arXiv subjects

Wenhui Wang

Publications and source records attributed to Wenhui Wang.

At least 19 recordsLinked to original sources

arXiv2026

VibeVoice-ASR-Streaming Technical Report

Traditional speaker-attributed ASR systems treated ASR and speaker diarization as two separate tasks. Recently, end-to-end models such as VibeVoice-ASR have unified the two tasks within a single model. However, existing unified models still mainly support offline recognition, making it difficult to meet the low-latency requirements of real-time voice assistants and agents. To tackle this issue, we present VibeVoice-ASR-Streaming, one of the first LLM-based end-to-end approaches to streaming speaker-attributed ASR. It interleaves fixed-size audio chunks, a small amount of lookahead audio and previous text. This allows the model to produce ''who said what'' as speech arrives, without a separate diarization stage. For transcription accuracy, our 7B model achieves the lowest average WER/CER across five evaluation sets. For speaker attribution, it achieves the best or tied-best on 12 of 13 evaluation settings. We release the 1.5B and 7B model weights together with inference code.

eess.AS↗

arXiv2026

VibeVoice-ASR-BitNet Technical Report

We present VibeVoice-ASR-BitNet, a compressed variant of VibeVoice-ASR optimized for real-time inference on edge CPUs. We apply heterogeneous quantization tailored to the computational characteristics of each stage: the VAE acoustic tokenizer uses full-pipeline INT8 quantization (I8_S) with kernel fusion and SIMD optimization, while the autoregressive language model adopts BitNet-style ternary weights (I2_S). To preserve accuracy under aggressive compression, we employ a progressive quantization-aware training strategy. For inference, we implement custom SIMD kernels and fused operators within the ggml framework targeting both ARM and x86 platforms, achieving real-time recognition (RTF < 1) on low-thread-count CPUs. VibeVoice-ASR-BitNet is 1.6--2.3x faster than Whisper.cpp at comparable model sizes (~1.6 GB), with only modest accuracy degradation compared to the FP16 baseline.

cs.SD↗

arXiv2026

Low-Latency Event-Based Object Detection with Spatially-Sparse Linear Attention

Event cameras provide sequential visual data with spatial sparsity and high temporal resolution, making them attractive for low-latency object detection. Existing asynchronous event-based neural networks exploit this low-latency advantage by updating predictions event by event, but still suffer from two bottlenecks: recurrent architectures are difficult to train efficiently on long sequences, and improving accuracy often increases per-event computation and latency. Linear attention is appealing because it enables parallel training and recurrent inference. However, its dense state updates make per-event computation scale with the state size, yielding a poor accuracy-efficiency trade-off for object detection, where accurate localization requires fine-grained spatial states. The key challenge is therefore to introduce sparse state activation that exploits the spatial sparsity of events while preserving efficient parallel training. We propose Spatially-Sparse Linear Attention (SSLA), which introduces a mixture-of-spaces state decomposition and a scatter-compute-gather training procedure, enabling state-level sparsity as well as training parallelism. Building on SSLA, we develop an end-to-end asynchronous linear attention model, SSLA-Det, for low-latency event-based object detection. On Gen1 and N-Caltech101, SSLA-Det achieves state-of-the-art accuracy among asynchronous methods, reaching 0.375 mAP and 0.515 mAP, respectively, while reducing per-event computation by over 20 times compared with the strongest prior asynchronous baseline, demonstrating the potential of linear attention for low-latency event-based vision.

cs.CV↗

arXiv2026

VIBEVOICE-ASR Technical Report

This report presents VibeVoice-ASR, a general-purpose speech understanding framework built upon VibeVoice, designed to address the persistent challenges of context fragmentation and multi-speaker complexity in long-form audio (e.g., meetings, podcasts) that remain despite recent advancements in short-form speech recognition. Unlike traditional pipelined approaches that rely on audio chunking, VibeVoice-ASRsupports single-pass processing for up to 60 minutes of audio. It unifies Automatic Speech Recognition, Speaker Diarization, and Timestamping into a single end-to-end generation task. In addition, VibeVoice-ASR supports over 50 languages, requires no explicit language setting, and natively handles code-switching within and across utterances. Furthermore, we introduce a prompt-based context injection mechanism that allows users to supply customized conetxt, significantly improving accuracy on domain-specific terminology and polyphonic character disambiguation.

cs.SD↗

arXiv2025

BitNet Distillation

In this paper, we present BitNet Distillation (BitDistill), a lightweight pipeline that fine-tunes off-the-shelf full-precision LLMs (e.g., Qwen) into 1.58-bit precision (i.e., ternary weights {-1, 0, 1}) for specific downstream tasks, achieving strong task-specific performance with minimal computational cost. Specifically, BitDistill incorporates three key techniques: the SubLN module, as introduced in BitNet; multi-head attention distillation, based on MiniLM; and continual pre-training, which serves as a crucial warm-up step to mitigate the scalability issue of the performance gap between finetuned full-precision and 1.58-bit LLMs on specific tasks. Experimental results show that BitDistill achieves performance comparable to the full-precision counterpart models across model size, while enabling up to 10x memory savings and 2.65x faster inference on CPUs. Code is available at https://github.com/microsoft/BitNet.

