Datasets:
Translation-EnKo
/
arxiv-translation

Dataset card Data Studio Files Community
2
text
stringlengths
4
4.47k
**Inference settings.** As a default configuration, we assign the weight terms \(\boxed{\text{IsRel}}\), \(\boxed{\text{IsNet}}\), \(\boxed{\text{IsNet}}\) values of 1.0, 1.0 and 0.5, respectively. To encourage frequent retrieval, we set the retrieval threshold to 0.2 for most tasks and to 0 for ALCE (Gao et al., 2023) due to citation requirements. We speed up inference using vllm (Kwon et al., 2023). At each segment level, we adopt a beam width of 2. For a token-level generation, we use greedy decoding. By default, we use the top five documents from Contriever-MS MARCO (Izacard et al., 2022); for biographies and open-domain QA, we use additional top five documents retrieved by a web search engine, following Luo et al. (2023); for ASQA, we use the author-provided top 5 documents by GTR-XXL (Ni et al., 2022) across all baselines for a fair comparison.
Figure 9: **Scaling on Hutter LLM for Varying \(k\).** A sample of size \(T_{0}\) is used to approximate the true distribution \(p\) via \(p_{\text{AI}}\). Then, a Hutter-type model is learned on a sample of size \(T\) from \(p_{\text{AI}}\), and evaluated on the true data distribution \(p\). Each horizontal line corresponds to the asymptote \(k^{-\beta c}\asymp T_{0}^{-c}\), for different values of \(T_{0}\). The diagonal line corresponds to \(T^{-c}\).
**Clarifications on Experiments**. In this version, we have included a number experiments on instruction-tuning (Table 9), overall ability evaluation (Table 16), and prompt engineering (Table 17). Due to the limit of computational resources, our experiments are not complete, limited to small-sized models or a few comparisons. Despite that, we feel that it might be meaningful to share the partial results to the public. We will try to include the missing results of larger models or more comparisons in the future versions.
Human evaluation setup.Evaluating models' performance on this evaluation set of diverse tasks is extremely challenging because different tasks require different expertise. Indeed, many of these tasks cannot be measured by automatic metrics or even be judged by normal crowdworkers (e.g., writing a program, or converting first-order logic into natural language). To get a more faithful evaluation, we asked the authors of the instructions to judge model predictions. Details on how we set up this human evaluation are described in Appendix B. The evaluators were asked to rate the output based on whether it accurately and effectively completes the task.
Ensure your student can answer the new question without the given question. If you want to use some numbers, conditions or background in the given question, please restate them to ensure no information is omitted in your new question.
Different from traditional LLMs, autonomous agent usually has to play as specific roles (_e.g._, program coder, researcher and chemist) for accomplishing different tasks. Thus, the capability of the agent for role-playing is very important. Although LLMs can effectively simulate many common roles such as movie reviewers, there are still various roles and aspects that they struggle to capture accurately. To begin with, LLMs are usually trained based on web-corpus, thus for the roles which are seldom discussed on the web or the newly emerging roles, LLMs may not simulate them well. In addition, previous research [30] has shown that existing LLMs may not well model the human cognitive psychology characters, leading to the lack of self-awareness in conversation scenarios. Potential solution to these problems may include fine-tuning LLMs or carefully designing the agent prompts/architectures [183]. For example, one can firstly collect real-human data for uncommon roles or psychology characters, and then leverage it to fine-tune LLMs. However, how to ensure that fine-tuned model still perform well for the common roles may pose further challenges. Beyond fine-tuning, one can also design tailored agent prompts/architectures to enhance the capability of LLM on role-playing. However, finding the optimal prompts/architectures is not easy, since their designing spaces are too large.
In Lemma C.2, we showed that orthogonal matrices \(\mathbf{U}\) commute with normalization.
To measure the time-invariant knowledge, we use InvariantLAMA (IL) because most of the slots to fill are at the end of the sentence. For light-tuning on behalf of IL, we use additional T-Rex data from Shin et al. (2020) which has a similar distribution as instances from IL. Among them, 5,000 instances with the same _time-invariant_ relations as IL are randomly sampled for _light-tuning_. On the other hand, unlike IL where most of the slots to fill are at the end of the sentences, the LAMA datasets for new knowledge in our CKL benchmark mostly have the slots at the beginning of the sentences. Therefore, we use the corresponding CBQA dataset of NewLAMA-Easy, NewQuestions-Easy (NQE) to roughly measure the new knowledge.15 For light-tuning on behalf of NQE, 5,000 instances are sampled from a set of QA pairs constructed from CC-RecentNews but not CC-RecentNews-Small to remove the test-train overlap.
The primary function, remove_unwanted_part, starts by splitting the input data into individual sentences. If the first sentence contains delimiters such as "\n\n" (indicating a new paragraph) or ":", the function checks the segment preceding the delimiter for the aforementioned unwanted elements. If these elements are detected, the preceding segment is removed. The entire revised content is then reconstructed and returned. In cases where no modifications are applicable, but we still have the flagged keywords, we remove the paraphrase completely.
Define the task: You need to define the specific task your assistant will perform. For example, answering questions about a particular topic or helping with a specific task such as scheduling. The more specific the task, the easier it will be to train the model.
