Datasets:
Translation-EnKo
/
arxiv-translation

Dataset card Data Studio Files
xet
Community
2
text
stringlengths
4
4.47k
In order to produce high-quality simulated dialogues at scale, we developed a novel multi-agent framework which comprised three key components:
While the authors of both RealNews and C4 explicitly attempted deduplication during dataset construction, the methods were insufficient to capture the more subtle types of duplicate text commonly found on the internet. In C4 and Wiki-40B, we qualitatively observe that much of the text identified as near-duplicated is computer-generated. The text is identical except for the names of places, businesses, products, dates, and so on. Because these examples frequently differ by just a few words at a time, deduplication strategies relying on exact string matching would fail to identify a match. Example duplicate pairs from each dataset can be found in Table 1 (more examples in the Appendix).
In this section, we introduce the concept of using a strong LLM as a codec, _i.e._, both encoder and decoder, for instruction generation.
**Query cost:** the total duration of the attack. Most APIs place a limit on the number of queries an adversary can make in any given interval, and so some attacks may be faster but cost more (by sending more tokens per query).
We also report evaluation results for ablated models in Tab. 2. 'SOLAR 10.7B + SFT + DPO' uses all the DPO data at once, _i.e.,_\(D_{1}+D_{2}\), same as the conventional DPO training setup. 'SOLAR 10.7B + SFT + sDPO Strat.' uses stratified sampling to sample \(\sim 16.67\%\) of the data points from the union of OpenOrca and Ultrafeedback Cleaned to form \(D_{1}\) and use the remaining \(\sim 83.33\%\) as \(D_{2}\) to mirror the dataset size of \(D_{1}\) and \(D_{2}\) used in SOLAR 10.7B + SFT + sDPO.
Atlas (Izacard et al., 2023) investigates few-shot learning approaches, including Attention Distillation and Perplexity Distillation, to steer the retriever toward retrieving more relevant documents. IRCOT (Trivedi et al., 2023) integrates retrieval with reasoning to improve the effectiveness of retrieval. SURGE (Kang et al., 2023) employs a subgraph retriever to extract relevant subgraphs from a knowledge graph, while AAR (Yu et al., 2023) modifies search preferences to help LLMs in fetching pertinent documents.
Xiao Wang,, Shiao Wang, Yuhe Ding, Yuehang Li, Wentao Wu, Yao Rong, Weizhe Kong, Ju Huang, Shihao Li, Haoxiang Yang, Ziwen Wang, Bo Jiang, Chenglong Li, Yaowei Wang, Yonghong Tian,, Jin Tang
To evaluate financial domain tasks, we compare FinPythia and Fynthia-1B exhibit superior performance on tasks FPB, Headline, and NER while showing comparatively lower results on the FiQA SA task compared with Pythia counterparts. DACP boosts the average task performance by 2.8% for the 1B model and 8.3% for the 6.9B model. These outcomes directly substantiate the impact of domain-adaptive continual pre-training on enhancing in-domain task performance. Furthermore, Pythia-6.9B outperforms OPT-7B, BLOOM-7B, and GPT-J-GB on average.
\(\bullet\)_Instruction following_. By fine-tuning with a mixture of multi-task datasets formatted via natural language descriptions (called _instruction tuning_), LLMs are shown to perform well on unseen tasks that are also described in the form of instructions [28, 66, 67]. With instruction tuning, LLMs are enabled to follow the task instructions for new tasks without using explicit examples, thus having an improved generalization ability. According to the experiments in [67], instruction-tuned LaMDA-PT [68] started to significantly outperform the untuned one on unseen tasks when the model size reached 68B, but not for 8B or smaller model sizes. A recent study [69] found that a model size of 62B is at least required for PaLM to perform well on various tasks in four evaluation benchmarks (_i.e._, MMLU, BBH, TyDiQA and MGSM), though a much smaller size might suffice for some specific tasks (_e.g._, MMLU).
Since LMs are limited by their context window, we summarize the articles. In particular, we prompt GPT-3.5-Turbo to distill the most relevant details from each article with respect to the forecasting question (Figure 13). Finally, we present the top \(k\) article summaries to the LM, ordered by their relevancy. We choose the ranking criterion, article count \(k\), and summarization prompt based on end-to-end Brier scores over the validation set; see Section 5.2 for the hyperparameter sweep procedure.
To approach GPT-3.5-matching human preferences, our finetuning dataset is curated from carefully selected multi-turn instruction-response pairs, annotated directly by our team of machine learning engineers then polished over multiple iterations of user feedback. As mentioned above, the size of our finetuning dataset is less than 10K, but improved over and over again across the model development timeline. Benefiting from the dataset's manageable size, we employed an extensive grid search to identify the optimal data composition, promote diversity, and discover effective hyperparameters. After 8-bit and 4-bit quantization, the final chat model can be deployed on consumer-grade GPUs nearly without performance degradation compared to the bf16 format.
A: Let's think step by step. How many fresh eggs remains after the three her eat for breakfast and four goes into the muffin mix for her friends. We can calculate it by 16-3-4. The amount equals 9 fresh eggs. So, the dollars it makes daily equals what multiplied by $2. 12 x 2. The answer is $24.
And more recently, **Mixture of Experts (MoE)**[130, 131] have become very popular in LLM space too. MoEs enable models to be pre-trained with much less compute, which means one can dramatically scale up the model or dataset size with the same compute budget as a dense model. MoE consists of two main elements: **Sparse MoE layers**, which are used instead of dense feed-forward network (FFN) layers, and have a certain number of "experts" (e.g. 8), in which each expert is a neural network. In practice, the experts are FFNs, but they can also be more complex networks. **A gate network or router**, that determines which tokens are sent to which expert. It is worth noting that, one can send a token to more than one expert. How to route a token to an expert is one of the big decisions when working with MoEs - the router is composed of learned parameters and is pretrained at the same time as the rest of the network. Fig 29 provides an illustration of a Switch Transformer encoder block, which are used in MoE.
Leveraging the insights from Section 6.2, we outperform the crowd by making selective forecasts.
Mixing HeadFor our mixing head, we use a three-layer MLP with ReLU activation, taking in a vector of two times the size of the hidden state of the language model (as we concatenate the two predictions to determine their weights), and outputting a scalar. This scalar is them used to weight the logits from the LM head with and without thinking to make a prediction from a given token.
As the field of synthetic data continues to evolve, there are several promising directions for future research and development.
