in-context learning to thousands of examples. Specifically, demonstration examples are separately encoded with well-designed position embeddings, and then they are jointly attended by the test example using a rescaled attention mechanism. So we can scale the number of exemplars with linear complexity instead of quadratic complexity with respect to length. Experimental results on a diverse set of tasks show that our approach improves end-task performance and reduces evaluation variance over conventional in-context learning as the number of demonstration examples increases. Code has been released at https://aka.ms/structured-prompting.

Transformers have become the state-of-the-art neural network architecture across numerous domains of machine learning. This is partly due to their celebrated ability to transfer and to learn in-context based on few examples. Nevertheless, the mechanisms by which Transformers become in-context learners are not well understood and remain mostly an intuition. Here, we argue that training Transformers on auto-regressive tasks can be closely related to well-known gradient-based meta-learning formulations. We start by providing a simple weight construction that shows the equivalence of data transformations induced by 1) a single linear self-attention layer and by 2) gradient-descent (GD) on a regression loss. Motivated by that construction, we show empirically that when training self-attention-only Transformers on simple regression tasks either the models learned by GD and Transformers show great similarity or, remarkably, the weights found by optimization match the construction. Thus we show how trained Transformers implement gradient descent in their forward pass. This allows us, at least in the domain of regression problems, to mechanistically understand the inner workings of optimized Transformers that learn in-context. Furthermore, we identify how Transformers surpass plain gradient descent by an iterative curvature correction and learn linear models on deep data representations to solve non-linear regression tasks. Finally, we discuss intriguing parallels to a mechanism identified to be crucial for in-context learning termed induction-head (Olsson et al., 2022) and show how it could be understood as a specific case of in-context learning by gradient descent learning within Transformers.

Pre-trained language models (PLMs) have exhibited remarkable few-shot learning capabilities when provided a few examples in a natural language prompt as demonstrations of test instances, i.e., in-context learning. However, the performance of in-context learning is susceptible to the choice of prompt format, training examples and the ordering of the training examples. In this paper, we propose a novel nearest-neighbor calibration framework for in-context learning to ease this issue. It is inspired by a phenomenon that the in-context learning paradigm produces incorrect labels when inferring training instances, which provides a useful supervised signal to calibrate predictions. Thus, our method directly augments the predictions with a $k$-nearest-neighbor ($k$NN) classifier over a datastore of cached few-shot instance representations obtained by PLMs and their corresponding labels. Then adaptive neighbor selection and feature regularization modules are introduced to make full use of a few support instances to reduce the $k$NN retrieval noise. Experiments on various few-shot text classification tasks demonstrate that our method significantly improves in-context learning, while even achieving comparable performance with state-of-the-art tuning-based approaches in some sentiment analysis tasks.

The recent GPT-3 model (Brown et al., 2020) achieves remarkable few-shot performance solely by leveraging a natural-language prompt and a few task demonstrations as input context. Inspired by their findings, we study few-shot learning in a more practical scenario, where we use smaller language models for which fine-tuning is computationally efficient. We present LM-BFF-better few-shot fine-tuning of language models 1 -a suite of simple and complementary techniques for finetuning language models on a small number of annotated examples. Our approach includes (1) prompt-based fine-tuning together with a novel pipeline for automating prompt generation; and (2) a refined strategy for dynamically and selectively incorporating demonstrations into each context. Finally, we present a systematic evaluation for analyzing few-shot performance on a range of NLP tasks, including classification and regression. Our experiments demonstrate that our methods combine to dramatically outperform standard fine-tuning procedures in this low resource setting, achieving up to 30% absolute improvement, and 11% on average across all tasks. Our approach makes minimal assumptions on task resources and domain expertise, and hence constitutes a strong task-agnostic method for few-shot learning. 2 * The first two authors contributed equally. 1 Alternatively, language models' best friends forever. 2 Our implementation is publicly available at https:// github.com/princeton-nlp/LM-BFF.