\(\bullet\)_GPT-1_. In 2017, the Transformer model [22] was introduced by Google, and the OpenAI team quickly adapted their language modeling work to this new neural network architecture. They released the first GPT model in 2018, _i.e._, GPT-1 [122], and coined the abbreviation term _GPT_ as the model name, standing for _Generative Pre-Training_. GPT-1 was developed based on a generative, decoder-only Transformer architecture, and adopted a hybrid approach of unsupervised pretraining and supervised fine-tuning. GPT-1 has set up the core architecture for the GPT-series models and established the underlying principle to model natural language text, _i.e._, predicting the next word.
Replace a commonly used requirement in the programming task with a less common and more specific one.
**Discussion on Effect of Data Quality.** For pre-training, the quality of pre-training data is vital to the model capacities of LLMs. Existing work has shown that pre-training on the low-quality corpus, such as noisy, toxic, and duplicate data, would largely hurt the performance of models [64, 214, 216, 219]. Recent studies, such as T5 [82], GLaM [112], and Gopher [64], have investigated the influence of data quality on the LLMs' capacities. By comparing the performance of models trained on the filtered and unfiltered corpus, they have reached the similar conclusion that pre-training LLMs on cleaned data can improve the model performance. More specifically, the duplication of data may result in "_double descent_" (referring to the phenomenon of performance initially deteriorating and subsequently improving) [214, 228], or even overwhelm the training process [214]. In addition, it has been shown that duplicate data degrades the ability of LLMs to copy from the context, which might further affect the generalization capacity of LLMs using in-context learning [214]. Therefore, as suggested in [56, 64, 78, 212], it is essential to utilize preprocessing methods like quality filtering, toxic filtering and deduplication to carefully clean the pre-training corpus (as illustrated in Section 4.1.2), to improve stability of the training process and avoid affecting the model performance.
ExperimentWe demonstrate that EntRec is a reasonable approximation of the internal recall of the bridge entity with indirect evidence. Note that EntRec\({}^{l}(e_{2},\tau_{\text{2H}})\) is calculated not at the last token of \(\tau_{\text{2H}}\) but at the last token of the bridge entity's descriptive mention, where it is grammatically natural to prepend a comma followed by the name of \(e_{2}\) (e.g., "The mother of the singer of 'Superstition', _Stevie Wonder_"). In the resulting string, grammatically \(\mu(r_{1}(e_{1}))\)) becomes the _antecedent_ and \(e_{2}\) becomes the _appositive_; an appositive is a noun phrase that follows another noun phrase in opposition to it and provides information that further identifies or defines it, and the antecedent is the noun phrase that the appositive describes. Then, if EntRec\({}^{l}(e_{2},\tau_{\text{2H}})\) reasonably approximates the internal recall of the bridge entity \(e_{2}\), it is expected that _there will be at least some layers \(l\) where increasing EntRec\({}^{l}(e_{2},\tau_{\text{2H}})\) increases the relative frequency of the LLM to generate \(e_{2}^{(0)}\) with a relative frequency higher than random chance_.
Both deduplication methods identify overlap between the train set and the validation set (Table 2). For example, 4.6% of the C4 validation set and 14.4% of the RealNews validation set examples had an approximate duplicate in their respective training sets. Such duplication is problematic since it could cause evaluation metrics to be unfairly inflated for models that are better at memorizing their train sets. We evaluate the effect of this leakage on publicly released models in Section 6.3.
This yields a set of 5,516 binary questions, including 3,762 for training, 840 for validation, and 914 for testing (Table 1(a)). See Table 12 for a sample data point and Appendix C for details about the curation process.
15. In Avalon Blufl, Marsiol Torres warns Iris Winnow and Thea Attwood about the regular threats from mythical creatures.
* **MinHash alone is insufficient**, as it doesn't match the zero-shot performance of exact deduplication. Conversely, combining it with exact deduplication doesn't improve performance further. * **Masking spanned duplicates degrades performance**, systematically underperforming other approaches. Dropping and cutting spans perform similarly, although it's likely that dropping documents slightly outperforms cutting.
Footnote 6: Here, \(N_{c}\)\(D_{c}\) and \(C_{c}\) are measured in the number of non-embedding parameters, the number of training tokens and the number of FP-days, respectively. According to the original paper [30], \(C_{c}\) and \(C\) should be denoted by \(C_{c}^{min}\) and \(C_{min}\), corresponding to the optimal use of compute. We use the simplified notations for ease of discussions.
* We investigate different data selection strategies for standard LLM pre-training setups where data has already been manually filtered / de-duplicated (e.g., MinHash), and where we do not know the target distribution for which we optimize performance. We argue that the performance of SSL Prototypes is affected by duplicate-driven clusters in the embedding space. In Section 3.4 we propose a new data selection strategy **D4** that utilizes SemDeDup to avoid getting impacted by such clusters. * In Section 4.1, we show that in the _compute-limited regime_ where we have "infinite" source data and train models with fixed token budgets, we can achieve better pre-training perplexity and downstream accuracy than random iid data selection and previously established methods. Furthermore, we show that our method D4 can achieve around 20% efficiency gains at the 6.7b model scale, and that the magnitude of efficiency gains increases with model scale. * In the _data-limited regime_, where we run out of data and must epoch over data, cleverly choosing what data to repeat can beat training on randomly selected new data, whereas randomly choosing data to repeat underperforms adding new data (Section 4.2). This calls into question the standard practice of single epoch LLM training, and suggests that epoching over intelligently subselected data might be a better approach.