_Remark 5.1_.: Result 1(a), 1(b), and 1(c) elucidate _three distinct facets_ of the scaling law.
To tackle the challenge of enhancing large language models' (LLMs) ability to sift through irrelevant text within the retrieval pipeline, our analysis revealed that training solely with oracle (highly relevant) documents can inadvertently diminish the model's ability to discern and disregard irrelevant information. To address this, our algorithm, RAFT, adopts a strategy that integrates oracle documents with a mix of irrelevant ones. This methodology prompts us to investigate the ideal fraction of negative (irrelevant) documents to incorporate throughout the training process and to assess how well this training approach adapts to different volumes of documents encountered by the Retrieval-Augmented Generation (RAG) during the test phase. Our aim is to refine the balance between relevant and irrelevant information to strengthen the model's efficiency in identifying and utilizing pertinent content. Notice that Sec 4.5 looked at what P% of training data should include distractors, while in this section, we study test-time scenarios.
**Emergent Abilities of LLMs**. In the literature [31], _emerget abilities_ of LLMs are formally defined as "the abilities that are not present in small models but arise in large models", which is one of the most prominent features that distinguish LLMs from previous PLMs. It further introduces a notable characteristic when emergent abilities occur [31]: performance rises significantly above random when the scale reaches a certain level. By analogy, such an emergent pattern has close connections with the phenomenon of _phase transition_ in physics [31, 63]. In principle, emergent abilities can be defined in relation to some complex tasks [31, 64], while we are more concerned with general abilities that can be applied to solve a variety of tasks. Here, we briefly introduce three typical emergent abilities for LLMs and representative models that possess such an ability8.
Detect malicious queries.Instead of preventing any queries that might leak model weights, an alternate strategy could be to implement standard anti-abuse tools to _detect_ any patterns of malicious behavior. Several proposals of this form exist for prior machine learning attacks, including model stealing (Juuti et al., 2019; Pal et al., 2021) and adversarial examples (Chen et al., 2020).
Retrieving information from an external knowledge source as described above is only one of the potential ways to augment an LLM. More generally, an LLM can access any number of external tools (e.g. an API to a service) to augment its functionality. In that regards, RAG can be seen as a specific instance of the broader category of the so called "tools".
In addition to the above library resources, existing deep learning frameworks (_e.g._, PyTorch [197], TensorFlow [198], MXNet [199], PaddlePaddle [200], MindSpore [136] and OneFlow [201]) have also provided the support for parallel algorithms, which are commonly used for training large-scale models.
Figure 6: The validation perplexity on the Pile validation set for 1B models trained on the default mixture and the optimized mixture of RedPajama for 100B tokens. Our optimized mixture achieves the performance of the default mixture only using 0.73 of the original number of training steps and eventually achieves a performance comparable to a default mixture trained with 1.48 times more tokens (estimated by the scaling law of training steps, shown as the dashed line). The specific mixture proportions are in the right table.
One way to address the problem at hand could be to train shallow small language models, increasing hidden dimension at the expense of other hyperparameters, such as layer count or feed-forward dimension. However, we believe that such research directions may not be promising in this context. Previous works have extensively explored and optimized the hyperparameter choices for various architecture sizes. The impact of width and depth has been extensively studied (Merrill et al., 2022; Tay et al., 2022; Petty et al., 2023), often showcasing the importance of depth in final performance and generalization capabilities.
Other reasoning tasks.Synthetic data also leads to impressive performance in other reasoning tasks. For instance, Wei et al. (2023a) augmented existing natural language datasets by replacing natural language labels with arbitrary symbols, generating over 500k synthetic examples. Using these synthetic data for supervised finetuning significantly improved model performance on unseen in-context learning and algorithmic-reasoning tasks. STaR (Zelikman et al., 2022) generates synthetic chain-of-thought rationales and filters out those leading to wrong answers for finetuning language models to improve their reasoning. In the domain of physics reasoning, Mind's Eye (Liu et al., 2022) takes a novel approach by training a text-to-code model with synthetic "text-description \(\rightarrow\) rendering code" data. This enables the model to convert textual questions into rendering code, which is then executed in a physical engine (i.e., DeepMind MuJoCo (Todorov et al., 2012)). The rendering results are injected into the context, allowing even small language models armed with Mind's Eye to achieve performance comparable to models 100 times larger.
Self-Rag trains an arbitrary LM to generate text with reflection tokens by unifying them as the next token prediction from the expanded model vocabulary. We train our generator LM on a diverse collection of text interleaved with reflection tokens and retrieved passages. Reflection tokens, inspired by reward models used in reinforcement learning (Ziegler et al., 2019; Ouyang et al., 2022), are inserted offline into the original corpus by a trained _critic_ model. This eliminates the need to host a critic model during training, reducing overhead. The critic model, in part, is supervised on a dataset of input, output, and corresponding reflection tokens collected by prompting a propriety LM (i.e., GPT-4; OpenAI 2023). While we draw inspiration from studies that use control tokens to start and guide text generation (Lu et al., 2022; Keskar et al., 2019), our trained LM uses critique tokens to assess its own predictions after each generated segment as an integral part of the generation output.
To find the value of \(f(-2)+f(-1)+f(0)\), we first need to evaluate each function at these specific values of x.
Footnote 4: We follow Liu et al. (2023a) in using a “perceived” utility value that is independent of retrieved passages.
Examining the parameters individually, \(E\) and \(\beta\) show highly significant differences, with \(p\)-values of \(1.5\times 10^{-6}\) and \(4.3\times 10^{-5}\), respectively. These small \(p\)-values indicate that the estimated values for \(E\) and \(\beta\) are significantly different from their respective true values, suggesting substantial deviations in these parameters' estimates from expected outcomes based on the model. On the other hand, the parameters \(A\), \(B\), and \(\alpha\) have estimated values that are not statistically significantly different from the values reported in Hoffmann et al.
* **ARC4**Clark et al. (2018): A collection of natural, grade-school science questions. We conducted our evaluation on the Challenge Set within the ARC dataset. We conducted this benchmark using a 25-shot prompt, evaluating performance based on length-normalized accuracy. Footnote 4: [https://allenai.org/data/arc](https://allenai.org/data/arc) * **Hellaswag5**Zellers et al. (2019): An evaluation of commonsense inference, presenting a task that is straightforward for humans but poses a challenge for state-of-the-art models.
Our cross-model analysis aims to understand the nuances and intricacies of model performance across different architectures and scales.
