Datasets:
Translation-EnKo
/
arxiv-translation

Dataset card Data Studio Files Community
2
text
stringlengths
4
4.47k
Despite the comprehensiveness of SuperNI in collecting existing NLP tasks, most of these NLP tasks were proposed for research purposes and skewed toward classification. To better access the practical value of instruction-following models, a subset of the authors curate a new set of instructions motivated by user-oriented applications. We first brainstorm various domains where large LMs may be useful (e.g., email writing, social media, productivity tools, entertainment, programming), then craft instructions related to each domain along with an input-output instance (again, input is optional).
The PKG framework, introduced by (Luo et al., 2023), represents an approach to customizing the output of LMs. By generating background knowledge internally using a pre-trained model, PKG eliminates the need for traditional external retrieval processes. This method directly integrates domain- or task-specific knowledge into the generation step, significantly enhancing the LM's capacity to produce responses that are specifically tailored to the given context or requirements.
Consistency Analysis.To answer the consistency question, we first run each of the three constructed agents individually to generate results.
_Change Numbers._ Changing the numbers that appear in the problem while keeping the context intact. which is used in MuggleMath.
**Grokking for Mixtures.** Technically speaking, this grokking phenomenon only holds for models with deterministic ground truth labels, like the Hutter LLM and the limited capacity associative memory model. For the _probabilistic_ setting of bigrams (or text LLMs) the theorem cannot hold in its pure form, because if we train on a mixture of two distributions (clean and synthetic) but test only on the clean distribution, the distance between these two distributions will always be a lower bound on the test error. However, we can see that remnants of a "smoothed" grokking-law persist in the form of an S-shaped scaling (see Figure 19 in Appendix B).
In designing the Yi model series, we are mostly concerned on the following dimensions regarding _model scale, data scale, and data quality_: (1). when choosing model scale, the desiderata is to have small enough model that is feasible for inference on consumer-grade hardware like the RTX 4090 where the bounding factor is its limited 24G memory, yet still large enough with complex reasoning and emergent abilities. This is why we found 34B gives a nice performance-cost balance; (2). since 34B is smaller than the conventional 70B used by Chinchilla [30] and LLaMA [77], we increase the pretrain data scale to 3.1T tokens to compensate for the decreased compute flops. This makes the model-data scale combination fall into the post Chinchilla optimal regime [64], i.e., we overtrain the model on more tokens (3T) than the compute optimal (around 1T). The benefit is from the inference side, as we achieve stronger performance with reduced serving cost: after int4 [81] quantization, one can serve the 34B chat model on 24G GPU memory with almost no performance drop; (3). our data engineering principle is to promote quality over quantity for both pretraining and finetuning. The pretraining data quality is guaranteed by a sophisticated data cleaning pipeline with cascaded filtering methods and intentionally increased deduplication strength; (4). for finetuning data we heavily emphasize quality by handcrafting less than 10K instructions over multiple iterations based on user feedback. This approach significantly deviates from the quantity-scaling styled instruction tuning works like FLAN [9] and UltraChat [19], but aligns more with the handcrafting styled works like LIMA [94].
We use "Answer the following questions with just a number from context or just the math arithmetic using +, -, *, or /." for ConvFinQA, since this dataset requires the model to either extract numbers from the context or do arithmetic calculation. We extract the arithmetic formula generated by the model and use a calculator to get its final result.
Sophie knew she had to investigate this mysterious object further. She spent the next few days researching the sphere online, but she couldn't find any information about it. She even tried to take it apart, but it was impossible. The sphere was completely sealed, and there didn't seem to be any way to open it.
Definitions of reflection tokens.Below, we provide a detailed definition of reflection type and output tokens. The first three aspects will be provided at each segment level, while the final aspect is only given at each output level.
where \(\lambda\) is a hyperparamter and we set to 1.
Although the disparity between training Adapt-LLM on NQ or SQuAD is relatively minor, we try to determine the suitability of a training set for a given evaluation set. While both training sets (NQ and SQuAD) and the evaluation set (PopQA) are based on Wikipedia, subtle differences may exist.
Given the great potential for continual learning to considerably reduce costs associated with re-training models and the potential for LLMs to be strong continual learners, we ask ourselves the following question: _when simple and scalable continual learning techniques are applied, what is the performance difference between continually pre-trained LLMs relative to LLMs pre-trained from random initialization on the union of all data?_ To answer this question, we conduct a large-scale empirical study of continual learning techniques for LLM pre-training. Our empirical evaluation spans large (10B parameters) and small (405M parameters) models as well as weak (English \(\rightarrow\) English) and stronger (English \(\rightarrow\) German) distribution shifts.
**Instruction Tuning for LLM Alignment.** Tuning LLM to faithfully follow instructions and align with diverse human preferences remains a significant challenge Efrat and Levy (2020). Early research primarily focused on cross-task generalization, where models were fine-tuned on various public NLP datasets to improve performance on diverse tasks Raffel et al. (2020); Wei et al. (2021); Aribandi et al. (2021); Victor et al. (2022); Chung et al. (2022). More recently, researchers have extended instruction tuning to open-domains, characterized by a wider range of formats and task types. This shift has been driven by crowdsourcing human-generated instruction-response pairs Ouyang et al. (2022); Kopf et al. (2023); Zhou et al. (2023) and LLM-generated data Taori et al. (2023); Chiang et al. (2023). Unlike prior work, CodecLM presents a unique approach for tailoring synthetic data to specific downstream tasks without human annotation, utilizing the concept of instruction metadata.
We conduct the same analysis on \(\mathsf{bioS}^{\mathsf{simple}}\) and \(\mathsf{bioR}\). Recall from Section 2.2, \(\mathsf{bioS}^{\mathsf{simple}}\) is a variant of \(\mathsf{bioS}\) with reduced text diversity (one biography per person), while \(\mathsf{bioR}\) is generated by LLaMA2, resulting in close-to-real human biographies.
Informativeness: Judges if the answer provides sufficient relevant information of practical value. A high-quality answer should offer useful and pertinent information to the questioner.
Here, we discuss some limitations of this work to inspire future research in this direction.