我们呈现隐藏状态优化（HSO），一种基于梯度的方法，用于提高推理时间的变压器语言模型的性能。类似于动态评估（KRAUE等，2018），HSO计算语言模型分配给评估文本的日志概率的渐变，但使用它来更新缓存的隐藏状态而不是模型参数。我们用预磨削的变换器-XL和GPT-2语言模型测试HSO，在困惑方面发现Wikitext103和PG-19数据集的改进，特别是在评估其培训分布之外的模型时。我们还通过在最近开发的基于少量拍摄评估设置中显示出口，再次展示下游适用性，没有额外的参数或培训数据。

Dynamic evaluation of language models (LMs) adapts model parameters at test time using gradient information from previous tokens and substantially improves LM performance. However, it requires over 3x more compute than standard inference. We present Fast Weight Layers (FWLs), a neural component that provides the benefits of dynamic evaluation much more efficiently by expressing gradient updates as linear attention. A key improvement over dynamic evaluation is that FWLs can also be applied at training time so the model learns to make good use of gradient updates. FWLs can easily be added on top of existing transformer models, require relatively little extra compute or memory to run, and significantly improve language modeling perplexity.

大型语言模型在各种任务上显示出令人印象深刻的几次结果。但是，当知识是此类结果的关键时，就像问题回答和事实检查之类的任务一样，似乎需要存储知识的大量参数计数。众所周知，检索增强模型可以在不需要多个参数的情况下在知识密集的任务上表现出色，但是目前尚不清楚它们是否在几个弹药设置中工作。在这项工作中，我们介绍了地图集，这是一个经过精心设计和预先训练的增强语言模型，能够通过很少的培训示例学习知识密集型任务。我们对包括MMLU，苏格兰短裙和归类等各种任务进行评估，并研究文档索引内容的影响，表明它可以很容易地进行更新。值得注意的是，在自然问题上仅使用64个示例在自然问题上达到超过42 \％的准确性，尽管参数少了50倍，但比540B参数模型的表现优于540b参数模型。

Current large language models can perform reasonably well on complex tasks that require step-by-step reasoning with few-shot learning. Are these models applying reasoning skills they have learnt during pre-training and reason outside of their training context, or are they simply memorizing their training corpus at finer granularity and have learnt to better understand their context? To tease apart these possibilities, we introduce ALERT, a benchmark and suite of analyses for assessing language models' reasoning ability comparing pre-trained and finetuned models on complex tasks that require reasoning skills to solve. ALERT provides a test bed to asses any language model on fine-grained reasoning skills, which spans over 20 datasets and covers 10 different reasoning skills. We leverage ALERT to further investigate the role of finetuning. With extensive empirical analysis we find that language models learn more reasoning skills such as textual entailment, abductive reasoning, and analogical reasoning during finetuning stage compared to pretraining state. We also find that when language models are finetuned they tend to overfit to the prompt template, which hurts the robustness of models causing generalization problems.

Through in-context learning (ICL), large-scale language models are effective few-shot learners without additional model fine-tuning. However, the ICL performance does not scale well with the number of available training samples as it is limited by the inherent input length constraint of the underlying language model. Meanwhile, many studies have revealed that language models are also powerful feature extractors, allowing them to be utilized in a black-box manner and enabling the linear probing paradigm, where lightweight discriminators are trained on top of the pre-extracted input representations. This paper proposes prompt-augmented linear probing (PALP), a hybrid of linear probing and ICL, which leverages the best of both worlds. PALP inherits the scalability of linear probing and the capability of enforcing language models to derive more meaningful representations via tailoring input into a more conceivable form. Throughout in-depth investigations on various datasets, we verified that PALP significantly enhances the input representations closing the gap between ICL in the data-hungry scenario and fine-tuning in the data-abundant scenario with little training overhead, potentially making PALP a strong alternative in a black-box scenario.

