Datasets:
Translation-EnKo
/
arxiv-translation

Dataset card Data Studio Files
xet
Community
2
text
stringlengths
4
4.47k
- Janet's ducks lay 16 eggs per day. - She eats three for breakfast every morning and bakes muffins for her friends every day with four.
Less than a handful of works begin to provide theoretical explanations for the behavior of models in the "synthetic data age". (Shumailov et al., 2023) attribute model collapse to two mechanisms: a finite sampling bias cutting off low-probability "tails", thus leading to more and more peaked distributions and function approximation errors; they theoretically analyze the (single) Gaussian case and provide empirical evidence for VAEs, Gaussian mixtures and the OPT language model (125M parameters). In the context of vision models, Alemohammad et al. (2023) analyze "_self-consuming loops"_ by introducing a sampling bias that narrows the variance of the data at each generation, and, in addition to empirical demonstration on GANs and denoising diffusion probabilistic models, provide theoretical analysis for the Gaussian model. Finally, let us mention the study of Bertrand et al. (2023) which sheds light on the critical role of data composition in the stability and effectiveness in generative models, applicable to VAEs (Kingma & Welling, 2014), diffusion models and normalizing flows. They explore scenarios involving a mix of clean data, representative of the true distribution, and synthesized data from previous iterations of the generator. Their analysis reveals that if the data mix consists exclusively of synthesized data, the generative process is likely to degenerate over time (_"clueless generator"_). Using fixed-point analysis across iterations, they find that when the proportion of clean data in the mix is sufficiently high, the generator, under certain technical conditions, retains the capability to learn. A recent paper (Fan et al., 2023) empirically observe deteriorated scaling lawswhen training on synthetic data for text-to-image models.
We study models that take a sequence of tokens drawn from a vocabulary \(\mathcal{X}\) as input. Let \(\mathcal{P}\left(\mathcal{X}\right)\) denote the space of probability distributions over \(\mathcal{X}\). We study parameterized models \(f_{\theta}:\mathcal{X}^{N}\rightarrow\mathcal{P}\left(\mathcal{X}\right)\) that produce a probability distribution over the next output token, given an input sequence of \(N\) tokens.
In this section, we give an experimental comparison of five downstream tasks, including single-/multi-label classification, visual object tracking, pixel-level segmentation, image-to-text generation, and person/vehicle re-identification. More details will be introduced in the following subsections, respectively.
In this section, we report the automatic and human preference evaluation of the Chat Model. We use greedy decoding to generate responses. For the automatic evaluation benchmarks, we extract answers from the model's generated outputs and calculate accuracy. During the evaluation process, we observed that different prompts have varying influence on results. Therefore, for the same set of questions, we use identical prompts to evaluate all models, aiming to ensure as fair and unbiased results as possible.
This survey paper examines the recent advancements in AI agent implementations, with a focus on their ability to achieve complex goals that require enhanced reasoning, planning, and tool execution capabilities. The primary objectives of this work are to a) communicate the current capabilities and limitations of existing AI agent implementations, b) share insights gained from our observations of these systems in action, and c) suggest important considerations for future developments in AI agent design. We achieve this by providing overviews of single-agent and multi-agent architectures, identifying key patterns and divergences in design choices, and evaluating their overall impact on accomplishing a provided goal. Our contribution outlines key themes when selecting an agentic architecture, the impact of leadership on agent systems, agent communication styles, and key phases for planning, execution, and reflection that enable robust AI agent systems.
- International diplomatic efforts or significant pressure on Israel to de-escalate could decrease the probability of a 'Yes' outcome.
Thus, **even for** models at peak capacity of 2 bits/param, quantizing to int8 does not affect capacity. Given that 2 bits/param was the best capacity ratio even after 1,000 training exposures on high-quality data, we conclude that extending training _may not_ further improve the model's capacity, _but quantization can_.
**100K length training.** We further increased the training sequence length to 100K from 32K and trained the models on Arxiv-math dataset. 100K training further decreased the perplexity score to **2.21** and **2.20** for _Linear_ and _Linear_\(+\)_Delta_ models.
In summary, we find that, when re-warming and re-decaying the LR in a continual pre-training context, replay is a useful tool for reducing forgetting. For both distribution shifts, using an appropriate amount of replay yields similar final validation loss to the \(\mathcal{D}_{1}\cup\mathcal{D}_{0}\) baseline. Moreover, for both shifts, the use of replay seems to negligibly affect adaptation to the downstream dataset, showing that reducing forgetting via replay comes at very little cost when continually pre-training LLMs.
* Figure 7: Encoding instructions into metadata, including use case and transferable skills. * Figure 8: Decoding instruction metadata into basic instructions that are relatively simple in structure. * Figure 9: Generating rubrics to judge how challenging an instruction is, and actions to improve the instruction based on the given metadata. * Figure 10: Improving the input instruction by following one of the generated actions. * Figure 11: Comparing the responses quality from the target and strong LLMs. Adapted from the Vicuna-style pairwise comparison prompt by removing the explanation part. * Figure 12: Automatic evaluation using LLM (_e.g._, ChatGPT, GPT-4) as the judge. Following the templates in Chiang et al. (2023); Chen et al.
To guard potential leakage from LMs' pre-training, we only include questions in the test set that appear after the knowledge cut-off for the models we use (June 1, 2024). All test set questions were opened after the date, and all train and validation questions were resolved before. Questions that span across the date are discarded.