Figure 15: 8-bit/4-bit quantization of GPT2 models trained on \(\mathsf{bioD}(N,K,C,D,L,T)\) data for **1000 exposures. Left:** Identical to Figure 2, showing only models with accuracy \(\geq 50\%\); **Middle:** After quantization to 8-bit; **Right:** After quantization to 4-bit, including models with all accuracies.
While there is speculation and some suggestive comments from McCarthy about the possibility of resignation, there is no concrete evidence or announcement that he will resign by the specified date. His history of fundraising and political influence, as well as the lack of a definitive statement about his future, suggest that a resignation by November 30 is not certain.
Reason and Act (ReAct)[176] prompts LLMs to generate not only verbal reasoning but also actionable steps, thus enhancing the model's dynamic problem-solving capabilities. ReAct is grounded in the principle of integrating reasoning with action. In this approach, the LLM is prompted to alternate between generating reasoning traces (explanations) and taking actions (steps or commands) in an interleaved manner. This approach allows the model to dynamically reason about a problem, and propose and take concrete actions simultaneously.
Table 12 shows the prediction results of T5-Vanilla and T5-Modular on three knowledge probing tasks: InvariantLAMA, UpdatedLAMA, and NewLAMA. We show the prediction for every training epoch for each model. The instances are selected from the predictions that T5-Modular got correct but T5-Initial got wrong on the final prediction, in order to see where the gap of the EM comes from.
Zhengxiao Du\({}^{1,2}\), Aohan Zeng\({}^{1,2}\), Yuxiao Dong\({}^{2}\), Jie Tang\({}^{2}\)
Experimental Framework We used the popular LM-Evaluation-Harness (Gao et al., 2021) repository to evaluate the performance of LLMs on the selected knowledge-intensive tasks. LM-Evaluation-Harness is a robust benchmarking tool that currently serves as the industry standard for model evaluation and is the basis of the HuggingFace leaderboard3. Leveraging this platform ensured a standardized evaluation framework and allowed consistent comparison across models, methods, and datasets. More importantly, by using the industry standard for evaluation, we could avoid any differences stemming from prompt engineering and formatting issues and replicate the reported baseline results for each model.
Figure 3: Evolution of the language modeling performance on the Wikipedia test set from the LM Evaluation Harness (Gao et al., 2023) and last-layer anisotropy of Pythia models along training (color).
**Output**[[**Retriee**]=No[The manager realizes that there is an extensive list of tasks that need to be completed in order to make the event a success.[[**Retriee**]=Yes]<p>Delegation-Delegation is the assignment of any responsibility or authority to another person (normally from a manager to a subordinate) to carry out specific activities. It is one of the core concepts of management leadership. However, the person who delegated the work remains accountable for the outcome of the delegated work. Delegation empowers a subordinate to make decisions, i.e. it is a shifting of decision-making authority from one organizational level to a lower one. Delegation, if properly done, is not fabrication. The opposite of effective delegation is micromanagement, where a manager provides too much input,<p>[[**Isstr**]=RelevantThis includes tasks such as creating a guest list, designing and sending invitations, booking a venue, coordinating with vendors, planning the event program, and monitoring the budget.[[[**Isstr**]=Partially Supported[**Retriee**]=No[Since it is not possible for the manager to handle everything alone, they must delegate tasks to their team members.[[**Retriee**]=No[The manager would then assign each team member specific responsibilities based on their individual skillsets and strengths.[[**Retriee**]=Yes]<p>Joint application design - Subject Matter Experts: These are the business users, the IS professionals, and the outside experts that will be needed for a successful workshop.
Footnote 1: Our code and trained models are available at [https://selffrag.github.io/](https://selffrag.github.io/).
In this work, we present a novel framework, **CodecLM**, which systematically generates tailored high-quality data to align LLMs for different downstream tasks. A high-level overview of CodecLM is shown in Figure 1. Inspired by the principles of Encode-Decode process Kramer (1991); Kingma and Welling (2013), we leverage a strong LLM as a codec to "encode" seed instructions from our target task into instruction _metadata_ and then "decode" the metadata into tailored synthetic instructions. The metadata serves as a word-level abstraction of the input instruction distribution, including the _use case_ and _skills_ for effective instruction following. It can be automatically generated by encoding seed instructions, or directly provided by users with a high-level anticipation of the downstream task.
Language is a prominent ability in human beings to express and communicate, which develops in early childhood and evolves over a lifetime [3, 4]. Machines, however, cannot naturally grasp the abilities of understanding and communicating in the form of human language, unless equipped with powerful artificial intelligence (AI) algorithms. It has been a longstanding research challenge to achieve this goal, to enable machines to read, write, and communicate like humans [5].
In this report, we discuss the full-stack development of the Yi language model family. Yi-34B achieves GPT-3.5 matching performance and is deployable (thank to the 4/8-bit quantization) on consumer-grade devices, making it an ideal model for local deployment.
their settings, at test time, the domain or documents can be different than the training time; whereas our paper studies a slightly opposite scenario where we only care about testing the LLM on the same set of documents.
We first study how the performance varies with different numbers of epochs. As shown in Tab. 2, training for 1 epoch is insufficient, and the performance of 3, 5, or 10 epochs is similar. We fix the number of epochs to 3 and arrange the order of QA pairs and corresponding documents as shown in Fig. 6. The interleaved arrangement cycles through all the data 3 times, ensuring that in each epoch, questions either precede or follow their associated documents. On the other hand, the grouped arrangement clusters each example's 3 appearances together, guaranteeing that the repeated questions are positioned either before or after their respective repeated documents. As shown in Tab. 2, positioning QA pairs before corresponding documents achieves better performance in both grouped and interleaved arrangements, indicating that during PIT, the learning mechanism prioritizes understanding how to access knowledge before learning to absorb information from the more complex and information-dense documents.
**Methods.** Due to cost constraints, we cannot afford to evaluate relevance scores using the full article texts.
_Add Constraint._ Adding one more constrain to the original question while keeping others unchanged, which is used in WizardMath and MuggleMath.
Our definition of terrorism includes but is not limited to the following ideologically defined groups. They have been extensively studied in previous work [179, 180].
where \(c\) represents the number of correct answers out of \(N\) responses. This metric considers a question to be solved if at least one correct answer is produced from \(N\) random generations. We employ this metric to reflect the potential or capability of a model in solving a math question. To enhance the diversity of the \(N\) generations, we set the temperature of the generation process to 0.73.
