Scaling Laws of Synthetic Data for Language Models

Large language models (LLMs) achieve strong performance across diverse tasks, largely driven by high-quality web data used in pre-training. However, recent studies indicate this data source is rapidly depleting. Synthetic data emerges as a promising alternative, but it remains unclear whether synthetic datasets exhibit predictable scalability comparable to raw pre-training data. In this work, we systematically investigate the scaling laws of synthetic data by introducing SynthLLM, a scalable framework that transforms pre-training corpora into diverse, high-quality synthetic datasets. Our approach achieves this by automatically extracting and recombining high-level concepts across multiple documents using a graph algorithm. Key findings from our extensive mathematical experiments on SynthLLM include: (1) SynthLLM generates synthetic data that reliably adheres to the rectified scaling law across various model sizes; (2) Performance improvements plateau near 300B tokens; and (3) Larger models approach optimal performance with fewer training tokens. For instance, an 8B model peaks at 1T tokens, while a 3B model requires 4T. Moreover, comparisons with existing synthetic data generation and augmentation methods demonstrate that SynthLLM achieves superior performance and scalability. Our findings highlight synthetic data as a scalable and reliable alternative to organic pre-training corpora, offering a viable path toward continued improvement in model performance.

One-shot In-context Part Segmentation

In this paper, we present the One-shot In-context Part Segmentation (OIParts) framework, designed to tackle the challenges of part segmentation by leveraging visual foundation models (VFMs). Existing training-based one-shot part segmentation methods that utilize VFMs encounter difficulties when faced with scenarios where the one-shot image and test image exhibit significant variance in appearance and perspective, or when the object in the test image is partially visible. We argue that training on the one-shot example often leads to overfitting, thereby compromising the model's generalization capability. Our framework offers a novel approach to part segmentation that is training-free, flexible, and data-efficient, requiring only a single in-context example for precise segmentation with superior generalization ability. By thoroughly exploring the complementary strengths of VFMs, specifically DINOv2 and Stable Diffusion, we introduce an adaptive channel selection approach by minimizing the intra-class distance for better exploiting these two features, thereby enhancing the discriminatory power of the extracted features for the fine-grained parts. We have achieved remarkable segmentation performance across diverse object categories. The OIParts framework not only eliminates the need for extensive labeled data but also demonstrates superior generalization ability. Through comprehensive experimentation on three benchmark datasets, we have demonstrated the superiority of our proposed method over existing part segmentation approaches in one-shot settings.

WildLong: Synthesizing Realistic Long-Context Instruction Data at Scale

Large language models (LLMs) with extended context windows enable tasks requiring extensive information integration but are limited by the scarcity of high-quality, diverse datasets for long-context instruction tuning. Existing data synthesis methods focus narrowly on objectives like fact retrieval and summarization, restricting their generalizability to complex, real-world tasks. WildLong extracts meta-information from real user queries, models co-occurrence relationships via graph-based methods, and employs adaptive generation to produce scalable data. It extends beyond single-document tasks to support multi-document reasoning, such as cross-document comparison and aggregation. Our models, finetuned on 150K instruction-response pairs synthesized using WildLong, surpasses existing open-source long-context-optimized models across benchmarks while maintaining strong performance on short-context tasks without incorporating supplementary short-context data. By generating a more diverse and realistic long-context instruction dataset, WildLong enhances LLMs' ability to generalize to complex, real-world reasoning over long contexts, establishing a new paradigm for long-context data synthesis.

