Baichuan Alignment Technical Report

We introduce Baichuan Alignment, a detailed analysis of the alignment techniques employed in the Baichuan series of models. This represents the industry's first comprehensive account of alignment methodologies, offering valuable insights for advancing AI research. We investigate the critical components that enhance model performance during the alignment process, including optimization methods, data strategies, capability enhancements, and evaluation processes. The process spans three key stages: Prompt Augmentation System (PAS), Supervised Fine-Tuning (SFT), and Preference Alignment. The problems encountered, the solutions applied, and the improvements made are thoroughly recorded. Through comparisons across well-established benchmarks, we highlight the technological advancements enabled by Baichuan Alignment. Baichuan-Instruct is an internal model, while Qwen2-Nova-72B and Llama3-PBM-Nova-70B are instruct versions of the Qwen2-72B and Llama-3-70B base models, optimized through Baichuan Alignment. Baichuan-Instruct demonstrates significant improvements in core capabilities, with user experience gains ranging from 17% to 28%, and performs exceptionally well on specialized benchmarks. In open-source benchmark evaluations, both Qwen2-Nova-72B and Llama3-PBM-Nova-70B consistently outperform their respective official instruct versions across nearly all datasets. This report aims to clarify the key technologies behind the alignment process, fostering a deeper understanding within the community. Llama3-PBM-Nova-70B model is available at https://huggingface.co/PKU-Baichuan-MLSystemLab/Llama3-PBM-Nova-70B.

Baichuan-Omni Technical Report

The salient multimodal capabilities and interactive experience of GPT-4o highlight its critical role in practical applications, yet it lacks a high-performing open-source counterpart. In this paper, we introduce Baichuan-Omni, the first open-source 7B Multimodal Large Language Model (MLLM) adept at concurrently processing and analyzing modalities of image, video, audio, and text, while delivering an advanced multimodal interactive experience and strong performance. We propose an effective multimodal training schema starting with 7B model and proceeding through two stages of multimodal alignment and multitask fine-tuning across audio, image, video, and text modal. This approach equips the language model with the ability to handle visual and audio data effectively. Demonstrating strong performance across various omni-modal and multimodal benchmarks, we aim for this contribution to serve as a competitive baseline for the open-source community in advancing multimodal understanding and real-time interaction.

Language-Queried Target Sound Extraction Without Parallel Training Data

Language-queried target sound extraction (TSE) aims to extract specific sounds from mixtures based on language queries. Traditional fully-supervised training schemes require extensively annotated parallel audio-text data, which are labor-intensive. We introduce a language-free training scheme, requiring only unlabelled audio clips for TSE model training by utilizing the multi-modal representation alignment nature of the contrastive language-audio pre-trained model (CLAP). In a vanilla language-free training stage, target audio is encoded using the pre-trained CLAP audio encoder to form a condition embedding for the TSE model, while during inference, user language queries are encoded by CLAP text encoder. This straightforward approach faces challenges due to the modality gap between training and inference queries and information leakage from direct exposure to target audio during training. To address this, we propose a retrieval-augmented strategy. Specifically, we create an embedding cache using audio captions generated by a large language model (LLM). During training, target audio embeddings retrieve text embeddings from this cache to use as condition embeddings, ensuring consistent modalities between training and inference and eliminating information leakage. Extensive experiment results show that our retrieval-augmented approach achieves consistent and notable performance improvements over existing state-of-the-art with better generalizability.

Joint Semantic Knowledge Distillation and Masked Acoustic Modeling for Full-band Speech Restoration with Improved Intelligibility

Speech restoration aims at restoring full-band speech with high quality and intelligibility, considering a diverse set of distortions. MaskSR is a recently proposed generative model for this task. As other models of its kind, MaskSR attains high quality but, as we show, intelligibility can be substantially improved. We do so by boosting the speech encoder component of MaskSR with predictions of semantic representations of the target speech, using a pre-trained self-supervised teacher model. Then, a masked language model is conditioned on the learned semantic features to predict acoustic tokens that encode low level spectral details of the target speech. We show that, with the same MaskSR model capacity and inference time, the proposed model, MaskSR2, significantly reduces the word error rate, a typical metric for intelligibility. MaskSR2 also achieves competitive word error rate among other models, while providing superior quality. An ablation study shows the effectiveness of various semantic representations.

EA-VTR: Event-Aware Video-Text Retrieval

Understanding the content of events occurring in the video and their inherent temporal logic is crucial for video-text retrieval. However, web-crawled pre-training datasets often lack sufficient event information, and the widely adopted video-level cross-modal contrastive learning also struggles to capture detailed and complex video-text event alignment. To address these challenges, we make improvements from both data and model perspectives. In terms of pre-training data, we focus on supplementing the missing specific event content and event temporal transitions with the proposed event augmentation strategies. Based on the event-augmented data, we construct a novel Event-Aware Video-Text Retrieval model, ie, EA-VTR, which achieves powerful video-text retrieval ability through superior video event awareness. EA-VTR can efficiently encode frame-level and video-level visual representations simultaneously, enabling detailed event content and complex event temporal cross-modal alignment, ultimately enhancing the comprehensive understanding of video events. Our method not only significantly outperforms existing approaches on multiple datasets for Text-to-Video Retrieval and Video Action Recognition tasks, but also demonstrates superior event content perceive ability on Multi-event Video-Text Retrieval and Video Moment Retrieval tasks, as well as outstanding event temporal logic understanding ability on Test of Time task.