Smith [162] is a novel spatiotemporal modeling approach focused on improving the efficiency and performance of modeling long sequence data. It combines the advantages of Convolutional Neural Networks (CNNs) and state space methods to effectively overcome the limitations of traditional models in dealing with complex spatial correlations and long time dependencies. ConvSSM significantly improves the speed of model training and prediction through parallel scanning and fast auto-regressive generation techniques. Meanwhile, it draws on state-space methods such as S4 and S5 to provide effective parameterization and initialization strategies for long-distance dependency modeling. This approach not only reduces the consumption of computational resources but also maintains a performance level comparable to that of complex models. Compared to other models, such as Transformer, whose computational cost grows significantly with sequence length, and ConvLSTM, which has a slower training speed, ConvSSM demonstrates its potential in long sequence spatiotemporal modeling. In addition, researchers are also exploring ways to further improve model performance by optimizing convolutional kernel design and combining the advantages of different models. [165] proposes an approach that fuses a hybrid Mixture of Expert (MoE) mechanism and a selective State Space Model (SSM) with the aim of optimising the efficiency and performance of sequence modeling. This approach introduces MoE on top of the Mampa model, achieving similar performance in fewer training steps and maintaining the inference performance advantage over the Transformer model. The MoE model utilizes dynamic gating, expert caching, and load balancing techniques to address the communication and memory issues faced by large models during the inference phase, while conditionally computing to extend the model without significantly increasing the computational cost while expanding the model capacity. It is shown that MoE models with instruction fine-tuning and task-specific fine-tuning are able to achieve better performance than denser models with the same computational complexity. The success of MoE-Mampa demonstrates that the performance and efficiency of sequence modeling tasks can be effectively improved by a well-designed MoE architecture, which opens up new possibilities for processing large-scale sequence data. Xu et al. [186] demonstrate the potential of Mampa models for classical information retrieval tasks by evaluating the performance of models based on the Mampa architecture in a document ranking task. Compared to Transformer-based language models, Mampa models achieve competitive performance in the same training configuration. However, the Mampa model falls short in terms of training throughput compared to efficient attention implementations, limiting its potential for efficient training and deployment. Although the study focuses on models with parameters less than 1B, they find that this may change when scaling to larger models and employing different training configurations. Thus, the effectiveness and performance of Mampa models in other classical information retrieval tasks remain to be further investigated. Amo et al. [193] introduce the SSM system and summarize the research progress in the field of control-theoretic. Sharma et al. [187] are inspired by the research found in the autoregressive transformer language model that knowledge recall can be specific to specific modules and marker positions, and aims to locate factual recall in Mampa. Olucha et al. proposed an overview of the state-of-the-art and comparative study of linear parameter-varying state-space (LPV-SS) model downscaling [194]. These comparisons can help to select the best downscaling method for a given LPV-SS model. Yang et al. propose the Rec-Mampa [190] model for sequence recommendation tasks. This model better models users' preference information that changes over time through the Mampa module, improving personalized recommendation performance and reducing computational complexity. Compared to SASRec, the training time is reduced by approximately \(70\%\). LaRocque et al. [183] compares the performance of Convolutional Neural Networks (CNNs) and the novel State Space Model (SSM)-based Mampa architecture, revealing intriguing insights into their respective strengths and the benefits of dataset fusion for improved terrain.
Chain StructurePAL (Gao et al., 2023) and PoT (Chen et al., 2022) introduce programming languages to describe the reasoning process, thereby converting the reasoning problem into the implementation of an executable program to obtain the final answer. Since the program execution is deterministic and performs arithmetic computations accurately, this approach shows excellent performance in mathematical reasoning. Besides, symbol sequence is another type of thought representation. Chain-of-Symbol (Hu et al., 2023) represents the complex environments with condensed symbolic chain representations during planning, which reduces the complexity of the simulation environment. Chain structure variants are shown in Figure 2(c,d) Algorithm of Thought (Sel et al., 2023) injects algorithmic capabilities into the model, making the model's reasoning more logical by adding examples based on algorithms. Its absence of the huge search space of tree search (Long, 2023; Yao et al., 2023) saves computational resources and achieves excellent performance.
The mayoral election in Indianapolis took place on the 7th of November, 2023, following primary elections that occurred on the 2nd of May. Joe Hogsett, the incumbent Democrat, successfully ran for a third term. Both Hogsett and his Republican challenger, Jefferson Shreve, made it through to the final round of the election.
\(\bullet\)_Task description._ A task description is typically a specific instruction that LLMs are expected to follow. In general, one should clearly describe the task goal in natural language. For the tasks with special input or output format, detailed clarifications are often needed, and one can further utilize keywords to highlight the special settings for better guiding LLMs in task completion.
Despite their remarkable capabilities, large language models (LLMs) often produce responses containing factual inaccuracies due to their sole reliance on the parametric knowledge they encapsulate. Retrieval-Augmented Generation (RAG), an ad hoc approach that augments LMs with retrieval of relevant knowledge, decreases such issues. However, indiscriminately retrieving and incorporating a fixed number of retrieved passages, regardless of whether retrieval is necessary, or passages are relevant, diminishes LM versatility or can lead to unhelpful response generation. We introduce a new framework called **Self-Reflective Retrieval-Augmented Generation (Self-RAG)** that enhances an LM's quality and factuality through retrieval and self-reflection. Our framework trains a single arbitrary LM that adaptively retrieves passages on-demand, and generates and reflects on retrieved passages and its own generations using special tokens, called _reflection_ tokens. Generating reflection tokens makes the LM controllable during the inference phase, enabling it to tailor its behavior to diverse task requirements. Experiments show that Self-RAG (7B and 13B parameters) significantly outperforms state-of-the-art LLMs and retrieval-augmented models on a diverse set of tasks.
These strategies in the above methods can be jointly used to improve the quantization performance. In order to achieve high-efficiency implementation, quantization methods also rely on hardware- or system-level support (_e.g.,_ efficient GPU kernels or hardware-friendly group partition).
**Calibration.** Calibration is important for accurate forecasting (Tetlock and Gardner, 2015). Hence, on competitive forecasting tournaments, forecasters are evaluated by proper scoring rules, such as Brier score (Brier, 1950), which incentivize calibration (Gneiting and Raftery, 2007). There is a vast literature on calibration in deep learning; see Gawlikowski et al. (2021); Wang (2023) for surveys.
propose to train deep and thin networks for SLMs, which aligns with our perspective. Therefore, we further make the architecture deeper and thinner for MiniCPM-1.2B.