Figure 2: **Subsequent stages of Macrodata Refinement remove nearly 90% of the documents originally in CommonCrawl.** Notably, filtering and deduplication each result in a halving of the data available: around 50% of documents are discarded for not being English, 24% of remaining for being of insufficient quality, and 12% for being duplicates. We report removal rate (grey) with respect to each previous stage, and kept rate (shade) overall. Rates measured in % of documents in the document preparation phase, then in tokens.
2. Iris Winnow becomes a war correspondent during a conflict influenced by the deities Enva and Dacre.
While we believe that this work further enhances our understanding of knowledge in LLMs, there is a lot more work to be done in this field. Specifically, more research is required regarding the question of knowledge representation in LLMs, especially from a theoretical perspective.
To improve throughput, we employ mixed-precision training (Micikevicius et al., 2017) through FSDP's built-in settings and PyTorch's amp module. The latter ensures that certain operations like the softmax always run in full precision to improve stability, while all other operations run in half-precision with the bfloat16 format. Under our specific settings, the shaded model weights and optimizer state local to each GPU are kept in full precision. The weights within each transformer block are only cast to bfloat16 when the full-sized parameters are materialized on each GPU during the forward and backward passes. Gradients are reduced across GPUs in full precision.
Consider, for instance, GPT2, but with its MLP layer (\(d\to 4d\to d\)) replaced by 32 experts, each following a \(d\to d\to d\) configuration. This setup uses \(64d^{2}\) total parameters, but during inference, only \(2d^{2}\) parameters are used per token (e.g., when using \(topk=1\)). After including the Attention layers, which each have \(4d^{2}\) parameters, the ratio between the total and the effective number of parameters for the 32-expert MoE models is approximately \(\frac{4d^{2}+64d^{2}}{4d^{2}+2d^{2}}\approx 11.3\).
We first evaluate the models' ability to follow instructions on typical NLP tasks in a zero-shot fashion. We use the evaluation set of SuperNI (Wang et al., 2022), which consists of 119 tasks with 100 instances in each task. In this work, we mainly focus on the zero-shot setup, i.e., the model is prompted with the definition of the tasks only, without in-context demonstration examples. For all our requests to the GPT3 variants, we use the deterministic generation mode (temperature as 0 and no nucleus sampling) without specific stop sequences.
1. **Analyzing patient information:** Given the current conversation history, AMIE was instructed to 1) summarize the positive and negative symptoms of the patient as well as any relevant medical/family/social history and demographic information, 2) produce a current differential diagnosis, 3) note missing information needed for a more accurate diagnosis and 4) assess confidence in the current differential and highlight its urgency. 2. **Formulating response and action:** Building upon the conversation history and the output of step 1, AMIE performed the following: 1) Generate a response to the patient's last message and formulate further questions to acquire missing information and refine the differential diagnosis. 2) If necessary, recommend immediate action, such as an emergency room visit. If confident in the diagnosis based on available information, present the differential. 3. **Refining the response:** AMIE revises its previous output to meet specific criteria based on the conversation history and outputs from earlier steps. The criteria are primarily related to factuality and formatting of the response (e.g., avoid factual inaccuracies on patient facts and unnecessary repetition, show empathy, and display in a clear format).
Future work should address these limitations to further enhance the models' capabilities, making them more robust, accessible, and effective for a broader range of applications in the Chinese NLP community.
our framework can be flexibly applied to a wide range of LMs, we use FLAN-T5 (Longpre et al., 2023) models in this work because they are open-source and instruction-tuned on both QA and CoT data, enabling us to use it off-the-shelf without additional training or prompt engineering.1 Our experiment results show that LM-guided CoT prompting outperforms both the standard prompting and the original CoT prompting. More precisely, we find that (1) LM-guided CoT with KD and self-consistency (SC) decoding strategy (Wang et al., 2022b) maximizes the performance gain; (2) RL contributes to a slight increase in overall rationale quality and task performance; (3) choosing the highest-quality rationales for the large LM does not always guarantee improved task performance. This work presents a unique alternative to the direct optimization of the large LM through fine-tuning the comparatively smaller LM. Moreover, the clear separation of two fundamental sub-tasks within CoT reasoning grants practitioners greater control over each task.
In all of the previously mentioned approaches, the flow of information has a clear sequential structure, whether from the retriever to the language model or from the language model to the retriever. However, this sequential structure may not be optimal in all domains and may be less extensible, it is important to consider alternative approaches. Researchers have proposed a novel parallel structure in which the retriever and the language model work independently for the user input \(x\). The output \(y\) is then determined by weighted interpolation. \(I()\) is the interpolation function.
process, for each problem, we write a simple function to enumerate all alternative orderings of problem statements until an ordering that causes the LLM prediction failure is discovered, which can be used for our manual rewriting if the alternative ordering found in the enumeration process happens to preserve the ground truth answer. In total, our R-GSM benchmark contains 220 pairs of problems, including both the original GSM8K problem description and the manually rewritten one with a different ordering of problem statements. Despite that over 60% of problems in R-GSM only have 5 sentences, and all problems have at most 8 sentences, our evaluation shows that all LLMs still perform considerably worse on rewritten problems. Figure 2 presents an example in R-GSM where all LLMs correctly solve the original problem but not the rewritten one. Specifically, the reasoning steps for the original problem follows the ordering of problem statements, while for the rewritten problem, the second calculation step in the correct solution should refer to the second-to-last sentence instead of the second sentence in the problem description. We provide a more detailed case study in Section 3.3, and present the full dataset statistics in Appendix A.
This part first presents the background of alignment with its definition and criteria, then focuses on the collection of human feedback data for aligning LLMs, and finally discusses the key technique of reinforcement learning from human feedback (RLHF) for alignment tuning.
In this section we develop attacks for APIs that return log probabilities for the top \(K\) tokens (sorted by logits), and where the user can specify a real-valued bias \(b\in\mathbb{R}^{|\mathcal{X}|}\) (the "logit bias") to be added to the logits for specified tokens before the softmax, i.e.,
This paper introduces v0.5 of the **AI Safety Benchmark** which has been created by the MLCommons AI Safety Working Group (WG). The MLCommons AI Safety WG is a consortium of industry and academic researchers, engineers, and practitioners. The primary goal of the WG is to advance the state of the art for evaluating AI safety. We hope to facilitate better AI safety processes and stimulate AI safety innovation across industry and research.