This section presents the researchers' efforts in language modelling, including Pre-Generation Retrieval Processing, Structural Model Optimization, and Post-Generation Output Enhancement.
Implicit MethodsZhu et al. (2020) proposed a new task of explicitly modifying specific facts without forgetting unmodified facts and provided several benchmark approaches without utilizing non-parametric memory, including constrained layer-wise finetuning. Wang et al. (2021) proposed K-Adapter, a method that adds adapters to frozen layers of pretrained LMs to inject factual and linguistic knowledge and improve performance on downstream tasks. Chen et al. (2020) proposed a new optimizer that simulates the pretraining optimization while finetuning on the target task without needing access to the pretraining corpus, improving performance on the GLUE benchmark. De Cao et al. (2021) propose using a hyper-network to edit factual knowledge.
We use the average accuracy of MATH and GSM8K as the standard for measuring downstream tasks performance of model.
**Reference [1]** Emperor Constantine then enacted the first Sunday Laws, for "the venerable Day of the Sun" in 321 A.D. On March 7, 321, the Roman Emperor Constantine issued a decree making Sunday a day of rest from labor, stating: We find a good description of this in sources such as the book Rest Days which states: Early Christian observance of both the spiritual seventh-day sabbath and a Lords Day assembly is evidenced in Ignatius] letter to the Magnesians ca. 110.
By contrast, our fitted model implies an optimal ratio of around 20 tokens per parameter, which is consistent with both how the Chinchilla model was trained and the findings from Approaches 1 and 2 in Hoffmann et al. (see Figure 5). The inconsistency between the prescriptions based on the estimated scaling law (Approach 3) and the results from Approaches 1 and 2 in Hoffmann et al. raises further concerns about the accuracy of their parameter estimates. The alignment of our estimates with the token-to-parameter ratio used in practice and obtained from their other approaches strengthens the case for the validity of our fit.
In a netshell, AIOS stands as a motivating body of work that brings a broad spectrum of research opportunities. Each outlined direction can not only build upon the foundational elements of AIOS but also contribute to the advancement of the field at large.
As for the concrete data-to-model ratio \(\frac{D_{opt}}{N_{opt}}\), we notice that there is a huge gap in compute optimal regime of between ours and Hoffmann et al. (2022) despite that the trend of \(\frac{D_{opt}}{N_{opt}}\), with compute \(C\) is aligned between ours and theirs. Specifically, the data size should be 192 times larger than the model size on average, as opposed to 20 times in Hoffmann et al. (2022). We note that this aligns with the observation in Section 4.3 and Figure 6.
Impact of front and back matter on the summary qualityBooks frequently contain additional information beyond the main narrative, including the author's biography, table of contents, dedications, and more, positioned at the beginning or the end of the book. Ideally, models should exclude this extraneous content, focusing solely on summarizing the core story. However, we have noted that models are sometimes unduly influenced by these elements, which can dominate a significant part of the summary and occasionally compromise its accuracy. Overall, between 19.23% (GPT-3.5-Turbo and GPT-4) and 34.62% (Mixtral) of summaries were affected by such content, either through focusing on this information,14 confusing story characters with names found in the front and/or back matter,15 or making up entire narrative based on a single mention.16Claude-3-Opus was the only model seemingly unaffected by the additional information. However, when faced with two summaries--where one primarily summarized the content of the back matter, since it represented the final chunk--the model declined to perform the task. We regard this cautious approach as preferable to introducing unfounded details or irrelevant content. Examples of such cases are shown in Table 24.
In Table 6, we show ablation studies on how the number of retrieved context/chunks, context ordering, and different retrievers affect the conversational QA results.
Various benchmarks and leaderboards have been proposed to address the most challenging question in the world of large language models: Which one is better? However not a simple answer can address this question. The answer depends on various aspects of large language models. Section V shows the categorical presentation of different tasks and the most important datasets in each category. We will follow the same categorization and provide a comparison based on each category. After providing comparison for each category, we will provide a broad overview of aggregated performance by averaging the reported performance metric on different tasks.
Moreover, We relate the selected tokens' loss to its downstream task performance via a power law in Figure 7, which is similar to a concurrent study (Gadre et al., 2024). Observing the curve fitted from the data points in the graph, the average loss of tokens selected by the SLM shows a positive correlation with the performance of downstream tasks, whereas the average loss of tokens not selected exhibits a negative correlation with downstream task performance. Therefore, it is not necessary for the all tokens' loss to decrease to benefit the model's ultimate performance. See Appendix D for more details.
Figure A.17: Rank-order agreement to specialist ratings of all 149 dialogue pairs, comparing various auto-evaluation prompting techniques. We choose to leverage the self-CoT technique for the auto-evaluation of clinical criteria.