In this study, we initially proposed the potential quality issues of gold references in the MT task, highlighting instances where advanced translation models can outperform these references. This finding not only challenges model training via SFT, but also the evaluation procedure that uses reference-based metrics. Subsequently, we introduce Contrastive Preference Optimization, a more efficient variant of of DPO. This method leverages both model-generated and reference data to guide the model in avoiding near-perfect yet flawed translations and learning superior ones. Our developed model, ALMA-13B-R, stands out as the first moderate-size LLM-based translation model to match, and in some cases surpass, the performance of GPT-4 and WMT competition winners, marking a significant advancement in the field of MT.
**Output**[[**Retriee**]=Yes]<p>Computer memory Computer memory – In computing, memory refers to the computer hardware integrated circuits that store information for immediate use in a computer; it is synonymous with the term “primary storage”. Computer memory operates at a high speed, for example random-access memory (RAM), as a distinction from storage that provides slow-to-access information but offers higher capacities. If needed, contents of the computer memory can be transferred to secondary storage; a very common way of doing this is through a memory management technique called “virtual memory”. An archic synonym for memory is store. The term “memory”, meaning “primary storage” or “main memory”, is<p>[[**Isstr**]=RelevantPrimary** memory, **also known as main memory, is the memory that the computer can directly access and manipulate. [[**Isstr**]=Fully Supported[**Retriee**]=Yes]<p>Computer memory-often associated with addressable semiconductor memory, i.e. integrated circuits consisting of silicon-based transistors, used for example as primary storage but also other purposes in computers and other digital electronic devices. There are two main kinds of semiconductor memory, volatile and non-volatile. Examples of non-volatile memory are flash memory (used as secondary memory) and ROM, PROM, EPROM and EEPROM memory (used for storing firmware such as BIOS). Examples of volatile memory are primary storage, which is typically dynamic random-access memory (DRAM), and fast CPU cache memory, which is typically static random-access memory (SRAM) that is fast but energy-consuming, offering lower memory<p>[[**Isstr**]=RelevantPrimary memory can be **accessed quickly, but it is volatile, meaning that the data is lost if the power is turned off.[[**Isstr**]=Partially Support[**].[[**Isstr**]=**0[The manager realizes that there is an extensive list of tasks that need to be completed in order to make the event a success.[[**Retriee**]=Yes]<p>Delegation-Delegation is the assignment of any responsibility or authority to another person (normally from a manager to a subordinate) to carry out specific activities. It is one of the core concepts of management leadership. However, the person who delegated the work remains accountable for the outcome of the delegated work. Delegation empowers a subordinate to make decisions, i.e. it is a shifting of decision-making authority from one organizational level to a lower one. Delegation, if properly done, is not fabrication. The opposite of effective delegation is micromanagement, where a manager provides too much input,<p>[[**Isstr**]=Relevant**]This includes tasks such as creating a guest list, designing and sending invitations, booking a venue, coordinating with vendors, planning the event program, and monitoring the budget.[[**Isstr**]=Partially Supported[**Retriee**]=No[**Since it is not possible for the manager to handle everything alone, they must delegate tasks to their team members.[[**Retriee**]=No[The manager would then assign each team member specific responsibilities based on their individual skillsets and strengths.[[**Retriee**]=Yes]<p>Joint application design – Subject Matter Experts: These are the business users, the IS professionals, and the outside experts that will be needed for a successful workshop.
Small LMs & SaturationBiderman et al. (2023) train Pythia, a suite of models of various sizes on 300B tokens from the Pile (Gao et al., 2020), and release the weights for an exhaustive number of checkpoints during training. They notice that smaller models suffer a performance decrease on the Lambada dataset (Paperno et al., 2016) in late training. The scaling laws (Kaplan et al., 2020; Hoffmann et al., 2022) predict that training smaller models on large corpora is suboptimal in terms of compute. However, recent initiatives (Zhang et al., 2024; Faysse et al., 2024; Team et al., 2024) have pretrained smaller language models on large datasets, motivated by inference cost reduction (Sardana & Frankle, 2023).
In constrast, the storage manager is responsible for the long-term preservation of data, overseeing the storage of information that needs to be retained indefinitely, beyond the active lifespan of any
Contrastive Filtering complements Self-Rubrics to select effective instruction-response pairs by calibrating the target LLM's instruction-following capability with the strong LLM's. Analogous to Contrastive Decoding (Li et al., 2022) at response-level, Contrastive Filtering can also be regarded as LLM-feedback (Madaan et al., 2023) with the interaction of two LLMs. While we adopt the strong LLM as scoring function to measure the quality gap, our framework can be compatible with and potentially benefit from the advances in more reliable and comprehensive scoring and feedback systems (Lee et al., 2023), and we leave it as promising future work.
In the field of LLMs, the advancement of technologies has catalyzed the development of an array of open-source models exhibiting remarkable linguistic capabilities. Notably, models such as LLaMA (Touvron et al., 2023; Zhu et al., 2023), Phi (Li et al., 2023; Gunasekar et al., 2023), Mistral (Jiang et al., 2023), and Gemma (Team et al., 2024) have emerged as frontrunners, underscoring the technological strides made in this arena. Amidst a globalized context, there's a rising demand for models proficient in bilingual or multilingual functionalities, particularly those accommodating the vast spectrum of Chinese language applications. This demand stems from the desire for localized solutions and the necessity to bridge linguistic divides worldwide. To address this need, several strategies have been employed to enhance the multilingual capabilities of LLMs, with a significant emphasis on incorporating a higher proportion of Chinese tokens during the pretraining phase or employing techniques such as supervised fine-tuning (SFT) to activate Chinese language functionalities (Zeng et al., 2023; Bai et al., 2023; Yang et al., 2023; Team, 2023; Young et al., 2024; Bi et al., 2024). An early example in this endeavor is ChatGLM (Zeng et al., 2023), which pioneered the use of an equal distribution of Chinese and English tokens during its pretraining phase, culminating in a proficient bilingual model. Following this, models like Owen (Bai et al., 2023) have expanded the linguistic horizon by integrating multilingual data in the pretraining process, thereby achieving broader language support. Furthermore, models such as Yi (Young et al., 2024) and DeepSeek (Bi et al., 2024) have demonstrated the efficacy of meticulous SFI applications in unlocking multilingual capabilities, with a notable prowess in Chinese language reasoning. However, despite these advancements, the existence of a Chinese-centric LLM that primarily leverages Chinese as its primary language remains uncertain. This gap highlights a critical area of interest for developing localized, open-source Chinese models, underscoring the significance of tailored approaches in the evolution of language technologies.