cs.LG↗

arXiv2025

VLA-Adapter: An Effective Paradigm for Tiny-Scale Vision-Language-Action Model

Vision-Language-Action (VLA) models typically bridge the gap between perceptual and action spaces by pre-training a large-scale Vision-Language Model (VLM) on robotic data. While this approach greatly enhances performance, it also incurs significant training costs. In this paper, we investigate how to effectively bridge vision-language (VL) representations to action (A). We introduce VLA-Adapter, a novel paradigm designed to reduce the reliance of VLA models on large-scale VLMs and extensive pre-training. To this end, we first systematically analyze the effectiveness of various VL conditions and present key findings on which conditions are essential for bridging perception and action spaces. Based on these insights, we propose a lightweight Policy module with Bridge Attention, which autonomously injects the optimal condition into the action space. In this way, our method achieves high performance using only a 0.5B-parameter backbone, without any robotic data pre-training. Extensive experiments on both simulated and real-world robotic benchmarks demonstrate that VLA-Adapter not only achieves state-of-the-art level performance, but also offers the fast inference speed reported to date. Furthermore, thanks to the proposed advanced bridging paradigm, VLA-Adapter enables the training of a powerful VLA model in just 8 hours on a single consumer-grade GPU, greatly lowering the barrier to deploying the VLA model. Project page: https://vla-adapter.github.io/.

cs.RO↗

arXiv2025

VibeVoice Technical Report

This report presents VibeVoice, a novel model designed to synthesize long-form speech with multiple speakers by employing next-token diffusion, which is a unified method for modeling continuous data by autoregressively generating latent vectors via diffusion. To enable this, we introduce a novel continuous speech tokenizer that, when compared to the popular Encodec model, improves data compression by 80 times while maintaining comparable performance. The tokenizer effectively preserves audio fidelity while significantly boosting computational efficiency for processing long sequences. Thus, VibeVoice can synthesize long-form speech for up to 90 minutes (in a 64K context window length) with a maximum of 4 speakers, capturing the authentic conversational ``vibe'' and surpassing open-source and proprietary dialogue models.

cs.CL↗

arXiv2025

Label-Noise Resistant Learning via Optimal Brain Damage Masking

Noisy labels are inevitable in real-world multimedia applications. Due to the strong memorization capacity of deep neural networks, these noisy labels cause significant performance degradation. Existing noise-robust methods have mainly focused on robust loss functions and sample selection strategies, with comparatively limited exploration of dynamic architectural adaptation. In this paper, we rethink the role of classifier connectivity under label noise. Intuitively, performance degradation stems from the backpropagation of noisy gradients. Since the final classifier layer acts as the primary gateway for this error propagation, selectively discarding redundant connections can restrict the backpropagation pathways of noisy gradients. Consequently, to identify redundant connections, we leverage the seminal Optimal Brain Damage (OBD) theory from model compression, which posits that parameters causing negligible loss perturbation can be removed. Guided by this principle, we show that masking low-activation edges limits the estimated loss perturbation to preserve the model's fitting capacity, while tightening a derived upper bound on noise-induced gradient error. To bridge this theoretical insight with practical training, we propose a novel Selective Edge Masking (SEM) mechanism for the fully connected (FC) layer to enhance noise robustness. It can adaptively retain critical edges for information propagation while suppressing gradient errors caused by noisy labels. As a plug-and-play component, SEM can be seamlessly integrated into various noise-robust methods. Additionally, we validate SEM's applicability by applying it to the newly emerged Kolmogorov-Arnold Network (KAN) employed as a classifier. Extensive evaluations on synthetic and real-world benchmarks demonstrate that our approach achieves state-of-the-art performance.

cs.LG↗

arXiv2025

Maximizing Asynchronicity in Event-based Neural Networks

Event cameras deliver visual data with high temporal resolution, low latency, and minimal redundancy, yet their asynchronous, sparse sequential nature challenges standard tensor-based machine learning (ML). While the recent asynchronous-to-synchronous (A2S) paradigm aims to bridge this gap by asynchronously encoding events into learned features for ML pipelines, existing A2S approaches often sacrifice expressivity and generalizability compared to dense, synchronous methods. This paper introduces EVA (EVent Asynchronous feature learning), a novel A2S framework to generate highly expressive and generalizable event-by-event features. Inspired by the analogy between events and language, EVA uniquely adapts advances from language modeling in linear attention and self-supervised learning for its construction. In demonstration, EVA outperforms prior A2S methods on recognition tasks (DVS128-Gesture and N-Cars), and represents the first A2S framework to successfully master demanding detection tasks, achieving a 0.477 mAP on the Gen1 dataset. These results underscore EVA's potential for advancing real-time event-based vision applications.