MaskSR: Masked Language Model for Full-band Speech Restoration

Speech restoration aims at restoring high quality speech in the presence of a diverse set of distortions. Although several deep learning paradigms have been studied for this task, the power of the recently emerging language models has not been fully explored. In this paper, we propose MaskSR, a masked language model capable of restoring full-band 44.1 kHz speech jointly considering noise, reverb, clipping, and low bandwidth. MaskSR works with discrete acoustic tokens extracted using a pre-trained neural codec. During training, MaskSR is optimized to predict randomly masked tokens extracted from the high quality target speech, conditioned on the corrupted speech with various distortions. During inference, MaskSR reconstructs the target speech tokens with efficient iterative sampling. Extensive experiments show that MaskSR obtains competitive results on both the full-band speech restoration task and also on sub-tasks compared with a wide range of models.

Dynamic Resolution Guidance for Facial Expression Recognition

Facial expression recognition (FER) is vital for human-computer interaction and emotion analysis, yet recognizing expressions in low-resolution images remains challenging. This paper introduces a practical method called Dynamic Resolution Guidance for Facial Expression Recognition (DRGFER) to effectively recognize facial expressions in images with varying resolutions without compromising FER model accuracy. Our framework comprises two main components: the Resolution Recognition Network (RRN) and the Multi-Resolution Adaptation Facial Expression Recognition Network (MRAFER). The RRN determines image resolution, outputs a binary vector, and the MRAFER assigns images to suitable facial expression recognition networks based on resolution. We evaluated DRGFER on widely-used datasets RAFDB and FERPlus, demonstrating that our method retains optimal model performance at each resolution and outperforms alternative resolution approaches. The proposed framework exhibits robustness against resolution variations and facial expressions, offering a promising solution for real-world applications.

CSST Strong Lensing Preparation: a Framework for Detecting Strong Lenses in the Multi-color Imaging Survey by the China Survey Space Telescope (CSST)

Strong gravitational lensing is a powerful tool for investigating dark matter and dark energy properties. With the advent of large-scale sky surveys, we can discover strong lensing systems on an unprecedented scale, which requires efficient tools to extract them from billions of astronomical objects. The existing mainstream lens-finding tools are based on machine learning algorithms and applied to cut-out-centered galaxies. However, according to the design and survey strategy of optical surveys by CSST, preparing cutouts with multiple bands requires considerable efforts. To overcome these challenges, we have developed a framework based on a hierarchical visual Transformer with a sliding window technique to search for strong lensing systems within entire images. Moreover, given that multi-color images of strong lensing systems can provide insights into their physical characteristics, our framework is specifically crafted to identify strong lensing systems in images with any number of channels. As evaluated using CSST mock data based on an Semi-Analytic Model named CosmoDC2, our framework achieves precision and recall rates of 0.98 and 0.90, respectively. To evaluate the effectiveness of our method in real observations, we have applied it to a subset of images from the DESI Legacy Imaging Surveys and media images from Euclid Early Release Observations. 61 new strong lensing system candidates are discovered by our method. However, we also identified false positives arising primarily from the simplified galaxy morphology assumptions within the simulation. This underscores the practical limitations of our approach while simultaneously highlighting potential avenues for future improvements.

CLAPSep: Leveraging Contrastive Pre-trained Models for Multi-Modal Query-Conditioned Target Sound Extraction

Universal sound separation (USS) aims to extract arbitrary types of sounds from real-world sound recordings. Language-queried target sound extraction (TSE) is an effective approach to achieving USS. Such systems consist of two components: a query network that converts user queries into conditional embeddings, and a separation network that extracts the target sound based on conditional embeddings. Existing methods mainly suffer from two issues: firstly, they require training a randomly initialized model from scratch, lacking the utilization of pre-trained models, and substantial data and computational resources are needed to ensure model convergence; secondly, existing methods need to jointly train a query network and a separation network, which tends to lead to overfitting. To address these issues, we build the CLAPSep model based on contrastive language-audio pre-trained model (CLAP). We achieve this by using a pre-trained text encoder of CLAP as the query network and introducing pre-trained audio encoder weights of CLAP into the separation network to fully utilize the prior knowledge embedded in the pre-trained model to assist in target sound extraction tasks. Extensive experimental results demonstrate that the proposed method saves training resources while ensuring the model's performance and generalizability. Additionally, we explore the model's ability to comprehensively utilize language/audio multi-modal and positive/negative multi-valent user queries, enhancing system performance while providing diversified application modes.

DFML: Decentralized Federated Mutual Learning

In the realm of real-world devices, centralized servers in Federated Learning (FL) present challenges including communication bottlenecks and susceptibility to a single point of failure. Additionally, contemporary devices inherently exhibit model and data heterogeneity. Existing work lacks a Decentralized FL (DFL) framework capable of accommodating such heterogeneity without imposing architectural restrictions or assuming the availability of public data. To address these issues, we propose a Decentralized Federated Mutual Learning (DFML) framework that is serverless, supports nonrestrictive heterogeneous models, and avoids reliance on public data. DFML effectively handles model and data heterogeneity through mutual learning, which distills knowledge between clients, and cyclically varying the amount of supervision and distillation signals. Extensive experimental results demonstrate consistent effectiveness of DFML in both convergence speed and global accuracy, outperforming prevalent baselines under various conditions. For example, with the CIFAR-100 dataset and 50 clients, DFML achieves a substantial increase of +17.20% and +19.95% in global accuracy under Independent and Identically Distributed (IID) and non-IID data shifts, respectively.