Figure 2 compares our model, Infini-Transformer, and Transformer-XL (Dai et al., 2019). Similar to Transformer-XL, Infini-Transformer operates on a sequence of segments. We compute the standard causal dot-product attention context within each segment. So the dot-product attention computation is local in a sense that it covers a total \(N\) number of tokens of the current segment with index \(S\) (\(N\) is the segment length).
(5) Accuracy: Evaluating whether the model can perform correctly in the corresponding for a given instruction.
First, we discuss the overall results obtained for the investigated tokenizers, followed by presenting the impact of the tokenizer library (Section 5.2.1), the impact of the tokenizer algorithm (Section 5.2.2), and the impact of the vocabulary size (Section 5.2.3).
\(\bullet\)_Honesty._ At a basic level, a LLM aligned to be honest should present accurate content to users instead of fabricating information. Additionally, it is crucial for the LLM to convey appropriate degrees of uncertainty in its output, in order to avoid any form of deception or misrepresentation of information. This requires the model to know about its capabilities and levels of knowledge (_e.g._, "know unknowns"). According to the discussion in [368], honesty is a more objective criterion compared to helpfulness and harmlessness, hence honesty alignment could potentially be developed with less reliance on human efforts.
DatasetTo the best of our knowledge, publically accessible Japanese VLM datasets are scarce. In response, we created a new open Japanese VLM benchmark and assessed our VLM on a widely recognized Japanese VQA dataset.
These results clearly show that _not all data is equal for continual pre-training_. In fact, all the data used in efficient DACP methods (10%) is a subset of the data in DACP. Since DACP's (100%) performance is lower than ETS-DACP/ETA-DACP-ent, adding more data on top of highly similar or high entropy data actually hurts the performance. The difference in results between hard and soft sampling adds more evidence to this observation. While there is variability across tasks, on an average, adding examples from outside the top decile of metrics hurts the performance with the notable exception of ETS-DACP-com which is a combination of all three metrics. Hence, we should carefully curate the data for any domain continual pre-training.
Figure A.8: **Specialist rated DDx accuracy by scenario specialty.** Top 1/3/5/10 accuracy for scenarios of each specialty. Accuracies are based on the specialist ratings for AMIE and PCP differential diagnoses with respect to the ground truth. Number of dialogues per OSCE agent: Cardiology (29), Gastroenterology (31), Internal Medicine (14), Neurology (30), Respiratory (30), OBGYN / Urology (15).
Here we present an ablation analysis for varying degrees of matching to the ground truth where for each differential, we only considered a diagnosis a match if the specialist indicated in the answer to this question that the match was at least as close as the specified degree of matching. Note that all other specialist-rated DDx evaluations in this paper used the "Relevant" threshold when computing accuracy. The differences between AMIE and PCPs in DDx accuracy were statistically significant for all values of k at the matching levels "Relevant", "Extremely Relevant", and "Exact Match".
Purpose of annotationWe are annotating 2,000 responses from a range of AI models used for chat. This is part of the ML Commons' Working Group on AI Safety efforts to benchmark the safety risks of AI models. You can find out more about our work HERE, and reach out to Bertie if you have questions.
\(\bullet\)_FLAN_[67] consists of 62 widely used NLP benchmarks in its original version. Recently, FLAN-v2 [183] is also proposed, which expands FLAN by mixing additional instruction datasets, including Muffin [67], NIV2 [88], T0-SF [28], and CoT [184, 185, 186]. Muffin contains 62 tasks from the original FLAN and additional 26 tasks, including conversation and code synthesis tasks. T0-SF is extracted from T0 [28] while ensuring no overlap with Muffin. NIV2 refers to the Natural-Instructions v2 dataset [88], and CoT [184, 185, 186] is a combination of nine reasoning tasks with corresponding chain-of-thought prompts and outputs.
MethodFollowing the methodology outlined by Kim et al. [38], our goal is to upscale our Yi-6B base model, which has 32 layers, to a 9B model named the Yi-9B base model, featuring 48 layers, by duplicating the original 16 middle layers 12-28. Depth up-scaling involves expanding the base model's depth and subsequently continuing the pretraining phase for the enhanced model.
The base version of the dataset is self-contained, but the multimodal version is interlaced with links to images-these are not distributed as part of the dataset, and constitute an external source.
Allen-Zhu and Li [3] suggested that rewriting pretraining data is crucial for making knowledge extractable rather word-by-word memorization.16 However, they did not explore the impact on the model's capacity. Our paper addresses this gap, indicating that rewriting pretraining data does not compromise -- and may even enhance -- the model's knowledge capacity.
**Method:** Let \(\mathbf{Q}\) be as defined in Algorithm 1. Now rewrite \(\mathbf{Q}=\mathbf{U}\cdot\mathbf{\Sigma}\cdot\mathbf{V}^{\top}\) with SVD. Previously we saw that the number of large enough singular values corresponded to the dimension of the model. But it turns out that the matrix \(\mathbf{U}\) actually directly represents (a rotation of) the final layer! Specifically, we can show that \(\mathbf{U}\cdot\mathbf{\Sigma}=\mathbf{W}\cdot\mathbf{G}\) for some \(h\times h\) matrix \(\mathbf{G}\) in the following lemma.
PersonA: Yes, you can preserve adobo sauce by freezing or canning it when you're done making it.