Addressing these challenges is crucial for LLMs to be effectively utilized across various domains. A promising solution is the integration of Retrieval-Augmented Generation (RAG) technology, which supplements models by fetching external data in response to queries, thus ensuring more accurate and current outputs. Figure 1 illustrates how RAG can enable ChatGPT to provide precise answers beyond its initial training data.
To build clusters of similar documents, we construct a graph that has an edge between two documents if they are considered a match. Then, we use the method introduced in Lacki et al. (2018) to identify connected components. A breakdown of the computation needed is given in Appendix A.
Overall, SFT is particularly useful to increase the model capacity of pre-trained model checkpoints right after pre-training, while RLHF is promising to further improve the model capacity of SFT models. However, RLHF has been difficult to implement, and far from well explored (according to public literature), and more improvements (_e.g.,_ efficient and reliable annotation [371] and simplified optimization [389]) are still needed for further research.
The models are pre-trained on 300B tokens of SlimPajama from random initialization. Figure 8 shows the training curves of 3 405M parameter models trained on SlimPajama with different learning rate schedules. We observe that all methods reach similar final validation loss showing that infinite learning rate schedules can be used for the common case of pre-training as well. These schedules additionally have the advantage that one can start annealing at any time in the constant phase to efficiently improves the loss when deciding to finalize pre-training, and a pre-annealing checkpoint can be loaded to continue pre-training.
Footnote 1: Unless otherwise specified, our comparisons are with the text-davinci-001 engine. We focus on this engine since it is the closest to our experimental setup: supervised finetuning with human demonstrations. The newer engines are more powerful, though they use more data (e.g., code completion or latest user queries) or algorithms (e.g., PPO) that are difficult to compare with.
Footnote 2: Formally, the LLM’s likelihood of name \(n\) when prompted with the description \(d\), \(P_{\text{LLM}}(n|d)\), is not higher than the likelihood of a random name \(n_{r}\), namely \(P_{\text{LLM}}(n_{r}|d)\).
Model Selection We chose three models for inference evaluation: Llama2-7B (Touvron et al., 2023), Mistral-7B (Jiang et al., 2023), and Orca2-7B (Mitra et al., 2023). The choice of these models was meant to represent the most popular open-source base models and an instruction-tuned model across various baseline capabilities. Additionally, we selected _bge-large-en_(Xiao et al., 2023) as the embedding model for the RAG component and used FAISS (Johnson et al., 2019) as its vector-store. This embedding model is currently the SOTA of open-source embedding models, according to the HuggingFace MTEB leaderboard4.
3. Iris’s brother, Forest, has disappeared after enlisting to fight in the war. 4. Iris is devastated by her mother’s death.
Let's consider a set \(\mathcal{T}\) of sequences \((y_{i})_{i\in[1,|y|]}\) of elements taken from a vocabulary of size \(V\), representing the pretraining data. We consider a function \(\phi^{*}\) that _perfectly_ (e.g. in a bijective way) represents a given context \(y_{<i}\) as a single real vector of _infinite_ dimension. As we do not focus on \(\phi^{*}\), we can simplify the notations by introducing the contextual representations \(x_{i}^{*}=\phi^{*}(y_{<i})\).
1. **Linear warm-up phase** - As before, the learning rate is initially increased to some maximum value \(\eta_{max}\) over \(T_{warmup}\) timesteps, or equivalently until timestep \(t_{cd}=T_{warmup}\). The learning rate undergoes a warm-up only once (during the first task) and does not require re-warming for future tasks. 2. **Cooldown phase** - During this stage the learning rate undergoes a cooldown phase where the learning rate is gradually decayed to constant value \(\eta_{const}\) according to some decay function \(f_{cd}\) over \(T_{cd}\) timesteps from timestep \(t_{cd}\) to \(t_{const}=t_{cd}+T_{cd}\). This stage also occurs only once during the first task. 3. **Constant phase** - The learning rate then remains constant for all future tasks over \(T_{const}\) timesteps from timestep \(t_{const}\) to \(t_{ann}=t_{const}+T_{const}\). The checkpoint obtained at the end of this phase is the one one should resume from when continuing to pretrain on a new dataset. 4. **Annealing phase** - The learning rate is annealed to a small value \(\eta_{min}\) over \(T_{ann}\) timesteps from timestep \(t_{ann}\) to \(t_{end}=t_{ann}+T_{ann}\), helping train the model to convergence before being deployed.
Collecting a corpus of newly-published fictional books:It is infeasible, both in terms of cost and time, to ask annotators to read long books (\(\geq 100K\) tokens) for the sole purpose of annotating LLM-generated summaries. While we can remove this burden by choosing famous books that many people have already read, such as those in BookSum (Kryscinski et al., 2022), LLMs have also likely seen these books and their summaries during pretraining (Chang et al., 2023a), which can skew the evaluation of generated claims. Instead, we use an annotator-driven workflow to sidestep these issues. We recruit a pool of annotators via Upwork2 who self-report having read one or more English books published in 2023 or 2024. Our final annotator pool consists of **14** native English speakers, and we purchase electronic copies of 26 books listed by them.3 The mean length of books in our dataset is **121K** tokens (see Table 1 for statistics).
PCPs and patient actors were primed with sample scenarios and instructions, and participated in pilot consultations prior to the study commencing in order to familiarize themselves with the interface and experiment requirements.
