Ce Ge

Daoyuan Chen, Yilun Huang, Zhijian Ma et al.

The immense evolution in Large Language Models (LLMs) has underscored the importance of massive, heterogeneous, and high-quality data. A data recipe is a mixture of data from different sources for training LLMs, which plays a vital role in LLMs' performance. Existing open-source tools for LLM data processing are mostly tailored for specific data recipes. To continuously uncover the potential of LLMs, incorporate data from new sources, and improve LLMs' performance, we build a new system named Data-Juicer, with which we can efficiently generate diverse data recipes, explore different possibilities in forming data mixtures, and evaluate their effects on model performance. Different from traditional data-analytics pipelines, Data-Juicer faces some unique challenges. Firstly, the possible data sources for forming data recipes are truly heterogeneous and massive with various qualities. Secondly, it is extremely expensive to precisely evaluate data recipes' impact on LLMs' performance. Thirdly, the end users of Data-Juicer, model developers, need sufficient flexibility to configure and evaluate different data recipes. Data-Juicer features a fine-grained abstraction of pipelines for constructing data recipes, with over 50 built-in operators for easy composition and extension. By incorporating visualization and auto-evaluation capabilities, Data-Juicer enables a timely feedback loop for both LLM pre-training and fine-tuning. Further, Data-Juicer is optimized and integrated with ecosystems for LLM training, evaluation, and distributed computing. The data recipes derived with Data-Juicer gain notable improvements on state-of-the-art LLMs, by up to 7.45% increase in averaged score across 16 LLM benchmarks and 17.5% higher win rate in pair-wise GPT-4 evaluations. Our system, data recipes, and tutorials are released, calling for broader data-centric research on training and understanding LLMs.

Daoyuan Chen, Haibin Wang, Yilun Huang et al.

The emergence of multimodal large models has advanced artificial intelligence, introducing unprecedented levels of performance and functionality. However, optimizing these models remains challenging due to historically isolated paths of model-centric and data-centric developments, leading to suboptimal outcomes and inefficient resource utilization. In response, we present a new sandbox suite tailored for integrated data-model co-development. This sandbox provides a feedback-driven experimental platform, enabling cost-effective iteration and guided refinement of both data and models. Our proposed ``Probe-Analyze-Refine'' workflow, validated through practical use cases on multimodal tasks such as image-text pre-training with CLIP, image-to-text generation with LLaVA-like models, and text-to-video generation with DiT-based models, yields transferable and notable performance boosts, such as topping the VBench leaderboard. A comprehensive set of over 100 experiments demonstrated the suite's usability and extensibility, while also uncovering insights into the interplay between data quality, diversity, model behavior, and computational costs. All codes, datasets, and models are open-sourced to foster future research and applications that would otherwise be infeasible due to the lack of a dedicated co-development infrastructure.

Ce Ge, Zhijian Ma, Daoyuan Chen et al.

Large language models have demonstrated remarkable capabilities across various tasks, primarily attributed to the utilization of diversely sourced data. However, the impact of pretraining data composition on model performance remains poorly understood. This paper introduces $\textbf{BiMix}$, a novel bivariate data mixing law that models the joint scaling behavior of domain proportions and data volume in LLM pretraining. $\textbf{BiMix}$ provides a systematic framework for understanding and optimizing data mixtures across diverse domains. Through extensive experiments on two large-scale datasets, we demonstrate $\textbf{BiMix}$'s high accuracy in loss extrapolation (mean relative error < 0.2%) and its generalization to unseen mixtures (R${}^{2}$ > 0.97). Optimization of domain proportions yields superior model performance compared to existing methods. Furthermore, we establish entropy-based measures as efficient proxies for data mixing, offering a computationally lightweight strategy. Our work contributes both theoretical insights into data mixing dynamics and practical tools for enhancing LLM training efficiency, paving the way for more effective scaling strategies in language model development.

Daoyuan Chen, Yilun Huang, Xuchen Pan et al.

Foundation models demand advanced data processing for their vast, multimodal datasets. However, traditional frameworks struggle with the unique complexities of multimodal data. In response, we present Data-Juicer 2.0, a data processing system backed by 100+ data processing operators spanning text, image, video, and audio modalities, supporting more critical tasks including data analysis, synthesis, annotation, and foundation model post-training. With seamless compatibility and dedicated optimization for popular dataset hubs like Hugging Face and computing engines like Ray, it improves upon its predecessor in terms of usability, efficiency, and programmability. It features an easily accessible user interface layer that supports decoupled Python interactions, RESTful APIs, and conversational commands. Its new runtime layer offers adaptive execution across diverse scales and environments, abstracting away system complexities. Extensive empirical evaluations demonstrate Data-Juicer 2.0's remarkable performance and scalability, highlighting its capability to efficiently process TB-level data with 10k+ CPU cores. The system is publicly available and has been widely adopted in diverse research fields and real-world products such as Alibaba Cloud PAI. We actively maintain the system and share practical insights to foster research and applications of next-generation foundation models.

Jiashi Li, Qi Qi, Jingyu Wang et al.

Channel pruning can significantly accelerate and compress deep neural networks. Many channel pruning works utilize structured sparsity regularization to zero out all the weights in some channels and automatically obtain structure-sparse network in training stage. However, these methods apply structured sparsity regularization on each layer separately where the correlations between consecutive layers are omitted. In this paper, we first combine one out-channel in current layer and the corresponding in-channel in next layer as a regularization group, namely out-in-channel. Our proposed Out-In-Channel Sparsity Regularization (OICSR) considers correlations between successive layers to further retain predictive power of the compact network. Training with OICSR thoroughly transfers discriminative features into a fraction of out-in-channels. Correspondingly, OICSR measures channel importance based on statistics computed from two consecutive layers, not individual layer. Finally, a global greedy pruning algorithm is designed to remove redundant out-in-channels in an iterative way. Our method is comprehensively evaluated with various CNN architectures including CifarNet, AlexNet, ResNet, DenseNet and PreActSeNet on CIFAR-10, CIFAR-100 and ImageNet-1K datasets. Notably, on ImageNet-1K, we reduce 37.2% FLOPs on ResNet-50 while outperforming the original model by 0.22% top-1 accuracy.