理解基于变压器的模型引起了极大的关注，因为它们是机器学习最近技术进步的核心。尽管大多数可解释性方法都依赖于输入的运行模型，但最近的工作表明，零通的方法，即直接解释参数而无需前进/向后传递，对于某些变压器参数是可行的，对于两层注意力网络是可行的。在这项工作中，我们提出了一个理论分析，其中通过将其投影到嵌入式空间（即它们操作的词汇量的空间）中来解释训练有素的变压器的所有参数。我们得出一个简单的理论框架来支持我们的论点，并为其有效性提供了充足的证据。首先，经验分析表明，可以在嵌入空间中解释审计和微调模型的参数。其次，我们提出了框架的两个应用：（a）对齐共享词汇的不同模型的参数，以及（b）通过``翻译''''''''分类器构建分类器的参数``翻译'''''''分类器的参数仅鉴定的不同模型。总体而言，我们的发现为解释方法打开了大门，至少部分地从模型细节中抽象出来，仅在嵌入空间中运行。

预处理的变形金刚记住事实知识的能力对于下游任务（例如封闭式问题答案）是必不可少的。现有的工作表明，经过审计的变压器可以回忆或利用在某种程度上出现的训练训练阶段中出现的事实知识。但是，由于模型能力的限制，预审预周仔的记忆知识的能力也受到限制。 Dai等。 （2022）发现经过验证的变形金刚中的馈电网络（FFN）以内存的方式存储事实知识。受这一发现的启发，我们提出了一个神经知识库（NKB），以存储预验证的变压器的额外事实知识。要具体而言，我们还将FFN视为键值记忆，并使用其他内存插槽扩展它们。在知识注入期间，我们将原始模型和事实知识注入扩展的存储插槽中，因此预验证的模型不会遗忘。此外，FFN作为钥匙值记忆的观点使NKB高度可解释。我们使用三个封闭式问题回答数据集来显示我们强大的存储额外事实知识的能力。另外，我们证明NKB不会通过两种代表性生成任务，摘要和机器翻译来降低验证模型的一般语言生成能力。此外，我们彻底分析了NKB以揭示其工作机制，并以人为可读的方式介绍其钥匙和价值观的含义。最重要的是，我们执行初步尝试，以直接更新NKB中的事实知识，而无需任何其他培训。

变压器注意机制的二次计算和内存复杂性限制了对长序列建模的可扩展性。在本文中，我们提出了Luna，一种线性统一嵌套关注机制，使Softmax注意力具有两个嵌套线性关注功能，仅产生线性（与二次）的时间和空间复杂度相反。具体地，通过第一注意功能，LUNA将输入序列包装成固定长度的序列。然后，使用第二关注功能未包装包装序列。与更传统的关注机制相比，LUNA引入具有固定长度的附加序列作为输入和额外的相应输出，允许LUNA线性地进行关注操作，同时还存储足够的上下文信息。我们对三个序列建模任务的基准进行了广泛的评估：长上下文序列建模，神经机平移和大型预磨损的屏蔽语言建模。竞争甚至更好的实验结果表明了Luna的有效性和效率与各种各样相比

迅速的学习方法通​​过诱导更好的几次表现，在他们仍然遵循基于参数的学习范式的同时，引起了自然语言处理的波动。学习中的遗忘和死记硬背的记忆问题可能会遇到不稳定的概括问题。具体而言，香草及时的学习可能难以利用死记硬背的非典型实例，在完全监督的培训或过度贴身模式的情况下使用低射击数据。为了减轻此类局限性，我们以将知识从记忆中解耦的动机发展为有助于模型在概括和记忆之间取得平衡。与香草及时学习相反，重新启动构造了培训实例中的开放式知识店，并在输入，培训和推理过程中实现检索机制，从而使该模型能够从培训语料库中检索相关环境作为能力为提示增强。广泛的实验表明，Retroppt可以在几次射击和零拍设置中获得更好的性能。此外，我们进一步说明，我们提出的撤退可以通过新数据集获得更好的概括能力。对记忆的详细分析确实显示逆转可以减少语言模型对记忆的依赖；因此，改善下游任务的概括。