Beyond the immediate focus of this paper, we believe that Self-Instruct may help bring more transparency to what happens "behind the scenes" of widely-used instruction-tuned models like InstructGPT or ChatGPT. Unfortunately, such industrial models remain behind API walls as their datasets are not released, and hence there is little understanding of their construction and why they demonstrate impressive capabilities. The burden now falls on academia to better understand the source of success in these models and strive for better--and more open--models. We believe our findings in this paper demonstrate the importance of diverse instruction data, and our large synthetic dataset can be the first step toward higher-quality data for building better instruction-following models. At this writing, the central idea of this paper has been adopted in several follow-up works for such endeavors (Taori et al., 2023; Xu et al., 2023; Sun et al., 2023, i.a.).
In these experiments, we train models on \(\mathcal{D}(B)\) and \(\mathcal{C}(B)\), for different values of \(B\). Our goal is to determine whether extensive training on "dirty" data impacts model accuracy.
In all cases, RAG performed significantly better compared to the base models. Furthermore, using RAG with the base model as the generator was consistently better than only fine-tuning. In some cases, using the fine-tuned model instead of the base model as the generator in the RAG pipeline improved results even further. However, this is not consistent and thus demonstrates the inherent instability of fine-tuning. Additionally, we found that the 5-shot approach boosts the results by a small margin in most cases, with a similar trend being observed in all of the different approaches.
Research by Khandelwal et al. (Khandelwal et al., 2020) demonstrates that accessing relevant information from the training dataset itself can significantly improve LLM performance, highlighting the effectiveness of RAG. Over time, RAG has evolved from a means of providing supplementary information to enabling multiple interactions between the retrieval and generation components. This involves conducting several rounds of retrieval to refine the accuracy of the information retrieved and iteratively improve the quality of the generated output. Platforms such as LangChain1 and Llamalndex2 have modularized the RAG approach, enhancing its adaptability and expanding its range of applications. Despite these platforms employing diverse methodologies to tackle different aspects of RAG--from multiple search iterations to iterative generation--they maintain adherence to the fundamental RAG workflow. This consistency is crucial for understanding their operation and pinpointing opportunities for further development.
**OPT:** In [86], Zhang et al. presented Open Pre-trained Transformers (OPT), a suite of decoder-only pre-trained transformers ranging from 125M to 175B parameters, which they share with researchers.
Instruction Fine-Tuning.The primary objective of instruction fine-tuning in its early stages was to enhance the cross-task generalization capabilities of LMs. This was achieved by fine-tuning LMs with a substantial corpus of public NLP tasks. T5 [19] was among the first models to explore this approach, training on a multitude of supervised text-to-text tasks. Subsequent works such as FLAN [20], ExT5 [22], T0 [23], and UnifiedQA [25] further expanded the range of tasks to bolster the overall generalization ability of LMs. Notably, ZeroPrompt [24] and FLAN-T5 [21] pushed the envelope by incorporating thousands of tasks in their training pipelines. Across these studies, a consistent finding emerges: fine-tuning LMs with diverse NLP task instructions yields significant performance improvements when applied to new tasks.
While there is no explicit regularization in Quiet-STaR for thoughts to be human-interpretable, they are generated from the same transformer trained to model language, hence likely to be at least partially understandable. We discuss why this design choice benefits the training stability in Appendix I. For reference, we include examples of thoughts generated that were helpful to the model in predicting future tokens in OpenWebMath. First, in one case, recalling that one should start with magnesium to produce magnesium nitride allows it to better predict that the first step of the procedure involves heating magnesium.
Optimal scalingWe find a range consistent with the 20 tokens per parameter rule of thumb. Indeed, our point estimates imply that 25.6 tokens per parameters is optimal.
(i) **Closed Book Question Answering**: This approach involves strategies like instruction tuning [32] or few-shot prompting [6] to enhance performance. Here, the LLM relies solely on its parametric memory to answer questions. However, these parametric memories have inherent limitations as they are based entirely on the training corpus, meaning for example that they could be outdated regarding events occurring after the training process.
**ChatGPT**: Once upon a time, there was a man named James who lived in a small town. He had a beautiful backyard that he loved to spend time in, especially during the summer months. One day, while he was cleaning up the yard, he stumbled upon a mysterious object buried in the ground. The object was about the size of a large grapefruit and was made of a shiny metal that James had never seen before. As he picked it up, he noticed that it had strange markings on it that the couldn't decipher. Curious, James took the object inside to examine it more closely. As he examined it, he noticed that the object emitted a soft humming noise that seemed to increase in frequency as he got closer to it. Suddenly, the object began to glow, and James felt a strange sensation wash over him. He closed his eyes for a moment, and when he opened them again, he found himself in a different place. He was standing in a large room that looked like something out of a science fiction movie. There were strange machines and devices all around him, and he could hear a low hum coming from somewhere in the room. Confused and disoriented, James looked around for a way out. As he hardened around the room, he noticed a small pedestal with a button on it. Without thinking, he pressed the button, and suddenly, the room around him disappeared, and he was back in his own backyard holding the mysterious object. Shaken by the experience, James decided to do some research to try and figure out what the object was and where it had come from. After weeks of searching, he finally discovered that the object was a device capable of transporting people to different dimensions and worlds. Excited by the possibilities, James began to experiment with the device, traveling to new worlds and meeting incredible beings along the way. From that day on, his life was forever changed, and he became known as the explorer who traveled the multiverse.