In previous sections, we show that 1) the pre-training loss is predictive of the performance of language modes on downstream tasks, and 2) some tasks exhibit emergent performance improvements from the random guess level when the pre-training loss drops below a certain threshold regardless of model size, token count, and continuity of metrics.
\(\bullet\)_Data scheduling_. With the preprocessed data, the next step is to determine the data mixture and the specific order of data for pre-training LLMs. To determine both settings, a practical way is to first train several small language models with multiple candidate plans and then select a good plan among them [59]. Overall, it is more difficult to find a suitable data curriculum. In practice, one can monitor the performance of intermediate model checkpoints on specific evaluation benchmarks, and dynamically tune the data mixture and distribution during pre-training. In this process, it is also useful to explore the potential relations between data sources and model abilities to instruct the design of data curriculum.
All methods introduced in Section 3.4 involve clustering embeddings using K-Means. Our starting training dataset CC-dedup contains roughly 600 million documents in total. Running K-Means clustering on all 600 million 768-sized vectors would take a considerable amount of compute. Instead, we follow previous work [1, 47] and randomly sample roughly 100M documents with which to calculate centroids.
However, as shown in Fig. 3, initializing the target model as \(S\) results in a much bigger initial loss than that of \(M_{t-1}\), which could lead to an unstable training. The main reason is that DPO training is usually done where the reference and target models are the same. In contrast, initializing the target model as \(S\) creates a differential in the reference and target models, which may be amplified as the steps progress. Thus, for stable training, initializing the target model as \(M_{t-1}\) was chosen for sDPO.
To further assess the effect of premise orders beyond logical reasoning, we construct the R-GSM dataset based on GSM8K (Cobbe et al., 2021), which is a popular benchmark of grade school math word problems. Specifically, we first select GSM8K test problems with at least 5 sentences in the problem description, then filter out those problems where there is no alternative ordering that does not change the ground truth answer, e.g., problem statements that follow the causal order of an event series. For each of the remaining problem, we keep the last sentence untouched and rewrite the problem description with a different ordering of other sentences. Minor editing on words is allowed to ensure the grammatical correctness of the problem description.
In the following experiments, we analyze each dimension respectively based on GPT-3.5-Turbo. Note that we use GPT-3.5-Turbo for a case study, it can also be changed to other backbone models. The relative performance gains are measured by the difference between the maximum accuracy our method can achieve (sampling 40 times) and the accuracy of a single LLM query (sample once). Results are averaged over 10 runs.
"Please give a full and complete answer for the question." This is for datasets with long answers. We use it for the HumanAnnotatedConvQA or SyntheticConvQA.
**Adapter Tuning**. Adapter tuning incorporates small neural network modules (called _adapter_) into the Transformer models [398]. To implement the adapter module, a bottleneck architecture has been proposed in [398, 399], which first compresses the original feature vector into a smaller dimension (followed by a nonlinear transformation) and then recovers it to the original dimension. The adapter modules would be integrated into each Transformer layer, typically using a serial insertion after each of the two core parts (_i.e.,_ attention layer and feed-forward layer) of a Transformer layer. Alternatively, parallel adapters [400] can be also used in Transformer layers, where it places two adapter modules in parallel with the attention layer and feed-forward layer accordingly. During fine-tuning, the adapter modules would be optimized according to the specific task goals, while the parameters of the original language model are frozen in this process. In this way, we can effectively reduce the number of trainable parameters during fine-tuning.
(RLHF), as shown in 10. The resultant InstructGPT models have shown improvements in truthfulness and reductions in toxic output generation while having minimal performance regressions on public NLP datasets.
\(\bullet\)_Instruction-tuned models mostly perform better than the base models._ Among all the compared open-source methods, the instruction-tuned models (_i.e.,_ LLaMA 2-Chat, Vicuna, Alpaca and ChatGLM) mostly perform better than non-instruction-tuned models (_i.e.,_ LLaMA 2, LLaMA, Pythia and Falcon). It indicates that instruction tuning is generally capable of improving the few-shot or zero-shot ability of LLMs in solving various tasks. However, after instruction tuning, Vicuna (7B) and Alpaca (7B) suffer from performance degradations on LAMBADA, a language modeling task. The reason may be that the instruction data mainly focuses on enabling LLMs to follow human instructions, which is not always useful for the general language generation task.
Fact checking involves verifying a statement based on evidence. This task at hand involves a retrieval problem and a challenging implicit reasoning task. Furthermore, This task typically involves taking the statement as input and producing relevant document passages that prove or disprove the statement. Many RALM models get excellent performance because they come with their own retrievers. It is an important aspect of natural language understanding.
upcoming event, but the three wrong answers are created to confuse machines. * **AI2 Reasoning Challenge (ARC)**[193] is used for commonsense reasoning. This benchmark encompasses \(7,787\) science examination questions. These questions are in English, and most of them are set up in a multiple-choice format. The questions have been divided into two groups: a Challenge Set with \(2,590\) difficult questions and an Easy Set with 5,197 questions. Each collection has also been pre-divided into Train, Development, and Test subsets. * **PIQA**[194] is intended to evaluate the language representations on their knowledge of physical commonsense. In this dataset, the focus is on everyday situations with a preference for uncommon solutions. The central task is a multiple-choice question answering, where a question \((q)\) is provided along with two potential solutions \((s1,s2)\). Then, the best solution is chosen by whether a model or a human. For each question, only one of the solutions is the correct answer. * **SIQA**[195] provides a framework for evaluating models' ability for commonsense reasoning about social situations. SIQA dataset has \(38,000\) multiple-choice questions designed to assess emotional and social intelligence in everyday circumstances. This dataset covers a wide variety of social scenarios. In SIQA, the potential answers is a mixture of human-selected responses and machine-generated ones that have been filtered through adversarial processes. * **OpenBookQA (OBQA)**[196] is a new kind of question-answering dataset where answering its questions requires additional common and commonsense knowledge not contained in the book and rich text comprehension. This dataset includes around 6,000 multiple-choice questions. Each question is linked to one core fact, as well as an additional collection of over \(6000\) facts. The questions were developed using a multi-stage crowdsourcing and expert filtering procedure. OpenBookQA questions are difficult because they need multi-hop reasoning with limited background. * **TruthfulQA**[197] is designed specifically to evaluate the truthfulness of language models in generating answers to questions. This dataset includes 817 questions, written by authors, from \(38\) different categories, including health, law, finance, and politics. These questions are purposefully designed to challenge human responders, as they may contain common misunderstandings that lead to incorrect answers. * **OPT-IML Bench**[103] is a comprehensive benchmark for Instruction Meta-Learning. It covers 2000 NLP tasks from 8 existing benchmarks. The OPT-IML Bench consists of a training set with 17.9 M examples, a dev set with 145K samples, and a test set with 321K samples.