Chain-of-Reasoning: Towards Unified Mathematical Reasoning in Large Language Models via a Multi-Paradigm Perspective

Large Language Models (LLMs) have made notable progress in mathematical reasoning, yet they often rely on single-paradigm reasoning that limits their effectiveness across diverse tasks. In this paper, we introduce Chain-of-Reasoning (CoR), a novel unified framework that integrates multiple reasoning paradigms--Natural Language Reasoning (NLR), Algorithmic Reasoning (AR), and Symbolic Reasoning (SR)--to enable synergistic collaboration. CoR generates multiple potential answers using different reasoning paradigms and synthesizes them into a coherent final solution. We propose a Progressive Paradigm Training (PPT) strategy that allows models to progressively master these paradigms, culminating in the development of CoR-Math-7B. Experimental results demonstrate that CoR-Math-7B significantly outperforms current SOTA models, achieving up to a 41.0% absolute improvement over GPT-4 in theorem proving tasks and a 7.9% improvement over RL-based methods in arithmetic tasks. These results showcase the enhanced mathematical comprehensive ability of our model, achieving significant performance gains on specific tasks and enabling zero-shot generalization across tasks.

Bootstrap Your Own Context Length

We introduce a bootstrapping approach to train long-context language models by exploiting their short-context capabilities only. Our method utilizes a simple agent workflow to synthesize diverse long-context instruction tuning data, thereby eliminating the necessity for manual data collection and annotation. The proposed data synthesis workflow requires only a short-context language model, a text retriever, and a document collection, all of which are readily accessible within the open-source ecosystem. Subsequently, language models are fine-tuned using the synthesized data to extend their context lengths. In this manner, we effectively transfer the short-context capabilities of language models to long-context scenarios through a bootstrapping process. We conduct experiments with the open-source Llama-3 family of models and demonstrate that our method can successfully extend the context length to up to 1M tokens, achieving superior performance across various benchmarks.

RedStone: Curating General, Code, Math, and QA Data for Large Language Models

Pre-training Large Language Models (LLMs) on high-quality, meticulously curated datasets is widely recognized as critical for enhancing their performance and generalization capabilities. This study explores the untapped potential of Common Crawl as a comprehensive and flexible resource for pre-training LLMs, addressing both general-purpose language understanding and specialized domain knowledge. We introduce RedStone, an innovative and scalable pipeline engineered to extract and process data from Common Crawl, facilitating the creation of extensive and varied pre-training datasets. Unlike traditional datasets, which often require expensive curation and domain-specific expertise, RedStone leverages the breadth of Common Crawl to deliver datasets tailored to a wide array of domains. In this work, we exemplify its capability by constructing pre-training datasets across multiple fields, including general language understanding, code, mathematics, and question-answering tasks. The flexibility of RedStone allows for easy adaptation to other specialized domains, significantly lowering the barrier to creating valuable domain-specific datasets. Our findings demonstrate that Common Crawl, when harnessed through effective pipelines like RedStone, can serve as a rich, renewable source of pre-training data, unlocking new avenues for domain adaptation and knowledge discovery in LLMs. This work also underscores the importance of innovative data acquisition strategies and highlights the role of web-scale data as a powerful resource in the continued evolution of LLMs. RedStone code and data samples will be publicly available at \url{https://aka.ms/redstone}.

Preference Optimization for Reasoning with Pseudo Feedback

Preference optimization techniques, such as Direct Preference Optimization (DPO), are frequently employed to enhance the reasoning capabilities of large language models (LLMs) in domains like mathematical reasoning and coding, typically following supervised fine-tuning. These methods rely on high-quality labels for reasoning tasks to generate preference pairs; however, the availability of reasoning datasets with human-verified labels is limited. In this study, we introduce a novel approach to generate pseudo feedback for reasoning tasks by framing the labeling of solutions to reason problems as an evaluation against associated test cases. We explore two forms of pseudo feedback based on test cases: one generated by frontier LLMs and the other by extending self-consistency to multi-test-case. We conduct experiments on both mathematical reasoning and coding tasks using pseudo feedback for preference optimization, and observe improvements across both tasks. Specifically, using Mathstral-7B as our base model, we improve MATH results from 58.3 to 68.6, surpassing both NuminaMath-72B and GPT-4-Turbo-1106-preview. In GSM8K and College Math, our scores increase from 85.6 to 90.3 and from 34.3 to 42.3, respectively. Building on Deepseek-coder-7B-v1.5, we achieve a score of 24.6 on LiveCodeBench (from 21.1), surpassing Claude-3-Haiku.