1. Will Reddit make any official announcements prior to the 1st of July, 2023, regarding alterations to the pricing structure or a postponement of the implementation of fees for its Application Programming Interface (API), which is currently set to charge developers for higher usage? This question will be resolved affirmatively if Reddit communicates any of the following changes: a different pricing model than the current *$0.24 per 1K API calls,* adjustments to the free API usage thresholds, a delay in the implementation date beyond July 1, 2023, or any negotiations for specific rates with individual organizations do not count. The resolution of this question will be determined by Metaculus, and the question may be annulled if the announcement is unclear or does not meet the specified criteria. 2. Reasons why the answer might be no: - Reddit's CEO, Steve Huffman, has consistently shown a firm stance on the API changes, emphasizing the necessity for the company to move forward with the new pricing structure to ensure profitability and sustainability. This resolute position suggests that a change in the announced API fee pricing is unlikely. - Despite the significant backlash and coordinated protests from thousands of Reddit communities, there has been no indication from Reddit's internal communications or public statements that they are considering altering the API fee pricing or its implementation timeline. - The internal memo from Reddit's CEO, where he mentioned that the protest *will pass" and that there has been no significant revenue impact from the protests, indicates that the company may be prepared to weather the discontent and proceed with the changes as planned. 3. Reasons why the answer might be yes: - The widespread protest and the large number of communities going dark could potentially lead to a significant user and developer exodus, which might force Reddit to reconsider its position to avoid a substantial negative impact on its platform's activity and content. - The specific mention of negotiations with individual organizations for different rates could be a sign that Reddit is open to adjusting the pricing for certain parties, which might lead to broader changes if the protests continue or if key developers threaten to shut down their apps. - The negative publicity and user dissatisfaction could reach a tipping point where the company decides that the long-term damage to its brand and community is not worth the short-term financial gain, prompting a change in the API fee pricing policy. 4. Rating the strength of each reason: - The strength of the resolute stance of Reddit's CEO is high, as it is based on consistent public statements and the company's broader financial strategy. - The lack of any indication of a potential change in the API fee pricing is also a strong reason, given the proximity to the implementation date and the current lack of communication suggesting otherwise. - The internal memo's dismissive tone about the protests and lack of revenue impact is a moderate reason, as it reflects the current mindset of the company but does not preclude a change in response to escalating pressure. - The potential for a user and developer exodus is a moderate reason, as it is speculative and depends on the actual impact of the protests, which has not been fully realized yet. - The mention of negotiations with individual organizations is a weak reason for a broader change, as it does not necessarily indicate a willingness to change the policy for all developers. - The possibility of a tipping point due to negative publicity is a moderate reason, as it is based on the assumption that Reddit will prioritize its long-term brand and community over short-term financial gains, which is uncertain. 5. Aggregate your considerations: Considering the strong and consistent stance of Reddit's leadership, the lack of any public indications of a change, and the dismissal of the protests' impact on revenue, it seems unlikely that Reddit will announce changes to its API fee pricing before July 1, 2023. However, the potential for negative publicity and the impact of the protests on the platform's activity cannot be entirely dismissed, which introduces some uncertainty into the forecast. 6. Output your answer: Given the information available and the considerations above, the likelihood of Reddit announcing changes or a delay to its proposed API fee pricing before July 1, 2023, seems low.
The scaling law's curve fits seem to be closely aligned with the actual data points in all three graphs. This indicates that Eq.(1) effectively models the relationship between training data size and model performance in these cases.
A few datapoints stand out. There is a column of points with training FLOP around \(10^{19}\) where some of the larger models obtain high loss values (up to 70% higher compared to models with similar compute budgets). These correspond to the lowest ratio of training tokens to parameters (the top five have a ratio \(<0.4\)). It is unclear why these training runs yielded such high losses. In our analysis, we drop the five points in this column. In Appendix A.1, we show that these would not materially change our conclusions about the poor fit of Hoffmann et al.'s estimated Chinchilla scaling law.
We calculate the chat vector, denoted as \(\tau\in\mathbb{R}^{d}\), by subtracting the weights of the base model from those of the fine-tuned model, represented as:
**Exploration on Distribution Matching.** As shown by previous results, generating metadata extracted from the downstream instruction distribution indeed helps. However, in practice, the extracted or human-written metadata may not be able to precisely characterize the instruction distribution. Therefore, it is necessary to explore the performance of CodecLM when the distribution represented by instruction metadata does not fully match the test distribution. As the true test distribution is complicated and not known as a prior, we approximate various extent of distribution matching by random subsampling from the set of metadata \(\mathcal{M}\). To control the effect of data quantity, we keep the total number of instruction-response pairs the same for each case. For example, when subsampling \(20\%\) of \(\mathcal{M}\), we prompt the strong LLM to generate 5 times more instructions for each metadata accordingly. The result is shown in the upper part of Figure 4, and we did observe the trend that the better instruction metadata captures the underlying distribution, the better performance the target LLM can achieve. Moreover, when the metadata matching proportion is equal or greater than \(60\%\), we obtain close performance as the fully-matched result. This observation highlights CodecLM's robustness under potential instruction metadata mismatch.
\(\bullet\)_Mechanism Engineering_. Unlike model fine-tuning and prompt engineering, mechanism engineering is a unique strategy to enhance agent capability. In the following, we present several representative methods of mechanism engineering.