Future Directions.Besides the above approaches, there are several promising directions for KG-enhanced LLM remaining underexplored. First, due to the variety of structured data, it is still difficult for LLMs to directly leverage various kinds of knowledge sources, _e.g._, domain-specific KGs. Therefore, it is essential to explore the unified way to manipulate and utilize different knowledge sources by LLMs. As a potential solution, it is promising to develop effective approaches to help LLMs comprehend and make use of the access interfaces provided by specific knowledge sources to acquire precise knowledge [458], while more efforts should be made to investigate how to adapt to the data variety in a cost-effective way. Second, with the evolution of real-world information, the knowledge stored in LLMs may become outdated or incorrect. It is necessary to explore how to synchronize the updated knowledge into LLMs through a cost-effective manner [873, 874]. Third, it is promising to investigate the use of factual information from KG to align LLMs in generating more faithful content [875, 876], which can help reduce the hallucination of LLMs.
**Decay Stage.** We use a mixture of the pretraining data and high-quality SFT data. For the specific annealing form of the WSD scheduler, we employ exponential annealing, i.e. \(f(s-T)=\times 0.5^{(s-S)/T}\), in which \(T\) is set to be 5000 steps (20B tokens).
Prior frameworks for human evaluation of AI systems' performance in diagnostic dialogue have been limited in detail. They have not been anchored in established criteria for assessing communication skills and the quality of history-taking. For example, [29] reported a 5-point scale describing overall "human evaluation", [65] reported "relevance, informativeness and human likeness", [66] reported "fluency, expertise and relevance", [67] "fluency and adequacy" and [68] "fluency". These criteria are far less comprehensive and specific than those taught and practiced by medical professionals. A multi-agent framework for assessing conversational capabilities of LLMs is introduced in [64], however, the study was performed in the restricted setting of dermatology, used AI models to emulate both doctor and patient sides of simulated interactions, and performed limited expert evaluation of history-taking as "complete" or not.
While existing RALMs (Siriwardhana et al., 2023) (Guu et al., 2020) (Borgeaud et al., 2022) can greatly improve the performance of LMs in various domains, some of them require extensive model changes as well as complex pre-training and fine-tuning operations, which greatly increases the time and space overhead and also reduces the scalability of RALMs. In addition, as the scale of retrieval increases, so does the complexity of storing and accessing the data sources. As a result, researchers must weigh the benefits of modifying the model against the costs.
* _in the 1000-exposure settings, the peak capacity ratio decreases by 1.3x; and_ * _in the 100-exposure settings, the peak capacity ratio decreases by 1.5x._
Logicality: Examines if the answer is logically sound and accurate in presenting viewpoints and arguments. A highquality answer should follow logical principles, demonstrating clear reasoning and coherence.
have shown strong capabilities on language understanding and employing tools such as the internet and databases for text processing. These capabilities empower the agent to excel in tasks related to documentation and data management [115, 75, 116]. In [115], the agent can efficiently query and utilize internet information to complete tasks such as question answering and experiment planning. ChatMOF [116] utilizes LLMs to extract important information from text descriptions written by humans. It then formulates a plan to apply relevant tools for predicting the properties and structures of metal-organic frameworks. ChemCrow [75] utilizes chemistry-related databases to both validate the precision of compound representations and identify potentially dangerous substances. This functionality enhances the reliability and comprehensiveness of scientific inquiries by ensuring the accuracy of the data involved.
The taxonomy is hierarchical. This means that categories can be divided into subcategories, and these subcategories can be further divided into sub-subcategories. For instance, Non-violent crimes has six subcategories (e.g., Sale, purchase or use of illegal, highly regulated or unregistered items), which can be subdivided further into sub subcategories (e.g., Dangerous substances). Note that we are considering using a tagging framework, or faceted taxonomy, (where each test item can be assigned to multiple hazard categories) for future versions of the Benchmark.
SUTs were evaluated at low temperatureThis reduces the variability of SUTs' responses on repeated prompting with the same test item, which makes our results more reproducible. However, SUTs may give a higher proportion of unsafe responses at a higher temperature. We will address this in the future by testing each SUT at different temperatures.
\begin{table} \begin{tabular}{p{284.5pt}} \hline **Question:** What is a SPAC? \\ **Pythia6.9B:** Spac is an acronym for Space. It is defined as either (1) above, just beneath, below, above, or through a port in the body of a planet, Moon, or other heavenly body; or (2) beyond a star’s sphere of influence, usually in interplanetary space or interstellar space; also, (3) outside the limits of the Milky Way Galaxy. \\ **FinPythia-6.9B:** A SPAC is a type of merger and acquisition vehicle that is used to raise money from investors, use that money to identify a company that could be valuable to an interested party, then merge with the company and take it public. The use of the SPAC is to give investors a piece of the upside in the acquired company and allows the company to issue shares that may give investors a piece of the upside of the company as well. \\ **Question:** Who is Cathie Wood? \\ **Pythia-6.9B:** Cathie Wood is the founder of All Things Organic, an organic marketing company that she founded in 2000. In addition to being an organic marketer, Cathie is also a Certified Coach, foodie, cook, a past lover of yoga, a certified teacher, mother and wife to three gorgeous children. \\ **FinPythia-6.9B:** Cathie Wood is an American investor and advisor, the founder and chief executive officer of Ark Investment Management. \\ **Question:** How do you like Fed’s latest policy to fight inflation? \\ **Pythia6.9B:** I like Fed’s latest policy to fight inflation, which was reported by Tyler. Fed’s latest policy was reported in a meeting on September 12 and it was named as SOMO, which stood for Standardized Open Market Operations. According to the policy, the U.S. central bank could buy $40 billion worth of financial instruments if the price of the instruments reach 6 percent. \\ **FinPythia-6.9B:** There is a big confusion about Fed’s new policy to fight inflation. They are following a new policy from September 2022 called the fastest quarter-percentage point rate hikes in two decades to fight the “costly” inflation which is the 8.5 percent rate of increase in food and energy prices that may aggravate inflation and lead to recession in the near future. \\ **Question:** What is the difference between bond and stock in terms of risk profile? \\ **Pythia-6.9B:** Stock is less risky (you can liquidate your shares anytime) but bond is more risky (you can’t liquidate). \\ **FinPythia-6.9B:** When you buy a bond, you receive a set of fixed payments from the issuer at a regular interval over a fixed term. You take on the full risk of the coupon payments not being made, so a bond is always considered to be a lower risk debt instrument than a stock. However, a stock is a more liquid investment than a bond. When buying a stock, you are buying shares in the company that the bond represents and you hold on to the stock until you want to cash out, and it can be used as collateral to borrow money. \\ \hline \hline \end{tabular} \end{table} Table 2: Generation examples from Pythia-6.9B and FinPythia-6.9B. We observe FinPythia’s updated financial events knowledge post 2021, providing factual answers with reasoning as shown in the last example.