After pre-training, instruction tuning (_a.k.a._, supervised fine-tuning) is an important method to enhance or unlock specific abilities of LLMs (_e.g._, instruction following). In this part, we introduce several widely used datasets for instruction tuning, and categorize them into three main types based on the construction method of formatted instruction instances, namely NLP task datasets, daily chat datasets and synthetic datasets. We show their details in Table III.
improvements compared to the single-sample approach. Multiple sampling with voting has become a common technique in current X-of-thought studies. Integrating reasoning chains into voting remains a significant area of research for the future.
**Comparing FT vs. RAG**. As interest grows in refining pre-trained language models for particular tasks, the comparison of FT and RAG strategies under equitable conditions is becoming increasingly important. Mosbach et al. (2023) explored the effectiveness of few-shot FT versus In-context Learning for classification tasks in general domains. de Luis Balaguer et al. (2024) compared FT and RAG in answering long, agriculture, and geography-specific questions. Ovadia et al. (2023) assessed the performance on multiple-choice questions in specialized areas like Anatomy, Astronomy, College Biology, and Prehistory. In contrast to these studies, we directly address the integration of less popular factual knowledge into LLMs, comparing various retrieval, data augmentation, and fine tuning methods.
The results indicate that the benefits of introducing high-quality data at the beginning of decay stage are much higher than simply adding it during the SFT phase.
**FastChat**[227] is an open platform for training, serving, and evaluating large language model based chatbots. FastChat's core features include: The training and evaluation code for state-of-the-art models (e.g., Vicuna, MT-Bench), and a distributed multi-model serving system with web UI and OpenAI-compatible RESTful APIs.
Now, to find the total number of gems in the chest, we add the number of diamonds, rubies, and emeralds together.
After constructing \(\mathcal{A}\), it is straightforward to identify duplicated training examples. Suppose that the sequence \(s\) was repeated exactly twice in the training dataset \(\mathcal{S}\) at positions \(i\) and \(j\), that is, \(\mathcal{S}_{i.i+|s|}=\mathcal{S}_{j.j+|s|}\). Then the indices \(i,j\) will occur adjacent to each other in the suffix array \(\mathcal{A}\).
Note that our confidence interval for \(a\) is consistent with Approaches 1 and 2 in Hoffmann et al. in the sense that our confidence intervals overlap with theirs. This implies that our optimal scaling policy is consistent with the scaling recommendations from those approaches.
However, model merging is considered by many to be a form of black art or alchemy, relying on the model maker's intuition and instincts about model selection and merging recipes to create and refine a new model that performs well for a particular task. Furthermore, the model maker is often required to have some domain knowledge for the various different benchmark tasks. Given the large diversity of open models and benchmarks in the community, human intuition can only go so far, and we believe a more systematic approach for discovering new model combinations will take things much further.
Scaling to about a Million SFT Math Data Considering the effectiveness of the synthetic approach, we substantially increase the scale of the SFT data for both GSM8K and MATH problems, to 960K and 480K, respectively. Figure 1 presents the main results utilizing various sizes of the LLaMA-2 series. The straightforward scaling strategy yields state-of-the-art accuracy.
Strong language models are being pre-trained on combinations of real and synthetic data. Using synthetic data enables baking in desirable attributes such as fairness, bias, and style (like instruction following) directly into the data, eliminating the need to adjust the training algorithm specifically. This offers an alternative approach to aligning language models to human values. The recent uptick in interest around synthetic data, especially for instruction-tuning language models, is noteworthy, with concurrent researchers also leveraging it for pre-training. As we transition into this paradigm, understanding the properties of the data fed to our models is paramount.
To accommodate real-world pretraining data that mostly contains more than two domains, we extend our investigation into multiple domains. For simplicity and the ease of visual aids, we start with the case of three domains.
Our idealized setting, free from irrelevant data, allows for more accurate scaling law computations -- we also discuss how "junk" data affects capacity later in Section 10. In contrast, it is difficult to quantify real-life knowledge; for instance, if LLaMA-70B outperforms LLaMA-7B by 30% on a benchmark, it doesn't necessarily mean a tenfold model scaling only boosts capacity by 30% (see Footnote 1).
As shown in Table XIII, we conduct experiments on two re-identification (re-ID) tasks, i.e., the person re-identification [257] and vehicle re-identification [256]. For the person re-ID, four widely used datasets are used, including MSMT17 [258], Market1501 [259], DukeMTMC [260], and Occluded-Duke [261] dataset. These datasets are captured from different scenes, and the samples are collected from surveillance systems with overlapping coverage of cameras, which has challenges such as cross-time span, occlusion, and background interference. For the vehicle re-ID, VeRi-776 [262] and VehicleID [263] datasets are utilized for the experimental validation. Different from pedestrian samples, the change of observation viewpoints also brings significant appearance differences for vehicles, for thus the vehicle datasets are additionally provided with viewpoint labels to mark the different viewpoints of the vehicle samples. For the above datasets, we use the Cumulative Matching Characteristic (CMC) curve and mean Average Precision (mAP) as evaluation metrics.
Footnote 7: Touvron et al. (2023b) report that Llama 2 was pretrained on data contaminated with MMLU test data.