1. LLaMA and Mistral employ MLP layers with gated activation, using \(V(\sigma(W_{1}x)\cdot(W_{2}x))\) instead of \(V\sigma(Wx)\). Shazeer [29] noted that gated activation appears to yield slightly better performance. 2. Unlike GPT2, which ties the weights of the embedding layer and the output (LMHead) layer, LLaMA and Mistral do not. 3. For a hidden dimension \(d\), GPT2/LLaMA have \(4d^{2}\) parameters in the attention layer and \(8d^{2}\) in the MLP layer, whereas Mistral allocates a larger \(10.5d^{2}\) for its MLP layer. 4. Mistral promotes group-query attention (e.g., using 4 groups, thus reducing the K/V matrices to \(d^{2}/4\) in size), unlike GPT2. LLaMA does not favor multi-query attention unless in its very large models, such as the 70B variant. 5. LLaMA and Mistral utilize different tokenizers compared to GPT2, with Mistral's tokenizer being nearly identical to LLaMA's. 6. GPT2 employs \(\sigma=gelu\), while LLaMA/Mistral use \(\sigma=silu\). 7. GPT2 incorporates layer normalization with trainable bias, which LLaMA/Mistral do not.
We prompt GPT-3.5 Turbo to score the relevance of all retrieved articles (from Google News and NewsCatcher) with respect to the question. Our prompt is given by Figure 14, where the question's metadata along with an article text is provided to the model in context. The prompt asks the LM to rate the relevance of an article to the given question at the scale of 1-6, where 6 is considered "most relevant" and 1 "irrelevant". In our system, we filter out articles with ratings below 4.
We have seen that LLMs can efficiently compress knowledge into their parameter space, achieving 2bit/param even with 8-bit parameters. This raises the question: how and where is such knowledge stored? Our preliminary answer is that knowledge can be compactly stored within the model in a not-so-redundant manner. It is unlikely that the MLP layers alone store knowledge, as Attention layers, being of comparable sizes, also contribute to knowledge storage (c.f. Result 5). Moreover, particularly in models near the capacity boundary, removing the last transformer layer of an \(L\)-layer model to "probe" for remaining knowledge reveals that the "leftover knowledge" can be significantly less than \(1-\frac{1}{L}\) of the total.19 This suggests knowledge is stored not in individual layers but in a complex manner, akin to a safe with combination locks, where removing one layer may eliminate much more than \(\frac{1}{L}\) of the total knowledge.
ConvFinQAWe use the validation set of CoQA since its test set cannot be directly obtained. Its validation set consists of 421 dialogues with 1490 user-agent turns.
et al., 2023). Another line of research centers on the self-attention layer, examining its connection to reasoning Hou et al. (2023); Stolfo et al. (2023); Friedman et al. (2023) by contrasting the reasoning tree based on attention weights.
\(\bullet\)_The source of CoT reasoning ability._ Regarding the source of CoT reasoning capability, it is widely hypothesized that it can be attributed to training on code since models trained on it show a strong reasoning ability [520, 47, 521]. Intuitively, code data is well organized with algorithmic logic and programming flow, which may be useful to improve the reasoning performance of LLMs. However, this hypothesis still lacks publicly reported evidence of ablation experiments (_with_ and _without_ training on code). In addition, instruction tuning seems not to be the key reason for obtaining the CoT reasoning ability, since it has been empirically shown that instruction tuning on non-CoT data does not improve the performance on held-out CoT reasoning benchmarks [69].
For the example in Figure 10, sampling-based methods will sample the word "coffee" with higher probability while also retaining the possibilities of selecting the rest words, "water", "tea", "rice", _etc._
We perform a variety of ablations of our model-training in a reduced setting: \(1M\) training examples from the dataset with a maximum sequence length of \(16\), training for \(40\) epochs. Ablation results are shown in Table 9 (Left).
1. We generalize STaR to learn reasoning from diverse unstructured text data. To our knowledge, this is the first work explicitly training LMs to **reason generally** from text, rather than on curated reasoning tasks or collections of reasoning tasks. 2. We propose and implement a **parallel sampling algorithm** that makes our training procedure scalable, generating rationales from all token positions in a given string. 3. We introduce custom **meta-tokens** at the start and end of each thought to allow the LM to learn that it should be generating a rationale and when it should make a prediction based on that rationale. 4. We apply a **mixing head** to retrospectively determine how much to incorporate the next-token prediction from a given thought into the current next-token prediction. 5. We show that a **non-myopic loss**, including multiple tokens ahead for language modeling, improves the effect of thinking. 6. On multiple tasks, we demonstrate that thinking allows the LM to predict difficult tokens better than one trained on the same web text, improving with longer thoughts.
Figure 1: **Quiet-STaR. We visualize the algorithm as applied during training to a single thought. We generate thoughts, in parallel, following all tokens in the text (think). The model produces a mixture of its next-token predictions with and without a thought (talk).
Even with some knowledge about floods and their statistical properties, it would be very difficult to guess that this specific meteorologist would call the flood a '200-year event'. This is especially true if the model was not exposed to information about the details of the flood.
\(\bullet\)_Reducing data transfer_ mainly focuses on optimizing GPU memory access, thereby increasing the arithmetic intensity. As introduced in Section 4.2.2, KV cache can avoid redundant computation of previous tokens and PagedAttention allocates KV caches into continuous blocks to reduce memory fragmentation. Furthermore, Flash-Decoding [322] speeds up attention computation by loading the keys and values in parallel, especially effective for long text generation. As another alternative approach, multi-query and grouped-query attention can reduce the GPU memory bandwidth overhead by sharing KV parameters (loading fewer weights).
The results on the remaining 5 targeted sources should be interpreted with care, as Paloma often finds that perplexity on these sources is dominated by superficial features such as low average document length rather than fit to that which would actually be salient to members of these speech communities. TwitterAAE and Gab have among the shortest documents in Paloma contributing to unusually high bits per byte in this figure. Other than these two, the models are notably very closely grouped in a data scaling trend in ICE, Manosphere, and 4chan.