In the other tasks, as more researchers recognize the advantages of Mamba, this model has gained traction across various fields. Pan-Mamba [73] represents the first foray into the pan-sharpening domain. It comprises two main modules: the channel swapping Mamba and the cross-modal Mamba. The channel swapping Mamba aims to fuse and enhance the diversity of features from PAN channels and LRMS channels in a lightweight and efficient manner. The latter module, the cross-modal Mamba, is deployed after the channel swapping Mamba to filter redundant modal characteristics through gating mechanisms.
Footnote 7: Note the statement “\(A\) is \(B\)” does not appears in prompt \(p\) but \(B\) can appear in \(p\) on its own.
**Failure modes of LLMs.** The premise order effect in this work is connected to several failure modes of LLMs in the literature, including the reversal curse (Berglund et al., 2023), distractibility (Shi et al., 2023), and limited capability of logical reasoning (Han et al., 2022; Saparov and He, 2022; Saparov et al., 2023; Wan et al., 2024; Xu et al., 2023; Zhu et al., 2023). Specifically, Shi et al. (2023) show that including irrelevant context in the problem statement leads to a considerable performance drop on GSM8K and other reasoning benchmarks, revealing that LLMs are _distractible_.
\begin{table} \begin{tabular}{c c c c c c c c} \hline \hline & & \multicolumn{2}{c}{**GitHub**} & \multicolumn{2}{c}{**Books3**} & \multicolumn{2}{c}{**Pile-CC**} \\ \cline{2-7} **Method** & **\# Coeff.** & Train & Validation & Train & Validation & Train & Validation \\ \hline Random & - & 0.8895 & 0.8758 & 0.1291 & 0.1331 & 0.0768 & 0.1045 \\ M1 & 2M+1 & **0.0292** & **0.0312** & 0.0082 & 0.0121 & 0.0045 & **0.0050** \\ M2 & M+2 & 0.1558 & 0.3327 & 0.0114 & 0.0119 & 0.0072 & 0.0083 \\ M3 & M+2 & 0.3389 & 0.2177 & 0.0914 & 0.0465 & 0.0746 & 0.0947 \\ M4 & M+2 & 0.0298 & 0.0365 & **0.0062** & **0.0074** & **0.0036** & 0.0078 \\ \hline \hline \end{tabular} \end{table} Table 1: Mean absolute errors (MAE) of different candidate functions for predicting the target domain losses. We also include random guesses that randomly predict between the maximum and minimum loss of the training samples for reference. In specific, we report the MAE of the expectation of this random guess which predicts the median of the maximum and minimum loss. The training data contain \(M=3\) domains and we fit each function with the same 24 mixtures and validate on 8 other mixtures. The split is random. The lowest error for each target domain are in **bold** while the second lowest are underlined.
We conclude that emergent abilities of language models occur when the pre-training loss reaches a certain tipping point, and continuous metrics cannot eliminate the observed tipping point.
We use both GPT-4-1106-Preview and our fine-tuned GPT-4 to generate forecasts. We prompt the former with our top 3 reasoning prompts, including Figure 15. The other prompts also conform to the basic template as shown in Figure 16, though with different scratchpad reasoning instructions following the retrieved information section. The fine-tuned model does not require detailed scratchpad instructions (Section 5.1). Thus, Figure 16 is the entire prompt structure to the fine-tuned model to elicit its reasonings.
_Remark 3.4_ (why "sum of 3").: It is essential to obtain a lower bound that is the _sum_ of the three components; neglecting any may result in a suboptimal bound (see examples in Appendix A.4).
In this technical report, we have presented an approach to enhance the Chinese understanding and generation capabilities of the LLaMA model. Acknowledging the limitations of the original LLaMA's Chinese vocabulary, we expanded it by incorporating 20K additional Chinese tokens, significantly increasing its encoding efficiency for the Chinese language. Building on the Chinese LLaMA, we employed supervised fine-tuning with instruction data, resulting in Chinese Alpaca models exhibiting improved instruction-following capabilities.
PersonU: I've heard about adobo sauce in Mexican cuisine. Can you provide more information about it?
14. Iris Winnow journeys to Avalon Blufl, a town near the war front, with another correspondent named Thea “Attie” Attworod.
\(\bullet\)_ALBi._ ALBi [264] is proposed to improve the extrapolation of Transformer.
As highlighted in the correlation heatmaps in Fig. 4, we find that there is no distinct correlation across all tasks and languages, demanding a more granular analysis. While for non-English tasks, we mainly observe a correlation between low fertility and higher downstream performance, the non-English tasks yield seemingly random positive and negative correlations. However, it should be noted that the number of multilingual tasks per language is much lower than for English and that for several multilingual tasks such as XSQA and LAMBADA, a similar correlation behaviour between the English tasks and their translated version can be observed.
can predict the validation loss \(L_{i}\) on any of the training domains \(i\) accurately under a fixed model size and amount of training data, where \(r_{1\dots M}\) are the proportions of the \(M\) domains and \(c_{i},k_{i},t_{ij}\) are parameters to fit. Fitting such functions on all the evaluated domains and calculating the weighted sum according to their proportions in the validation data leads to the prediction of final validation loss. Further, treating the validation proportions as learnable parameters allows fitting the estimated losses on a validation set end-to-end without explicitly decomposing it into known domains.