* User's questions can refer to his/her previous (or multiple turns before) question. * User's questions can also refer to the previous (or multiple turns before) answer from the agent. * Try to use pronouns or common nouns to replace the entities mentioned before. * Try to make the user's questions diverse. For the same type of questions, try to use different ways to express it in different dialogue annotations. * Other user question types we need * Given the answer from agent, ask for extra information (e.g., What else...; any other...; is there more...). * Switch the topic and starting a new thread in conversation. * Ask two questions at the same time.
Footnote 9: We note that scores for _Unt faithful_ claims on a per-model level should be taken with a grain of salt due to the small sample size, particularly for Claude-3-Opus summaries.
but got this error: cp: 80: No such file or directory How do I specify the port number in a scp command?
Retrieval-augmented generation (RAG) is a widely used approach to incorporate new knowledge into LLMs by augmenting fixed LLMs with retrieved information from external sources (Chen et al., 2017; Guu et al., 2020; Lewis et al., 2020; Borgeaud et al., 2022; Wang et al., 2023; Alon et al., 2022; He et al., 2021; Sachan et al., 2021; Izacard et al., 2023; Lee et al., 2022; Jiang et al., 2022; Shi et al., 2023; Jiang et al., 2023; Asai et al., 2023; Nakano et al., 2021; Qin et al., 2023; Lin et al., 2023). While RAG is effective in reducing hallucinations commonly experienced when relying solely on knowledge stored in parameters, its retrieval and generation process adds extra latency and complexity. In contrast, continued pre-training to store knowledge in parameters and utilizing the stored knowledge to answer questions in a closed-book manner are simpler and faster at inference time. Enhancing this capability is also scientifically significant, as it represents a fundamental step in employing LLMs as dependable assistants for accessing information. Therefore, this paper focuses on exploring parametric approaches.
**Seeking for Advice**. Despite all our efforts, this survey is still far from perfect: we are likely to miss important references or topics, and might also have non-rigorous expressions or discussions. We will continuously update this survey, and improve the quality as much as we can. For us, survey writing is also a learning process for LLMs by ourselves. For readers with constructive suggestions to improve this survey, you are welcome to leave comments on the GitHub page of our survey or directly email our authors. We will make revisions following the received comments or suggestions in a future version, and acknowledge the readers who have contributed constructive suggestions in our survey.
Similar to prior work Xu et al. (2023); Zhao et al. (2023), we compute the total ratio of wins and ties of a target LLM against the strong LLM, to indicate how much model capacity the target LLM recovers from the strong LLM (often treated as the upper bound performer). CRR simplifies the combinatorial pairwise comparisons between all target LLMs. We name the metric as _Capacity Recovery Ratio_ (CRR), where \(\texttt{CRR}=\frac{\texttt{wins}+\texttt{ties}}{\texttt{total comparisons}}\). In experiments, we observe that the number of ties often dominates the number of wins, since the strong LLM is much capable than the target model. So we do not put additional weights on wins in the calculation. To demonstrate CRR faithfully reflects model performance, we show the exact number of wins, ties and losses in Appendix A.5 on Evol-Instruct. We would like to emphasize our focus on the gap in CRR between different methods instead of the absolute value, since the absolute value may based on the specific LLM evaluator we choose.
1. Iris Winnow's aspiration to become a renowned journalist is driven by her desire to uncover the truth about the war that resulted in the death of her mother and brother, Forest. 4.The conflict between the Skywards and the Underlings has caused chaos in the realm of Cambria, a place where magic and mythology interact with the mortal world. 5. Iris Winnow has a professional rival named Roman Kitt, a skilled storyteller from a wealthy family. 6. Roman Kitt is under pressure from his father to join the family business. 7. Zeb Autry, the editor of the Oath Gazette, does not want to cover the war, against Iris Winnow’s wishes. 8. After the loss of her mother and brother, Iris Winnow’s family is facing financial and personal struggles. 9. The Inkridation Tribune, a competing newspaper, advertises a job for war correspondents, leading both Iris Winnow and Roman Kit to compete for the position. 10. Iris Winnow maintains a mysterious correspondence with an unidentified individual who leaves her letters in her wardebro. 11. Iris Winnow’s brother, Forest, introduced her to journalism. 12. Iris Winnow and her brother Forest had differing views on the war and the Skyward goddess, Enva. 13. Despite various challenges, Iris Winnow achieves her goal of becoming a war correspondent for the Inkridden Tribune. 14. Iris Winnow embarks on a dangerous journey to cover the war. 15. Iris Winnow continues to write letters to the unknown person who becomes her confidante during her journey. 16. The town of Avalon Bluff is preparing for a battle against an unknown enemy. 17. Dacre, the god of the Underlings. 18. Amidset the chaos, Iris Winnow and Roman Kitt rekindle their connection, and Roman Kitt proposes to Iris Winnow. 19. Iris Winnow experiences the horrors of war firsthand, including the loss of innocent lives and the destruction of her childhood home, Clover Hill. 20. Despite the hardships, Iris Winnow continues to report on the war, using her writing to document the battles and search for her brother, Forest. 21. Divine Rivals explores themes of identity, courage, and the power of words during war. 22. Iris Winnow’s determination to make a difference by writing about the truth and seeking answers about her brother is a central theme in “Divine Rivals”. 23. The author of “Divine Rivals” acknowledges her literary agent, film agent, subrights team, editors, cover designer, copy editor, and fellow authors for their assistance in publishing the book.
There is further evidence for the Reversal Curse in Grosse et al. (2023), which is contemporary to our work. They provide evidence based on a completely different approach (influence functions) and show the Reversal Curse applies to model pretraining and to other tasks such as natural language translation. See Section 3 for more discussion.
To increase the diversity of document lengths, we collect five conversational QA datasets with short documents (less than 1.5K words). On average, 1 word will be tokenized into 1.5 tokens. Hence, the document can be directly fitted into LLMs with a sequence length of 4K tokens.