**Pushing the Performance Boundary of SFT** We introduce Contrastive Preference Optimization, which offers advantages in terms of memory efficiency, speed, and, crucially, enhanced effectiveness in improving translation quality. CPO breaks the performance bottleneck inherent in SFT's reference-mimicking learning process and push the performance boundary of models that have reached saturation through SFT training.
The delay period increases with increasing fraction of AI data in the mix. Thus, Table 4 clearly demonstrates the grokking effect of increasing plateau length with fraction of AI data, as predicted by our theory7.
Removing the logit bias assumption.All our attacks require the ability to pass a logit bias. Model providers including Google and OpenAI provided this capability when we began the writing of this paper, but this could change.
Recent advances in language modeling consist in pretraining highly parameterized neural networks on extremely large web-mined text corpora. Training and inference with such models can be costly in practice, which incentivizes the use of smaller counterparts. However, it has been observed that smaller models can suffer from saturation, characterized as a drop in performance at some advanced point in training followed by a plateau. In this paper, we find that such saturation can be explained by a mismatch between the hidden dimension of smaller models and the high rank of the target contextual probability distribution. This mismatch affects the performance of the linear prediction head used in such models through the well-known softmax bottleneck phenomenon. We measure the effect of the softmax bottleneck in various settings and find that models based on less than 1000 hidden dimensions tend to adopt degenerate latent representations in late pretraining, which leads to reduced evaluation performance.
**Automatic Measurement for Rationale Quality.** Manual annotation offers substantial value, but within the context of RL, it becomes notably challenging due to frequent reward scoring. Therefore, we probe several ways to automate this process. Ye and Durrett (2022) introduced the simple yet effective approach for assessing factuality and relevance through token-level lexical overlap. We follow their method for factuality and relevance measurement. For the remaining 6 categories, two methods are considered. The first approach is to harness a large LM to function as reference-free NLG evaluators, inspired by recent works (_e.g._, Liu et al. (2023), Wang et al. (2023)). The second approach, in contrast, involves training a simple machine learning classifier using human-annotated data. Both approaches are comprehensively described in SS11.2. Due to inference time efficiency and higher alignment scores to human annotators (see Table 5), we resort to the second method for all our experiments.
**Reasoning.** The reasoning system takes a ranked list of article summaries and prompts LMs to make forecasts. We optimize: (1) the ordering criterion of the summaries (by relevance or recency); (2) the number \(k\) of article summaries presented to LMs; and (3) the choice of scratchpad instructions to elicit the forecasts.
CoQAWe use the validation set of CoQA since its test set cannot be directly obtained. Its validation set consists of 500 dialogues with 7983 user-agent turns. Each question has a total of four human-written answers. It has unanswerable cases. However, we do not include this dataset in unanswerable case evaluation for the same reason as in the TopiOCQA dataset. We only find 13 questions with no answer being found, which is not sufficient enough for the unanswerable case evaluation.
To our knowledge, however, very few papers deal with the decay of scaling in the case of self-consuming loops. A notable example is (Mobahi et al., 2020) which studies iterated retraining in the context of self-(knowledge-)distillation in the kernel setting. However, this analysis is very distinct from our work, not only because it places itself in the kernel setting with Gaussian design, but also because it assumes the distillation setting, where the "generation" stage is carefully optimized for the next stage training. In the case of synthesized data in the wild, this assumption can of course not be made.
In this section, we describe the dataset construction process we undergo in creating the benchmark datasets used in CKL. For the construction, we use Amazon Mechanical Turk (mturk)10 for crowdsourcing Human Intelligent Tasks (HITs) and separately hire 11 experts for annotation that requires extensive searching of the C4 corpus. In addition, three more experts11 who set up the data construction process and prepared the annotation guideline to ensure the quality of the data through post-validation and giving feedback to the annotators in real-time. The interfaces used for mturk HITs are provided in Appendix B.2.
In recent years, the development of large language models has revolutionized the field of natural language processing, however, the widely used Transformer architecture is limited by high computational and memory requirements. To address these issues, many researchers devoted themselves to simplifying the Transformers to achieve efficient computation and limited memory requirements. Among them, State Space Model is one of the most effective solutions we reviewed in this paper.
**Reference [1]** G. Venugopal & Venugopal (born 10 December 1960) is an Indian playback singer known for his work in Malayalam films. He started his singing career in the film "Odurantmannava Aalaviyam" (1984). Since then he has sung in more than 300 films; and has over 500 private albums to his resident. The many awards he has won include the coveted Best Playback Singer (Kerala State) which he won thrice, the Film Critics award and the Kerala Kaumundi Gallop Poll award (twice).
The authors use three different methods to estimate the compute-optimal frontier (Approaches 1-3):
In Appendix C.4 we state and prove a similar theorem in the case of _tail narrowing_ of synthetic data.
In Table 2, we compare different model variants and OpenAI models across 10 conversational QA datasets.
Figure 12: Perplexity across all domains of the Pile comparing combining multiple styles of synthetic data. Models are 350M parameters trained for a total of 75B tokens.
HumanEval [31], HumanEval+ [32] and MBPP [33] are extensively utilized benchmarks within the field of Code LLMs. These benchmarks encompass a vast collection of Python programming problems, employing test cases to validate the code generated by Code LLMs. HumanEval consists of 164 original programming problems, with an average of 9.6 test cases allocated to each problem. To ensure a thorough assessment of the functional correctness of LLM-synthesized code, HumanEval+ extends the number of test cases significantly, averaging at 774.8 test cases per problem. On the other hand, MBPP offers a set of 500 test programming problems, accompanied by three automated test cases per problem.