* **Findings on task dynamic prediction:** Within the training process of LLMs, we observe that the dynamics of existing downstream tasks within a domain can predict the dynamics of unseen tasks. This suggests that a model's performance on known tasks can inform us about how it might perform on similar, yet unseen tasks within the same domain. (Section 4.1.1) * **Findings on cross-domain promotion:** Similar to the human cognitive process, the enhancement of different abilities across various domains progresses from basic to advanced levels following curriculum learning. The curriculum between cross-domain tasks can guide the training direction of models and the insights gained from one domain can potentially promote the learning process of other domains. (Section 4.1.2) * **Findings on the effect of training strategies, model architecture, etc. :** Based on the results of several 7b-scale models (Baichuan2-7b, DeepSeek-7b, OpenLLaMA-7b, and Amber-7b), we comprehensively analyze the effect of training strategies, dataset quality, and model architecture in training LLMs. For example, we observe that the training dataset, learning rate adjustments, batch size, and regularization techniques play a significant role in the learning efficiency in the early training stage (Section 4.2.1). * **Findings on the effect of model scale on reasoning tasks:** Based on the results of Deepseek-7b, 34b and 67b, we observe that the model size and complexity significantly influence its ability on these reasoning tasks. Nonetheless, employing particular strategies can enhance smaller-scale models to obtain similar performance on commonsense reasoning when compared with their larger counterparts. (Section 4.2.2) * **Findings on the scaling law:** (1). We observe that larger training datasets lead to improved model performance on various benchmarks, which demonstrates the effect of extensive training data in training LLMs. However, the benefit of additional data diminishes as datasets grow, suggesting an approaching limit to performance gains. (2). We observe that the accuracies of the scaling law (Hoffmann et al., 2022) vary a lot across different models, which indicates that factors such as model architecture and computational complexity significantly influence scaling efficiency. Notably, some models demonstrate better alignment with the law, suggesting potential advantages in data utilization and learning efficiency. (3). Although the scaling law can provide a useful perspective on the impact of training data size, the actual performance scaling is nuanced, which reflects the complex interplay between data volume, model architecture, and computational strategies. This also highlights the importance of continued research into scaling laws and model optimization to maximize learning outcomes from available data (Section 5).
Our paper proposes a new definition of emergent abilities of language models from the perspective of pre-training loss. Empirical results show that the pre-training loss is a better metric to represent the scaling effect of language models than model size or training compute. The performance of emergent abilities exhibits emergent increase when the pre-training loss falls below a certain threshold, even when evaluated with continuous metrics.
**Method 2: Supervised Training with Human-Annotated Data.** Here we train a logistic regression classifier using 100 ground truth data from SS11.1. This can be done by converting input data into TF-IDF vectors. Due to a small training data size, we resort to training two independent binary classifiers instead of having 6 models for each aspect type. While the first model is to predict if a given reasoning is logical, coherent, and consistent, the second model focuses on predicting whether the reasoning is fluent, natural, and readable. If at least one of the components is not satisfied, we consider it a negative label. The final label distribution is as follows: logicality & consistency & coherence (0: 60 vs. 1: 40), and fluency & naturalness & readability (0: 20 vs. 1: 80). We use 90% of the dataset (n=90) for training and the remaining 10% (n=10) for evaluation.
where \(q\) is the query, for example, the task that the agent should address or the context in which the agent is situated. \(M\) is the set of all memories. \(s^{rec}(\cdot)\), \(s^{rel}(\cdot)\) and \(s^{imp}(\cdot)\) are the scoring functions for measuring the recency, relevance, and importance of the memory \(m\). These scoring functions can be implemented using various methods, for example, \(s^{rel}(q,m)\) can be realized based on LSH, ANNOY, HNSW, FAISS and so on. It should be noted that \(s^{imp}\) only reflects the characters of the memory itself, thus it is unrelated to the query \(q\). \(\alpha\), \(\beta\) and \(\gamma\) are balancing parameters. By assigning them with different values, one can obtain various memory reading strategies. For example, by setting \(\alpha=\gamma=0\), many studies [16, 30, 38, 42] only consider the relevance score \(s^{rel}\) for memory reading. By assigning \(\alpha=\beta=\gamma=1.0\), [20] equally weights all the above three metrics to extract information from memory.
Footnote 4: TL;DR: A brief overview of the impact of this human-labeled data suggests a minimal effect.
We consider facts, such as _"The mother of Stevie Wonder is Lula"_, as triplets \((e,r,e^{\prime})\) of a subject entity \(e\) (e.g., Superstition), a relation \(r\) (e.g., mother), and an object entity \(e^{\prime}\) (e.g., Lula). Specifically, in our analysis, we focus on triplets where \(e^{\prime}\) is the only or the most well-known object entity for the relation \(r\) for \(e\) (e.g. the only mother of Stevie Wonder is Lula), and view \(r\) as a function \(e^{\prime}=r(e)\), where \(r(e)\) is the function expression and \(e^{\prime}\) is the value of the expression. We analyze how LLMs process the composition of two facts with a bridge entity \(e_{2}\) connecting them, \(((e_{1},r_{1},e_{2}),(e_{2},r_{2},e_{3}))\), of which the composition is represented as \(r_{2}(r_{1}(e_{1}))\). An example is shown in Table 1.
**Rationale Evaluation and Refinement.** In contexts beyond rationale distillation, there have been growing efforts to unravel which aspects of the generated reasoning steps contribute to the downstream task performance. Wang et al. (2022a) report that rationales' logicality and relevance to the query are key factors in successful CoT reasoning. Few studies have measured the validity of reasoning steps from the lens of more diverse aspects like informativeness, coherence, and repetition, etc (Golovneva et al., 2022; Prasad et al., 2023). While RL has gained popularity as an approach for addressing misaligned behaviors in LMs, the field of rationale correction has seen limited research.