The AIOS SDK is designed to equip developers with a versatile toolkit for crafting sophisticated agent applications within the AIOS. This SDK encompasses a broad spectrum of functionalities, from initializing agents and managing agent lifecycles to facilitating complex operations like resource monitoring, and generation plan for an agent task. Like any operating system, enriching the SDK to be comprehensive and developer-friendly is a long-term and never-ending endeavor. The current SDK functions supported in AIOS are shown in Table 3, which will be continuously updated and expanded to fulfill the needs of evolving agent applications. This development effort aims to provide developers with the tools they need to harness the full potential of their agent applications within the AIOS framework.
We measure performance on a wide variety of downstream tasks, which can be broadly categorized as follows.
'<s> # Magnesium reacts with nitrogen to form magnesium nitride. The chemical formula for this reaction is Mg+N_2-> MgN_2.
**Closed-Book QA.** Closed-book QA tasks [652] test the acquired factual knowledge of LLMs from the pre-training corpus, where LLMs should answer the question only based on the given context without using external resources. For evaluating this ability, there are several datasets that can be leveraged, including Natural Questions [554], Web Questions [557], and TriviaQA [558], where the accuracy metric is widely adopted. Empirical results have revealed that LLMs can perform well in this setting and even match the performance of state-of-the-art open-domain QA systems [56]. Also, the performance of LLMs on closed-book QA tasks shows a scaling law pattern in terms of both model size and data size: scaling the parameters and training tokens can increase the capacity of LLMs and help them learn (or memorize) more knowledge from the pre-training data [56]. Further, under a similar parameter scale, LLMs with more pre-training data relevant to the evaluated tasks would achieve better performance [81]. Also, the closed-book QA setting provides a testbed for probing the accuracy of the factual knowledge encoded by LLMs. However, as shown in existing work [55], LLMs might perform less well on QA tasks relying on fine-grained knowledge, even when it exists in the pre-training data.
FLAN(Chung et al., 2022) is one of the most systematic and comprehensive approaches among the many studies on instruction tuning. This approach fine-tunes the language model on the instruction-optimised dataset, scales the number of tasks and model size, and incorporates chain-of-thought data in the fine-tuning. Although the authors did not consider a specific approach to tuning instructions in RALM architecture, their work provides a valuable reference for future research. In the instruction fine-tuning of RALM, Lin et al. (2023) integrated in-context retrieval augmentation. This greatly reduces the likelihood of the language model being misled by irrelevant retrieval content. SAIL(Luo et al., 2023) builds language generation and instruction tracking capabilities on complex search results generated by internal and external search engines. Using a corpus of instruction tuning, they collect search results for each training case from different search APIs and domains, and construct a search-based training set containing a triplet of (instruction, grounding information, response). In contrast to training on instruction-tuned datasets, Madaan et al. (2022) and Lazaridou et al. (2022) propose to prompt large models directly from retrieved knowledge. Madaan et al. (2022) used GPT-3 to clarify memory pairings of recorded cases where the model misinterpreted the user's intention, as well as user feedback. This ensures that their system can generate enhanced prompts for each new query based on user feedback. In contrast, Lazaridou et al. (2022) uses few-shot prompts and answer reordering to improve inference computation.
Considering that \(0\leq\pi_{w}(y_{l}|x)\leq 1\), the loss can be upper bounded as follows:
Recent studies have proposed a mixture of synthetic data and real human data to train more robust reward models (Gao et al., 2023). Constitutional AI (Bai et al., 2022) proposes to use a small set of principles to steer the AI generated critiques and feedback, and use such synthetic data to replace the real human data in the typical RLHF pipeline. The model trained with this RLAIF (i.e., reinforcement learning from AI feedback) method shows similar strong performance as RLHF baselines. In general, synthetic data offers a powerful solution for human values and preferences alignment by allowing researchers to generate large-scale, diverse, and controlled training datasets in a low-cost way (Cui et al., 2023; Ganguli et al., 2022). By simulating a wide range of scenarios involving ethical dilemmas (Perez et al., 2022), social interactions (Liu et al., 2023c), and cultural norms (Ziems et al., 2023), synthetic data enables comprehensive and systematic testing of AI models' alignment with human values (Askell et al., 2021). This approach helps identify and mitigate issues related to bias (Liu et al., 2021; Ntoutsi et al., 2020), fairness (Landers and Behrend, 2023; Zhao et al., 2018), and unintended consequences before AI systems are deployed in real-world settings (Ye et al., 2024).
I want you act as a Prompt Rewriter. Your objective is to rewrite a given prompt into a more complex version to make those famous AI systems (e.g., ChatGPT and GPT4) a bit harder to handle. But the rewritten prompt must be reasonable and must be understood and responded by humans. Your rewriting cannot omit the non-text parts such as the table and code in #Given Prompt#:. Also, please do not omit the input in #Given Prompt#. You SHOUTLD complicate the given prompt using the following method: If #Given Prompt# can be solved with just a few simple thinking processes, you can rewrite it to explicitly request multiple-step reasoning.
These models are usually composed of a neural network \(f_{\theta}\) that takes sequences of tokens \((y_{<i})\in[1,V]^{i-1}\) as inputs and produces a relatively low-dimensional contextual representation in \(\mathbb{R}^{d}\), where \(d\) is the _hidden dimension_ of the model. They then rely on a _language modeling head_ that produces logits for contextual token probabilities. A common choice for the language modeling head is a linear layer with parameter \(W\in\mathbb{R}^{V\times d}\), where \(V\) is the number of possible tokens.
The TF-IDF algorithm utilises term frequency (TF) and inverse document frequency (IDF) to represent relevance, which has the advantages of simplicity and speed, and even if the corpus is unchanged, the TF-IDF value for each word can be computed in advance. In RALM, Lazaridou et al. (2022) utilise the TF-IDF algorithm to match information obtained from user queries and calls to the Google search API. Hua and Wang (2018) also employ the algorithm to score the generated results. The BM25 represents an enhancement over the TF-IDF. It considers the user's query and calculates the relevance score as the weighted sum of the relevance of each query word to the document. The IDF algorithm is used to derive the weight of each word, but it is improved by two moderating factors to prevent the strength of the influence of a certain factor from being infinite.
In the paper, we adopt a new approach for highlighting improvements within tables, moving beyond the standard practice of specifying a static improvement threshold in metric \(y\) by score \(x\). Instead, our threshold is dynamic, calibrated to the minimal metric difference \(x\) in metric \(y\) that yields a perceptible distinction between two systems as recognized by humans (Kocmi et al., 2024). For instance, to align with human judgments at an 80% concordance rate, the required improvement margin is \(\geq 1.24\) for both KWI-XXL and COMET-XXL, and \(\geq 0.53\) for KWI-22. A comprehensive delineation of these thresholds can be found in Table 13.