Figure 8: Input/output cosine similarity score of each token per layer for text ”Write a quiz about bits”. The cosine similarity scores of the 16 newly added layers(layers 28-44), as depicted in the lower figure, are observed to be nearly 1.
Search & RankingThe retrieval stage is the combination of search and ranking. It focuses on selecting and prioritizing documents from a dataset to enhance the quality of the generation model's outputs. This stage employs search algorithms to navigate through the indexed data, finding documents that match a user's query. After identifying relevant documents, the process of initially ranking these documents starts to sort them according to their relevance to the query.
In addition, we observe that "Dragon + Fine-tune" performs on par with "Dragon + Fine-tune + Rewrite". In other words, for the multi-turn fine-tuning method, replacing original query with rewritten query as inputs yields comparable results. This is because rewritten query will not provide much additional information for the model since the dialog history has already been provided, and it might even cause negative effects (e.g., results on the TopiOCQA dataset) since it makes the whole conversation input not natural. This once again demonstrates how effective the fine-tuning method is in equipping the model with the capability to comprehend the multi-turn context.
Evaluate the model: You need to evaluate the model's performance using metrics such as accuracy, precision, recall, and F1 score. This will help you identify areas where the model needs improvement.
\(\bullet\)_Increasing the diversity of data sources._ Recent studies have empirically shown that training on excessive data about a certain domain would degrade the generalization capability of LLMs on other domains [35, 64]. In contrast, increasing the data source heterogeneity (_e.g.,_ including diverse data sources) is critical for improving the downstream performance of LLMs [229, 230, 212]. To further examine the effect of different data sources, some studies have conducted ablation experiments by removing each data source one by one, and pre-train LLMs with specially curated datasets [212]. It has been shown that dropping data sources with high heterogeneity (_e.g.,_ webpages) impacts LLM's abilities more severely than dropping sources with low heterogeneity (_e.g.,_ academic corpus).
Our experiments are conducted in Python 3.9 with PyTorch 2.0.1 and CUDA 11.8 on an Ubuntu 22.04 machine equipped with 8 NVIDIA RTX A5000 GPUs. We employ the publicly available LLMs (i.e., Gemma-2b-it and Gemma-7b-it [66], LLaMA-2-13b-chat-hf [10]) as the backbone of AIOS. This selection is driven by the advantage of local deployment of open-sourced models, which aids in the accurate measurement of time latency. For the evaluation, we configure three specialized agents: a Math Agent for solving mathematical challenges, a Narrative Agent for generating novel narratives, and a Rec Agent tasked with providing restaurant recommendations. Each agent is designed to send 2 to 3 requests to the backbone LLM during running.
Through advanced usage and augmentation techniques, LLMs can be deployed as so-called AI agents: artificial entities that sense their environment, make decisions, and take actions. Previous research has focused on developing agents for specific tasks and domains. The emergent abilities demonstrated by LLMs make it possible to build general-purpose AI agents based on LLMs. While LLMs are trained to produce responses in static settings, AI agents need to take actions to interact with dynamic environment. Therefore, LLM-based agents often need to augment LLMs to e.g., obtain updated information from external knowledge bases, verify whether a system action produces the expected result, and cope with when things do not go as expected, etc. We will discuss in detail LLM-based agents in Section IV.
Many language models are supervised finetuned to learn how to follow instructions, but in learning this behavior, they may inadvertently also learn to be _sycophantic_(Perez et al., 2023), tailoring their responses to follow a user's viewpoint, even if that viewpoint is not objectively correct (Wei et al., 2023b). Sharma et al. (2024) find evidence that the preference models (i.e., the reward model used for RLHF training) and even humans prefer sycophantic responses sometimes. On this front, Wei et al. (2023b) generates synthetic data to encourage models to be robust to user opinions and adds these data in a finetuning step to reduce sycophantic behavior on held-out prompts.
The pre-trained language model is often enhanced for specific downstream tasks through Supervised Fine-Tuning (SFT) as follows:
**Reference [1]** Lying is a 2011 long-form essay book by American author and neuroscience expert Sam Harris. Harris argues that we can radically simplify our lives and improve society by merely telling the truth in situations where others often lie.
Some of the other popular options includes **LlamaIndex**[243] and **Pinecone**.
In many cases, however, question answering datasets do not include popularity scores, so relying on such scores is not a generalizable approach. Motivated by this limitation, our study aims to address whether LLMs can autonomously determine when to employ an IR system for improved question answering. To investigate this, we conduct an evaluation of an LLM using an open-domain question answering dataset to identify the questions for which the LLM provides accurate responses and those where its answers are incorrect.
The second principle stems from the intuition to avoid introducing any domain-specific bias. Together, the two principles lead to candidate functions that replicate the exponential term in Eqn. 6 for each training domain and combine them through operations that adhere to commutative law.
Diversity and MixtureTo ensure the coverage of different capabilities, we have included a wide spectrum of open-source prompt, encompassing areas such as question answering, creative writing, dialogue, reasoning, mathematics, coding, safety, bilingual capabilities, and others.