Given the success with in-context learning of large pre-trained language models, we introduce in-context learning distillation to transfer in-context few-shot learning ability from large models to smaller models. We propose to combine in-context learning objectives with language modeling objectives to distill both the ability to read in-context examples and task knowledge to the smaller models. We perform in-context learning distillation under two different few-shot learning paradigms: Meta In-context Tuning (Meta-ICT) and Multitask In-context Tuning (Multitask-ICT). Multitask-ICT performs better on multitask few-shot learning but also requires more computation than Meta-ICT. Our method shows consistent improvements for both Meta-ICT and Multitask-ICT on two benchmarks: LAMA and CrossFit. Our extensive experiments and analysis reveal that in-context learning objectives and language modeling objectives are complementary under the Multitask-ICT paradigm. In-context learning objectives achieve the best performance when combined with language modeling objectives.

几乎没有射击的内在学习（ICL）使预训练的语言模型能够通过为输入的一部分提供少量的培训示例来执行以前的任务，而无需任何基于梯度的培训。 ICL会产生大量的计算，内存和存储成本，因为它每次进行预测时都涉及处理所有培训示例。参数有效的微调（PEFT）（例如，适配器模块，提示调谐，稀疏更新方法等）提供了替代范式，其中训练了一组少量参数以启用模型来执行新任务。在本文中，我们严格地比较了几个ICL和PEFT，并证明后者提供了更好的准确性，并大大降低了计算成本。在此过程中，我们引入了一种称为（IA）$^3 $的新PEFT方法，该方法通过学习的向量来扩展激活，从而获得更强的性能，同时仅引入相对少量的新参数。我们还提出了一个基于称为T-FEW的T0模型的简单食谱，可以将其应用于新任务，而无需特定于任务的调整或修改。我们通过将T-FEW应用于木筏基准，首次实现超人性能，并以6％的绝对性能优于最先进的方法来验证T-FEW对完全看不见的任务的有效性。我们实验中使用的所有代码均可公开使用。

Recent work has demonstrated substantial gains in pre-training large-scale unidirectional language models such as the GPT-2, GPT-3, and GPT-neo, followed by fine-tuning on a downstream task. In this paper, we evaluate the performance of the GPT-neo 1.3 billion model for commonsense reasoning tasks. We assess the model performance on six commonsense reasoning benchmark tasks and report the accuracy scores for these tasks. When fine-tuned using the right set of hyperparameters, we obtain competitive scores on three of these tasks but struggle when the dataset size is significantly smaller. The low model performance on a few of these tasks suggests the inherent difficulty in these datasets and since it fails to establish coherent patterns given their limited training samples. We also investigate and substantiate our results using visualization and conduct numerous inference tests to understand the model performance better. Finally, we conduct thorough robustness tests using various methods to gauge the model performance under numerous settings. These findings suggest a promising path for exploring smaller language models than the GPT-3 175 billion model to perform tasks requiring natural language understanding.

专家层（MOES）的混合物通过条件计算实现语言模型的高效缩放。本文提出了一个详细的实证研究，自回归鞋语言模型与广泛的设置中的密集模型相比：在域外语言建模，零和少量射击和全部微调。除了微调外，我们发现Moes基本上更加计算效率。在更适度的培训预算下，MOES可以使用$ \ SIM值4倍的计算，符合密集模型的性能。该差距在比例下变窄，但我们最大的MOE模型（1.1T参数）始终如一地优于计算等效的密集模型（6.7b参数）。总体而言，这种表现差距在任务和域中有很大差异，表明MOE和密集模型以不值得研究的方式概括不同的方式。我们使我们的代码和模型公开可用于研究使用。