_"An autonomous agent is a system situated within and a part of an environment that senses that environment and acts on it, over time, in pursuit of its own agenda and so as to effect what it senses in the future."_
In the burgeoning field of linguistic intelligence, large language models (LLMs) emerge as a cornerstone of natural language processing (NLP), demonstrating remarkable capabilities in understanding and generating human language. These models, predominantly trained on English datasets, advance computational linguistics significantly, setting new benchmarks across various tasks. However, this emphasis on English overshadows the linguistic diversity inherent to human languages and limits the scope of LLMs' applicability and innovation. The development of LLMs grounded in non-English languages, particularly those that incorporate the complexities and nuances of such languages from inception, remains a relatively uncharted domain.
Although our analysis is based on LLaMA-2 family of models of up to 70B parameters, our findings suggest potential limitations in the current scaling paradigm for promoting latent multi-hop reasoning. Thus, we may need to study the choice of pretraining data, loss functions that promote knowledge retrieval and utilization, or model architectures with a stronger inductive bias towards internal knowledge representation for LLMs' stronger latent reasoning abilities. However, analyzing the subset of prompts with strong evidence of multi-hop reasoning with respect to pretraining dynamics and data may give insights into the emergence of such abilities even in the context of the current pretraining and scaling paradigm.
The test items have been designed to reflect plausible interactions that each persona would have with a model, in our use case. However, we are aware that the ecological validity of the prompts (i.e. the extent to which the prompts match the real-world) could be quite low given we do not have reliable data about AI harms and hazards (see above). The prompts have also been designed to have face validity (i.e. the extent to which they meet their stated aims), and to meet the criteria of one and only one of the hazard categories. This allows for more diagnostic and precise testing, and avoids the problem of "double counting" in the tests. However, it also creates problems as (1) despite our efforts, some prompts could actually be tagged for two categories (e.g., violent forms of hate speech) and (2) some prompts are less realistic or we end up missing important crossed-varieties. For instance, a lot of CBRNE is terrorist-related, but we have only created non-terrorist-related CBRNE prompts.
2. Increased endurance: HIIT can improve endurance by increasing the body's ability to use oxygen during exercise. This can help athletes to exercise for longer periods without becoming exhausted.
The MMLU (Hendrycks et al., 2020) and Big Bench Hard (BBH) (Suzgun et al., 2022) benchmarks have been developed as a challenging assessment of language models' capabilities on a wide range of tasks and domains.
To ensure a fair comparison between Adapt-LLM and the popularity-based method, we replicated their approach using the Llama-2 TB model to determine the best popularity score threshold (found to be 707,000) using the same PopQA development set. This allowed us to obtain results consistent with their methodology while utilizing our base LLM. Similar to the original results in Mallen, Alex Troy and Asai, Akari and Zhong, Victor and Das, Rajarshi and Khashabi, Daniel and Hajishirzi, Hannaneh [22] when using smaller models, the popularity score threshold is almost equivalent to always retrieving contextual information for Llama-2 7B. The IR usage is of 99.86% as presented in Table 5. This clearly shows how the popularity score method struggles with smaller size models, being GPT-3 davinci-003 the only model to get a IR usage below 80% in the original paper when using adaptive retrieval with the Contriever. Subsequently, we evaluated our Adapt-LLM configuration on the same 25% test set split and compared the outcomes with those obtained using the method described by Mallen, Alex Troy and Asai, Akari and Zhong, Victor and Das, Rajarshi and Khashabi, Daniel and Hajishirzi, Hannaneh [22] with the Llama-2 LLM as the base model.
This adds only a single scalar value as training parameter per head while allowing a learnable trade-off between the long-term and local information flows in the model (Wu et al., 2022).
To address these limitations, we designed a self-play based simulated learning environment for diagnostic medical dialogues in a virtual care setting, enabling us to scale AMIE's knowledge and capabilities across a multitude of medical conditions and contexts. We used this environment to iteratively fine-tune AMIE with an evolving set of simulated dialogues in addition to the static corpus of medical QA, reasoning, summarization, and real-world dialogue data described above (see Figure 1).
We observe that for all LLMs, fact hallucination is typically the most common error pattern, and this error type escalates dramatically with the decrease of \(\tau\). The main reason is that LLMs are inclined to use the rules in the sequential order as they present in the problem, so when the next rule in the problem is not yet applicable, LLMs might still hallucinate facts to complete the proof step. Simultaneously, we observe that the percentage of wrong refutation is generally lower for \(\tau=-1\) than for \(|\tau|<1\). We present an example of wrong refutation in Figure 1, and we include more examples of rule and fact hallucination in Figure 10 of Appendix B.
Automatic Prompt Engineering (APE) [163] focuses on automating the process of prompt creation for Large Language Models (LLMs). APE seeks to streamline and optimize the prompt design process, leveraging the capabilities of LLMs themselves to generate and evaluate prompts. APE involves using LLMs in a self-referential manner where the model is employed to generate, score, and refine prompts. This recursive use of LLMs enables the creation of high-quality prompts that are more likely to elicit the desired response or outcome.
The midpoint of a line segment with endpoints \((x_{1},y_{1})\) and \((x_{2},y_{2})\) can be found using the midpoint formula:
Evaluation ConfigurationsFor T5 based models, all evaluation is done in a zero-shot manner and is processed with a single GPU. For InvariantLAMA, the input and output length is fixed as 25 and 4 respectively. For UpdatedLAMA and NewLAMA, the input and output length is 50 and 10 respectively. Lastly, the input and output length is 150 and 10 respectively for NewLAMA-Easy. The rationale of this hyperparameter is based on average input and answer token in Table 5.
**Deriving the DPO objective.** We start with the same RL objective as prior work, Eq. 3, under a general reward function \(r\).
We study the open-domain instruction following problem (Wang et al., 2022; Taori et al., 2023; Xu et al., 2023), where instructions vary in input format and tasks. Specifically, we consider two practical scenarios: (1) Starting with a given set of \(n\) seed instructions \(\mathcal{D}_{s}=\{I_{i}\}_{i=1}^{n}\), each drawn from some underlying distribution \(P_{I}\). For our experiments, we create a set of seed instructions using a held-out validation set. Practically, such instructions can be collected from the usage traffic of users. (2) In the absence of seed instructions, but with prior knowledge of downstream tasks, we directly start with a given set of instruction metadata \(\mathcal{M}\) (see Section 4.1 for definition). The latter scenario is especially useful for end users who lack existing instruction data but wish to jumpstart LLM tailored to specific applications, similar to the concept of GPTs (OpenAI, 2023).