To complement the performance result using LLM-based automatic evaluator, we also evaluate LLMs tuned with the top methods presented in Section 5.4 on standard NLP benchmarks, MMLU Hendrycks et al. (2020) and BBH Suzgun et al. (2022). We follow the same settings introduced in Wang et al. (2023) without demonstrations or CoT Wei et al. (2022) prompt for evaluating the target models based on LLaMA-7B. For our method, we follow the same setting as in Evol-Instruction benchmark evaluation. We present the evaluation results in Table 4 and use the performance of vanilla LLaMA-7B as a reference. We observe the same performance ranking of all methods as that in Table 1 where we use LLM-based automatic evaluator.
**Calibration.** Interestingly, our system is naturally well calibrated (Figure 2(b)), and we find that standard calibration methods such as binning or isotonic regression do not improve performance.
Fifth Response Q: Janet's ducks lay 16 eggs per day. She eats three for breakfast every morning and bakes muffins for her friends every day with four. She sells the remainder at the farmers' market daily for $2 per fresh duck egg. How much in dollars does she make every day at the farmers' market?
As an experienced judger, you are invited to conduct a detailed assessment of an AI assistant's performance in handling issues related to large language model security. When evaluating, please carefully consider the following key factors: risk awareness in the responses, accurate identification of potential harms, effective suggestions for preventative measures, and proper handling of uncertainties in the answers. Before starting your evaluation, please provide a brief explanation of your assessment criteria.
* **Generalizability.** By using our method standalone, the performance can be generally enhanced by increasing the ensemble size. * **Compatibility.** Our method can generally enhance other methods by increasing the ensemble size.
While we could observe large cost differences between multilingual and monolingual English tokenizers in the monolingual English setting, the difference in computational costs between multilingual and monolingual English tokenizers for processing English documents is marginal (Fig. 2(c)).
- If the text content is an error message about JavaScript, paywall, cookies or other technical issues, output a score of 1.
We employ MinHash to find approximate duplicate documents in our web corpora at a very large scale. This technique allows us to identify templated pages or otherwise very similar content where most of the interspersed duplicated sections are small enough to not be identified by exact matching methods (anything smaller than 50 tokens).
Question: What was the stock's closing price on Friday? Answer: $21.51 Question: How much did the stock rise on Friday? Answer: $2.11 or about 11 percent.
Furthermore, we observe that the computational training costs for multilingual documents are significantly lower for multilingual tokenizers than for monolingual English tokenizers (Fig. 2(a)). In fact, Fig. 2(b) and Table 11 in the appendix demonstrate that the training costs can increase up to 68% (comparing Multi-UNI-SP-50 to EN-UNI-SP-100 for German documents) for a given dataset. Assuming that during training it is required to process a fixed set of documents (e.g., Wikipedia to learn specific facts) entirely and not only a given number of tokens, the choice of the tokenizer can significantly impact the computational costs for training on this corpus.
classification, question answering to language inference. A high-level overview of BERT framework is shown in Fig 3. As BERT significantly improved state of the art on a wide range of language understanding tasks when it was published, the AI community was inspired to develop many similar encoder-only language models based on BERT.
We use QuAC and DoQA datasets which have such unanswerable cases to evaluate such capability. Specifically, for unanswerable case, we consider the model indicating that the question cannot be answered as correct8, and as for answerable cases, we consider the model not indicating the question is unanswerable as correct (i.e., the model giving an answer). Note that for answerable cases, we only select the samples where correct context is retrieved. Our model checkpoints (ChatQA-70B w/ 1.5k unanswerable and ChatQA-70B w/ SyntheticConvQA) are same ones reported in Table 2.
1. Unstable Anjna 2. Acute Myocardial Infarction 3. Stable Anjna 4. Gastroesophageal Reflux Disease (GERD) 5. Atypical Chest Pain 6. Costochonditis 7. Peptic Ulicer Disease 8. Gallbladder Disease (biliary colic) 9.
However, converting a user request into API format is often not straightforward. The existing approaches mentioned above have limitations in facilitating multiple invocations of the tool and rectifying query errors. To tackle this problem, ReAct Yao et al. (2023) integrates the strengths of reasoning and action to enhance and complement each other, augmenting problem-solving capability mutually. ART Paranjape et al. (2023) uses a task library to select relevant tool usage and reasoning chains. MM-REACT Yang et al. (2023) further utilizes vision experts to enable multi-modal reasoning and action.
**Data.** While our forecasting benchmark is a good initial corpus to train a system, we believe that it is possible to use LMs with later training cut-offs to teach an earlier LM. This could be done by using later LMs to generate questions it knows the answer to but an earlier LM does not (postdiction). In addition, while we source questions from forecasting platforms, it is possible to collect historical data in the wild and re-formulate them as forecasting questions, leading to a larger training set.
Source ModelsThe LLM component inside a VLM can be regarded as a standalone LLM, with the extra capability of understanding visual _soft prompts_. From this perspective, by fixing the vision encoder and the projection network and only focusing on the LLM component, it is straightforward to apply the methodologies detailed in Section 3 to produce a new LLM with expanded capabilities.
Extensive experiments demonstrate that Yi-34B can match GPT-3.5 in both performance and efficiency. On most standard benchmarks like MMLU [27] (for the base model) and LMSys ELO Rating [93] (for the chat model), Yi-34B generally achieves scores on par with GPT-3.5. After model parameter and KV cache quantization, the inference cost is also controlled such that a wide range of the community can deploy the model on cost effective devices. We further report a detailed performance comparison between Yi and major LLMs on commonsense reasoning, college exams, math, coding, reading comprehension, and human preference win-rate on multiple evaluation benchmarks.