**Preference Data Components**: Our preference data selection involves choosing preferred and dis-preferred translations from a triplet consisting of outputs from GPT-4, ALMA, and the gold reference. In the right of Figure 4, we emphasize the significance of the data generated by both ALMA and GPT-4. The results indicate a notable decline in performance when ALMA data is excluded in the en\(\rightarrow\)xx direction. Conversely, omitting GPT-4 data leads to a significant performance decrease in the xx\(\rightarrow\)en direction. This demonstrates that data generated by both systems plays a helpful role in enhancing model performance.
Rewritten Prompt(MUST contain a specific Shell cmd as input): I'm trying to scp a file from a remote server to my local machine. Only port 80 is accessible.
As a special prompting form, in-context learning (ICL) is first proposed along with GPT-3 [55], which has become a typical approach to utilizing LLMs.
that \(M^{L}\) answered correctly and an additional 1000 from questions where the model provided incorrect responses. This results in a total of 2000 samples for evaluation.
Code-switching is a phenomenon in which multilingual speakers switch between languages within a single utterance. Previous work on the performance of multilingual language models on code-switching tasks has shown mixed results. Some studies have suggested that pretrained models fine-tuned for specific code-switching scenarios can achieve state-of-the-art performance for certain language pairs such as English-Spanish and English-Hindi Khanuja et al. (2020), while others have found that using meta-embeddings can yield better results with fewer parameters Winata et al. (2019); Winata et al. (2019, 2021). In another line of research, code-switching-based methods have been presented to improve the capability of multilingual language models Jiang et al. (2020); Tan and Joty (2021); Krishnan et al. (2021).
Figure A.18: Rank-order agreement to specialist ratings of 139 dialogue pairs (excluding cases without multiple specialist ratings) for alternative specialists compared to the self-CoT auto-evaluation technique. Auto-evaluation agreement to the first specialist is comparable to inter-specialist agreement. The black dashed line shows the rank-order agreement one would get with a random ranking of the AMIE and PCP dialogues, while the green dashed line shows the rank-order agreement with a strategy of randomly guessing according to the distribution of specialist preferences for each criteria.
For the clinical condition, [Condition], is the following a good description of common demographics/symptoms/management plans (Yes/No)?
Advanced Scheduling Algorithms. The scheduling function of AIOS lays the groundwork for the development of more advanced algorithms. Future research could focus on algorithms that perform dependency analysis among agent requests, optimizing the allocation of computational resources. Additionally, some of the tool resources are locally deployed models, which can also be incorporated into the scheduling paradigm. This includes the management of tool status and snapshots, suggesting a move towards a unified scheduling framework that encompasses both agents and their tools.
* Improving the ability to recognize hallucination phenomena in the reasoning processes. * Improving the accuracy of external knowledge retrieval and utilization to reduce factual mistakes. * Improving the ability to recognize and correct contextual inconsistencies and logical mistakes, which is more challenging. * How to fundamentally eliminate hallucination phenomena from alternative approaches, e.g. specific pre-training.
ically, the Yi-6B model's superior performance against models with significantly more parameters challenges the notion that parameter count alone is a sufficient predictor of model efficacy.
For the detection task, Chen et al. [80] propose a Mamba-in-Mamba (MiM-ISTD) structure to detect the infrared small targets. In this structure, the images are evenly divided into "visual sentences" (patches) and further subdivided into "visual words" (sub-patches), and a pure Mamba-based MiM pyramid encoder is designed to extract global and local features. Chen et al.
* **Two datasets, weak shift**: In this variation, we consider \(\mathcal{D}_{0}\) to be the Pile (Gao et al., 2020) and \(\mathcal{D}_{1}\) to be pre-training on SlimPajama (Soboleva et al., 2023). SlimPajama is an extensively deduplicated version of RedPajama (Computer, 2023) which is built based on the LLaMA dataset (Touvron et al., 2023a). We consider this to be a weak but realistic distribution shift as both datasets are English-language, contain common domains(CommonCrawl, GitHub, Arxiv, Wikipedia, and StackExchange) as well as other non-overlapping domains. Additionally, SlimPajama (2023) is a newer dataset than Pile (2020) and is therefore likely to have newer data within the overlapping domains. Therefore, despite the potential overlap, we believe this transition is realistic and is likely to be of interest to many practitioners wishing to update an LLM on a similar distribution to pre-training (e.g., newly collected data of the same sources with higher quality filtering). * **Two datasets, stronger shift**: In this variation, we consider \(\mathcal{D}_{0}\) to be pre-training on the Pile (Gao et al., 2020) and \(\mathcal{D}_{1}\) to be pre-training on German Common Crawl. German Common Crawl is a \(\sim\) 200B token dataset taken from the Oscar dataset (Laippala et al., 2022). We note that this constitutes a stronger shift given the change of language. This setting is of particular interest for practitioners wishing to augment an LLM with a new natural language, programming language, or specific domain that is notably different in vocabulary from pre-training. We note, however, that as the domain strays farther and farther away from the tokenizer's training corpus, the tokenizer may become a key bottleneck to performance. We leave the treatment of the tokenizer to future work.