ManiBox: Enhancing Spatial Grasping Generalization via Scalable Simulation Data Generation

Learning a precise robotic grasping policy is crucial for embodied agents operating in complex real-world manipulation tasks. Despite significant advancements, most models still struggle with accurate spatial positioning of objects to be grasped. We first show that this spatial generalization challenge stems primarily from the extensive data requirements for adequate spatial understanding. However, collecting such data with real robots is prohibitively expensive, and relying on simulation data often leads to visual generalization gaps upon deployment. To overcome these challenges, we then focus on state-based policy generalization and present \textbf{ManiBox}, a novel bounding-box-guided manipulation method built on a simulation-based teacher-student framework. The teacher policy efficiently generates scalable simulation data using bounding boxes, which are proven to uniquely determine the objects' spatial positions. The student policy then utilizes these low-dimensional spatial states to enable zero-shot transfer to real robots. Through comprehensive evaluations in simulated and real-world environments, ManiBox demonstrates a marked improvement in spatial grasping generalization and adaptability to diverse objects and backgrounds. Further, our empirical study into scaling laws for policy performance indicates that spatial volume generalization scales positively with data volume. For a certain level of spatial volume, the success rate of grasping empirically follows Michaelis-Menten kinetics relative to data volume, showing a saturation effect as data increases. Our videos and code are available in https://thkkk.github.io/manibox.

Self-Boosting Large Language Models with Synthetic Preference Data

Through alignment with human preferences, Large Language Models (LLMs) have advanced significantly in generating honest, harmless, and helpful responses. However, collecting high-quality preference data is a resource-intensive and creativity-demanding process, especially for the continual improvement of LLMs. We introduce SynPO, a self-boosting paradigm that leverages synthetic preference data for model alignment. SynPO employs an iterative mechanism wherein a self-prompt generator creates diverse prompts, and a response improver refines model responses progressively. This approach trains LLMs to autonomously learn the generative rewards for their own outputs and eliminates the need for large-scale annotation of prompts and human preferences. After four SynPO iterations, Llama3-8B and Mistral-7B show significant enhancements in instruction-following abilities, achieving over 22.1% win rate improvements on AlpacaEval 2.0 and ArenaHard. Simultaneously, SynPO improves the general performance of LLMs on various tasks, validated by a 3.2 to 5.0 average score increase on the well-recognized Open LLM leaderboard.

HiDe-PET: Continual Learning via Hierarchical Decomposition of Parameter-Efficient Tuning

The deployment of pre-trained models (PTMs) has greatly advanced the field of continual learning (CL), enabling positive knowledge transfer and resilience to catastrophic forgetting. To sustain these advantages for sequentially arriving tasks, a promising direction involves keeping the pre-trained backbone frozen while employing parameter-efficient tuning (PET) techniques to instruct representation learning. Despite the popularity of Prompt-based PET for CL, its empirical design often leads to sub-optimal performance in our evaluation of different PTMs and target tasks. To this end, we propose a unified framework for CL with PTMs and PET that provides both theoretical and empirical advancements. We first perform an in-depth theoretical analysis of the CL objective in a pre-training context, decomposing it into hierarchical components namely within-task prediction, task-identity inference and task-adaptive prediction. We then present Hierarchical Decomposition PET (HiDe-PET), an innovative approach that explicitly optimizes the decomposed objective through incorporating task-specific and task-shared knowledge via mainstream PET techniques along with efficient recovery of pre-trained representations. Leveraging this framework, we delve into the distinct impacts of implementation strategy, PET technique and PET architecture, as well as adaptive knowledge accumulation amidst pronounced distribution changes. Finally, across various CL scenarios, our approach demonstrates remarkably superior performance over a broad spectrum of recent strong baselines.