However, the language model does not generate the plan entirely on its own. Instead, it generates goals and steps to work aside a classical planner which executes the plan using Planning Domain Definition Language (PDDL). The paper found that "LLMs currently lack the ability to directly translate a natural language instruction into a plan for executing robotic tasks, primarily due to their constrained reasoning capabilities" [1]. By combining the LLM planning capabilities with a classical planner, their approach significantly improves upon other purely language model-based approaches to robotic planning.
Figure 3: Bits per byte for each of the 7 remaining Paloma data sources not aggregated in Figure 2.
In Figure 0(b), we fit a scaling law in the style of Hoffmann et al. (2022) on data points from models ranging from 410M parameters, only optimizing for model-related constants (\(A\) and \(\alpha\)) while reusing all other values (\(B=410.7\), \(\beta=0.28\), \(E=1.69\)). We recall the relation between parameter count \(N\) and token count \(T\) given in Hoffmann et al.
World knowledge is mostly about general knowledge questions, for example, in Wikifact dataset questions such as "Who is the author of a specific well-known book" can be found and references are also provided. Table VII shows the results.
Although it is expedient to leverage an existing tokenizer (_e.g.,_ OPT [90] and GPT-3 [55] utilize the tokenizer of GPT-2 [26]), using a tokenizer specially designed for the pre-training corpus can be highly beneficial [78], especially for the corpus that consists of diverse domains, languages, and formats. Therefore, recent LLMs often train the customized tokenizers specially for the pre-training corpus with the SentencePiece library [226], which includes Byte-level BPE and Unigram tokenization. A note is that normalization techniques in BPE, such as NFKC [227], may degrade the tokenization performance [34, 64, 78]. When extending existing LLMs (_i.e.,_ continual pre-training or instruction tuning), we should be also aware of the potential side effect with customized tokenizers. For example, LLaMA trains the BPE tokenizer based on a pre-training corpus mainly consisting of English texts, and the derived vocabulary might be less capable in processing non-English data, _e.g.,_ taking longer inference latency to generate Chinese texts.
15. This revelation comes after Forest saves Iris from a gas attack by Dacre and then deceives her by leaving Roman behind.
**Output** [[Retriev=Yes] <p>Reference [1]<p>I [IsRel] =Relevant] A Question and Answer Guide to Astronomy is a non-fiction book that falls under the genre of Astronomy and Cosmology.
Figure 3: **Parallel Generation**. By constructing an attention mask that allows all thought tokens to pay attention to themselves, all preceding thought tokens within the same thought, and the preceding text, we can generate continuations of all of the thoughts in parallel. Each inference call is used to generate one additional thought token for all text tokens.
Our goal is to facilitate scientific development and empower the scientific community, so we favor permissive licenses that give users flexibility in using our resources and artifacts. As such, all code and weights are released under the Apache 2.0 License.27 Some licenses used by other organizations for recent model releases prohibit using the outputs from their models to train artificial intelligence or machine learning systems, while we expressly allow users to do so. We also do not limit commercial use. We hope that our models can make other models better. We recognize that the risk for misuse of our models is relatively low since they are mainly designed as scientific artifacts not as products with broad public adoption (our models have not been adapted as chatbots). In addition, over the past year there have been a number of comparable models released with very permissive licenses, so using a more strict license for our work will not remove the overall risk in the field. We believe this tradeoff on the side of being more open is the best option.
In this section, we apply our method for inducing instruction data to GPT3 as a case study. We use the largest GPT3 LM ("davinci" engine) accessed through the OpenAI API.6 The parameters for making queries are described in Appendix A.2. Here we present an overview of the generated data.
Cheng et al. (2024) present Selfmem, a model designed to tackle the issue of low corpus quality. Selfmem utilises a retrieval-enhanced generator to create an infinite pool of memory, which is then used by a memory selector to choose an output for subsequent generations. This approach enables the model to use its own output to enhance generation. Peng et al. (2023) propose an AI agent that formulates human system dialogue as a Markov Decision Process (MDP) described by a quintuple. The quintuple includes an infinitely large set of dialogue states, a collection of historical behaviours, a probability of state transfer, external rewards obtained, and a variable parameter.
The proofs are straightforward and we defer them to Appendix A.5. The first lemma is a well-known under-specification issue with the Plackett-Luce family of models [30]. Due to this under-specification,we usually have to impose additional identifiability constraints to achieve any guarantees on the MLE estimates from Eq. 2[4]. The second lemma states that all reward functions from the same class yield the same optimal policy, hence for our final objective, we are only interested in recovering an arbitrary reward function from the optimal class.
While creating our tokenizer and model training datasets, we ensure that the mixture proportions of data domains (Wikipedia, books, web text) follow the same distribution to avoid a domain shift between tokenizers training and model training. We created _two datasets_ with 70B words where one of the datasets is monolingual, containing English documents, and the second is a multilingual dataset comprised of English, German, French, Italian, and Spanish documents. Our datasets are filtered and deduplicated and consist of web-crawled data (80%) and curated data (20%), comparable to related datasets used to train LLMs. In the multilingual dataset, the amount of web-crawled data is equally distributed across languages in terms of number of words. Further details about our data pipeline and the data composition are described in Appendix A.
Fig. 34: An illustration of LoRA reparametrizan. Only \(A\) and \(B\) trained during this process. Courtesy of [142].
The stock rose $2.11, or about 11 percent, to close Friday at $21.51 on the New York Stock Exchange.
During our investigations, we found a significant limitation in the original prompts of LegalBench. The complex nature of these prompts, combined with the challenges encountered by open source LLMs in adhering to instructions - particularly in handling formatting - leads to a substantial drop in performance (as measured by accuracy). The generated sentences are often verbose and difficult to parse, rendering LegalBench in its current form too stringent and failing to accurately gauge improvement on the task.
Among the various datasets offering factual QA (Christmann et al., 2023; Sciavolino et al., 2021), we selected one of the most recent ones, named PopQA (Mallen et al., 2023). PopQA dataset comprises approximately 14,000 templated questions aimed at eliciting single-entity answers, based on 16 types of relationships extracted from Wikipedia. To assess the popularity of each entity, we followed the previous studies (Mallen et al., 2023; Sciavolino et al., 2021; Chen et al., 2021), using Wikipedia pageviews as a metric for popularity. We accumulated the pageviews for each entity from the start to the end of 2023.