**Safety Prompts Evaluation**Sun et al. (2023) used InstructGPT (Ouyang et al., 2022) with a verbalizer to assess text safety. In our method, we simplify things. Instead of using a verbalizer to understand the output, we use the function call feature of the OpenAI GPT 3.5 8. This makes the results clearer and easier to interpret. For those who are interested, we have detailed the specifics of this function call and the related assessment prompts in Appendix A.2.
This approach ensures that the model effectively learns to discern when context is necessary for answering questions, or to provide a direct response when it suffices, as well as answer directly when provided with context.
CoT is based on making the implicit reasoning process of LLMs explicit. By outlining the steps required for reasoning, the model is directed closer to a logical and reasoned output, especially in scenarios demanding more than simple information retrieval or pattern recognition.
We pre-train the Chinese LLaMA model with the standard Causal Language Modeling (CLM) task. Given an input token sequence \(\mathbf{x}=(x_{0},x_{1},x_{2},\ldots)\), the model is trained to predict the next token \(x_{i}\) in an autoregressive manner.
In sDPO, we propose to use the available preference datasets in a stepwise manner instead of using them all at once. The comparison of the overall flow of DPO and sDPO is presented in Fig. 1.
LLMs acquire language generation and instruction-following capabilities through pretraining and instruction tuning. However, English holds a dominant position in the field of natural language processing, possessing the most abundant collection of text data from various domains. LLMs trained on English-dominant corpora exhibit inferior performance on other non-English languages. Extrapolating LLMs to non-English languages poses a highly valuable research challenge. Common extrapolation approaches consist of the following three steps: (1) extending the vocabulary to add tokens of the target language, and thus enhancing encoding expressiveness to that language. (2) further pretraining to transfer language generation capabilities of LLMs to the target language. The required training scale for this step is generally on the order of billions of tokens, significantly less than the trillions of tokens needed for training from scratch. (3) conducting SFT in the target language to transfer instruction-following capabilities of LLMs.
As shown in Figure 2, the implementation of RL enables the model to achieve additional improvements in both rationale qualities and task performance. However, in line with Joshi et al. (2023)'s findings, a slight decrease in task performance is observed at the cost of maximized rationale qualities when selecting top-quality rationales. There may be several underlying factors involved (e.g., models' unfaithfulness), but this is not the scope of this work.
We perform a simple ablation on our 12-thought-token-4-ahead baseline, namely asking whether sampling multiple thoughts per sequence is necessary. We find that although simply computing the reward as the difference between the losses with and without thought proves to be a strong baseline, using multiple thoughts consistently outperformed it (by roughly 0.5% on GSM8K generalization and 3% on CommonsenseQA generalization). However, the exact number of thoughts had little impact: varying between 2, 3, and 4 thoughts per sequence appeared to result in a consistent improvement with additionalthoughts, but additional thoughts per sequence beyond two improved performance by only 0.1-0.3% on both GSM8K and CommonsenseQA. Similarly, we found that predicting more than one token ahead helped by 0.3% on GSM8K and 3.1% on CommonsenseQA, but with 12-thought-tokens, we did not find that additional tokens ahead, beyond two, improved performance. However, qualitatively, we found that the rationales appeared more coherent with additional tokens-ahead of supervision. This suggests a need for more specific evaluations of generation quality in future work.
Footnote 4: More concretely, we regard tasks that have a small limited output label space as classification tasks.
Given that our model operates in a many-to-many translation format and the additional data is specific to only de and zh directions, we anticipate changes in performance when translating into these languages, but not in others. To assess the impact of the human-labeled data, we conducted a comparison between models exclusively fine-tuned on triplet data and those fine-tuned on both triplet and human-labeled data. The training approach remained consistent, utilizing the ALMA-13B-LoRA model fine-tuned via CPO. It's important to note that tied data were excluded from this analysis due to their lack of clear preference.
Figure 2: Comparative performance analysis of Baichuan-7B, DeepSeek-7B, DeepSeek-67B, Amber-7B, OpenLLaMA-7B, and Yi-34B models across standard benchmarks.
**Question:** What was a key issue that led to the 2023 United Auto Workers strike?
Figure 18: We extract the training loss data from Llama2 paper (left part) and estimate the compute optimal \(\frac{D_{opt}}{N_{opt}}\) in their paper using the right part. The straight lines are plotted to estimate the optimal loss envelope assuming using WSD Scheduler.
We then manually examine to eliminate all questions of low quality. This includes those with few community forecasts or trading engagement on platforms such as Manifold and Polymarket, as well as any ill-defined questions that GPT-3.5 is unable to identify during the initial screening.
**RLHF** (reinforcement learning from human feedback) and **RLAIF** (reinforcement learning from AI feedback) are two popular approaches. RLHF uses a reward model to learn alignment from human feedback. This reward model, after being tuned, is able to rate different outputs and score them according to their alignment preferences given by humans. The reward model gives feedback to the original LLM and this feedback is used to tune the LLM further [137]. Reinforcement learning from AI feedback on the other hand, directly connects a pretrained and well-aligned model to the LLM and helps it to learn from larger and more aligned models [138].