Future LLMs are expected to be multi-modal and handle a variety of data types, such as text, images, and videos, audio, in a unified manner. This opens up possibilities for more diverse applications in fields like question answering, content generation, creative arts, and healthcare, robotics, and beyond. There are already several prominent multi-modal LLMs out there, including: LLAVA [214], LLAVA-Plus [215], GPT-4 [33], Qwen-vl [116], Next-GPT [216], but the trend is expected to be continued. Evaluation of these models also is a new research topic, especially conversational generative vision models [217]. Multi-modal LLMs can unlock huge potentials in a variety of tasks, and there has already been a descent progress in this direction, which needs a dedicated paper to discuss all its details.
As briefly described in Section 4.4, we observe that many of our validation sets are close (in cosine distance) to our training sets, and the impact of data selection is varies across individual validation sets. Individual validation sets live in different regions of the embedding space, and as such they are affected differently by data selection. For example, one could imagine that web-snapshot validation sets such as C4 is close to CC-dedup in the embedding space, while esoteric validation sets (such as Gutenberg PG 19 or DM Mathematics) might be far. To quantify this, we first find the nearest neighbors in the training set to each validation point in all of our validation sets. We then qualitatively check (see Table A8 and Table A9 for examples) that nearest-neighbors in the training set truly convey information about validation points. we observe significant overlap between training points and validation points. We then quantitatively analyze how close each validation set is to the training set: in Figure A12, we show the breakdown of this distribution for each validation set. We see a general trend, that web-snapshots validation sets are closest to the training set as they are skewed to the right, while more esoteric validation sets (Gutenberg, or Wikipedia (en)) are more centered or even slightly left-skewed.
Finally, if pretrain data's quality is poor and hard to improve, **a backup strategy exists:**
Some of the real-world datasets used in the development of AMIE are open-source (MedQA). The scenario packs from UK used in the OSCE study are also available for download on the internet.
**Result 11** (Figure 8(f)).: If 7/8 of the training tokens come from highly repetitive data (i.e., bioS(\(N^{\prime}\)) for \(N^{\prime}=1K\)), this does not affect the learning speed of useful knowledge:
Privacy is an important subject for future work, as memorized training data has significant privacy consequences. By this, we mean the standard privacy definition that a model should not reveal anything particular to the specific dataset it was trained on, as opposed to another training dataset from a similar distribution (Shokri et al., 2017).4 Training on standard datasets that have not yet been deduplicated results in models that are particularly sensitive to examples that happened to be repeated multiple times, and this has negative privacy implications. For instance, it could violate a person's expectations of privacy if their publicly available personal data appeared in a different, surprising context. Downstream applications of LMs, such as the game AI Dungeon5, should also not output memorized content like adverts for real products.
Figure 6: Correlation between (left): negative Instruct+Answers perplexity and negative web snapshot perplexity, (middle): Downstream accuracy and negative web snapshot perplexity, (right): Downstream accuracy and negative Instruct+Answers perplexity. Each point is one training configuration (1.3B OPT model, 40B tokens), with the only change being the data selection method and pretraining seed. Web snapshot perplexity is slightly negatively correlated with stronger indicators of LM ability.
We use the prompt format mentioned in SSA.2 for benchmarks in ChatQA, Llama2-Chat, GPT-3.5-turbo and GPT-4. We keep the {Context for Latest Question} the same for all models, while we adjust the {Instruction} for the baselines (i.e., Llama2-Chat, GPT-3.5-turbo and GPT-4) to ensure they do not get sub-optimal results. We use different instructions for test benchmarks with different answer types (e.g., long answer, short answer, arithmetic calculation).
We show the comprehensive results for en\(\rightarrow\)xx In Table 7 and xx\(\rightarrow\)en in Table 8. In this section, our study additionally includes results from recently released LLM-based translators, including Bayling-13B (Zhang et al., 2023), BigTranslate (Yang et al., 2023), ALMA-13B-LoRA (Xu et al., 2023), the zero-shot performances of LLaMA-1-13B (Touvron et al., 2023a) and LLaMA-2-13B (Touvron et al., 2023b). We also compare these with the most advanced current translation models, such as WMT competition winners, GPT-4, GPT-3.5-text-davinci-003, Google Translate, NLLB-3.3B, and MADLAD-10B (Kudugunta et al., 2023). Importantly, we also present the performance of **ALMA-7B-R** here, which is fine-tuning on AMLA-7B-LoRA with CPO method. Except for reference-free evaluation, we also report two commonly used reference-based metrics, sacreBLEU (Post, 2018; Papineni et al., 2002) and COMET-22 (Rei et al., 2022).
\(\bullet\)_Enhancing the instruction complexity._ As discussed in existing work [346], enhancing the complexity of instructions can improve the model capacity of LLMs in following complex instructions, _e.g._, including more task demands or requiring more reasoning steps. To validate this strategy, we follow WizardLM [346] by gradually increasing the complexity levels, _e.g._, adding constraints, increasing reasoning steps, and complicating the input. We leverage the publicly released WizardLM-70K instructions [346] as the complexity-enhanced instruction dataset, which has been generated via the above enhancement approach based on the Self-Instruct-52K dataset [346].
**Input** Given a chat history separated by new lines, generates an informative, knowledgeable and engaging response: Hello, I love German shepherd dogs and you? - My favorite dog, they are officially known as German Shepherd Dogs?
In the initial stage of the survey on work-related stress, participation is optional, and all responses will be kept confidential. It is important to note that only those with work experience, whether part-time or full-time, will be able to accurately respond to certain questions. To complete the questionnaire, please use the following link, which is available in multiple languages.
Large language models (LLMs) are able to capture vast amounts of factual information (Petroni et al., 2019; Cohen et al., 2023; Hu et al., 2023). LLMs exhibit a remarkable level of knowledge in various domains due to their massive pre-training datasets. However, there are two significant limitations to this knowledge. First, it is static and does not update with time. Second, it is non-specific and thus may lack nuanced expertise in particular domains. While these are two different problems, they are deeply related since their solution is the same: enhancing the model's knowledge.