In this paper, we conduct a comprehensive survey of the field of LLM-based autonomous agents. Specifically, we organize our survey based on three aspects including the construction, application, and evaluation of LLM-based autonomous agents. For the agent construction, we focus on two problems, that is, (1) how to design the agent architecture to better leverage LLMs, and (2) how to inspire and enhance the agent capability to complete different tasks. Intuitively, the first problem aims to build the hardware fundamentals for the agent, while the second problem focus on providing the agent with software resources. For the first problem, we present a unified agent framework, which can encompass most of the previous studies. For the second problem, we provide a summary on the commonly-used strategies for agents' capability acquisition. In addition to discussing agent construction, we also provide an systematic overview of the applications of LLM-based autonomous agents in social science, natural science, and engineering. Finally, we delve into the strategies for evaluating LLM-based autonomous agents, focusing on both subjective and objective strategies.
Footnote 3: We use the term “professional law” here as defined in Hendrycks et al.
Such distributions are ubiquitous in natural datasets, from Zipf's law (Zipf, 1935) in distribution of word frequencies, to biological data, earthquake magnitudes, financial data etc. - this is the data being consumed by large models at scale, like LLMs. But what distribution do AI-models generate when trained on such data? Figure 2 provides an empirical answer for a large scale LLM (Llama2-7B) and a transformer model trained on an arithmetic task. Regenerating heavy-tailed data affects the distribution in two possible ways: (1) "Cutting off" the tail of the distribution and/or (2) "Narrowing" the tail (see Figure 1 for a cartoon illustration). The mechanisms leading to this, apart from finite sampling bias (as already proposed in Shumailov et al. (2023) - see Section 2 for a derivation in the Zipf-setting), stem from deliberate choices in the generation algorithms of the models: in LLMs via truncated next token prediction at inference (e.g. selecting more likely tokens via _top-_ or _top-_\(k\)_ truncation, concentrating the probability distribution by lowering the temperature); in vision models like GANs via truncation or in diffusion models through guidance.
The quantization level is strictly bound to the memory usage and inference speed, and thus a trade-off must be made when choosing a proper quantization level. As we can see, the 8-bit quantization method has almost the same or even lower perplexities compared to the original FP16 model, demonstrating that it is a good choice for deploying LLMs on personal computers, with only half size of the FP16 one. The 6-bit models also achieve decent PPLs comparable to the 8-bit one, making it a better balance of speed and performance. When we use a more aggressive quantization level, the performance drastically decreases (i.e., higher PPL), especially for 3-bit and 2-bit. We also discover that larger models are less sensitive to quantization methods than smaller ones. For example, the performance of 33B models changes much more mildly than the others. A similar result is also observed when comparing Plus-7B and Plus-13B models. This might indicate that though 2-bit and 3-bit quantization are less effective for smaller models, it might be a promising way to deploy larger models without significant performance loss. This is extremely helpful when the users only have limited computing resources and still want to try large language models. This might also imply that the quantized training method may become a main-stream approach for training large language models, especially for those with limited training resources.
Figure 2: Impact of deduplicating the training set on validation perplexity. We plot the results from T5 XL (see Appendix for base-sized model). For C4, we evaluate on _C4 Original_, the original validation set; _C4 Unique_, a subset of the validation set identified by NearDup as having zero matches across C4; and _C4 Duplicates_, a subset of the validation set identified by NearDup as having a match in the C4 train set.
Nonetheless, it remains elusive to figure out an ideal training data mixture. Most existing practices tune the mixture through heuristics to upsample a proportion of high-quality or underrepresented data without disclosing the concrete criteria in detail (Gao et al., 2020; Touvron et al., 2023; Bai et al., 2023; Bi et al., 2024) and it is hard to predate whether these data strategies are effective before finishing the training run. Encouraged by advances in scaling laws that show model losses on a given set of evaluation data are quantitatively predictable for a wide range of variables (Kaplan et al., 2020; Hoffmann et al., 2022), we wonder whether this also holds for mixture proportions, so that _we can estimate the outcome model performance given any mixture before actually training on them, including the desired one that reaches minimum loss._In this paper, we answer this proposition affirmatively.
where \(y_{s}\) represents the generated tokens at the current step \(s\). Among the researchers who have employed this method, the most renowned are Jiang et al. (2023), He et al. (2022), and Khattab et al. (2022).
Finally, in section 7, we illustrate LR re-warming can cause unwanted forgetting, introduce infinite learning rate schedules as a promising way to circumvent it, and compare these schedules to baselines.
We now study mixing of clean and synthesized data in the bigram setting. Figures 17 and 18 add top-p tail-cutting when synthesizing, and start with \(T_{0}=10,000\) original data samples, which are successively blended with synthesized data from the largest model. Note that in this setting we observe a reversion of scaling laws with increased AI data. This needs to be compared with the orange curve in Figure 20 in the deterministic Hutter setting. The probabilistic nature of the bigram models leads to a new effect here.
(2) Knowledgeable: Whether the model can accurately use various and detailed knowledge for problem-solving.
6. Evaluate whether your calculated probability is excessively confident or not confident enough: The calculated probability seems confident, but it is based on concrete data and reports from reliable sources. However, it is important to consider the possibility of unforeseen factors that could affect the final outcome. While the probability is high, it should not be absolute to account for any potential uncertainties.