cs.LG↗

arXiv2025

Atmospheric Newtonian Noise May Constrain Third-Generation Gravitational-Wave Detectors

Advanced gravitational-wave detector designs are pushing towards lower frequencies, where certain types of noise, previously considered negligible, may come to dominate the detectors' noise budgets. In particular, we revisit atmospheric Newtonian noise, caused by the fluctuating gravitational field as regions of high and low gas density move past the detector. We consider density perturbations both due to pressure waves (infrasound) and due to advected temperature fluctuations. In the absence of detailed site-specific models of topography, airflow, and building design, we present general scaling formulae that estimate the spectrum of atmospheric Newtonian noise, and show how it is affected by broad detector design choices. We confirm previous analyses that show that atmospheric Newtonian noise is not likely a strong contributor to advanced LIGO; however, it will be a dominant factor for third-generation detectors targeting low frequencies, with the low-frequency cutoff principally constrained by the depth of the detector underground.

gr-qc↗

arXiv2024

Multimodal Latent Language Modeling with Next-Token Diffusion

Multimodal generative models require a unified approach to handle both discrete data (e.g., text and code) and continuous data (e.g., image, audio, video). In this work, we propose Latent Language Modeling (LatentLM), which seamlessly integrates continuous and discrete data using causal Transformers. Specifically, we employ a variational autoencoder (VAE) to represent continuous data as latent vectors and introduce next-token diffusion for autoregressive generation of these vectors. Additionally, we develop $\sigma$-VAE to address the challenges of variance collapse, which is crucial for autoregressive modeling. Extensive experiments demonstrate the effectiveness of LatentLM across various modalities. In image generation, LatentLM surpasses Diffusion Transformers in both performance and scalability. When integrated into multimodal large language models, LatentLM provides a general-purpose interface that unifies multimodal generation and understanding. Experimental results show that LatentLM achieves favorable performance compared to Transfusion and vector quantized models in the setting of scaling up training tokens. In text-to-speech synthesis, LatentLM outperforms the state-of-the-art VALL-E 2 model in speaker similarity and robustness, while requiring 10x fewer decoding steps. The results establish LatentLM as a highly effective and scalable approach to advance large multimodal models.

cs.CL↗

arXiv2024

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}.

cs.CL↗

arXiv2024

GLoCIM: Global-view Long Chain Interest Modeling for news recommendation

Accurately recommending candidate news articles to users has always been the core challenge of news recommendation system. News recommendations often require modeling of user interest to match candidate news. Recent efforts have primarily focused on extracting local subgraph information in a global click graph constructed by the clicked news sequence of all users. Howerer, the computational complexity of extracting global click graph information has hindered the ability to utilize far-reaching linkage which is hidden between two distant nodes in global click graph collaboratively among similar users. To overcome the problem above, we propose a Global-view Long Chain Interests Modeling for news recommendation (GLoCIM), which combines neighbor interest with long chain interest distilled from a global click graph, leveraging the collaboration among similar users to enhance news recommendation. We therefore design a long chain selection algorithm and long chain interest encoder to obtain global-view long chain interest from the global click graph. We design a gated network to integrate long chain interest with neighbor interest to achieve the collaborative interest among similar users. Subsequently we aggregate it with local news category-enhanced representation to generate final user representation. Then candidate news representation can be formed to match user representation to achieve news recommendation. Experimental results on real-world datasets validate the effectiveness of our method to improve the performance of news recommendation.

cs.AI↗

arXiv2024

You Only Cache Once: Decoder-Decoder Architectures for Language Models

We introduce a decoder-decoder architecture, YOCO, for large language models, which only caches key-value pairs once. It consists of two components, i.e., a cross-decoder stacked upon a self-decoder. The self-decoder efficiently encodes global key-value (KV) caches that are reused by the cross-decoder via cross-attention. The overall model behaves like a decoder-only Transformer, although YOCO only caches once. The design substantially reduces GPU memory demands, yet retains global attention capability. Additionally, the computation flow enables prefilling to early exit without changing the final output, thereby significantly speeding up the prefill stage. Experimental results demonstrate that YOCO achieves favorable performance compared to Transformer in various settings of scaling up model size and number of training tokens. We also extend YOCO to 1M context length with near-perfect needle retrieval accuracy. The profiling results show that YOCO improves inference memory, prefill latency, and throughput by orders of magnitude across context lengths and model sizes. Code is available at https://aka.ms/YOCO.

cs.CL↗

arXiv2024

Multi-Head Mixture-of-Experts

Sparse Mixtures of Experts (SMoE) scales model capacity without significant increases in training and inference costs, but exhibits the following two issues: (1) Low expert activation, where only a small subset of experts are activated for optimization. (2) Lacking fine-grained analytical capabilities for multiple semantic concepts within individual tokens. We propose Multi-Head Mixture-of-Experts (MH-MoE), which employs a multi-head mechanism to split each token into multiple sub-tokens. These sub-tokens are then assigned to and processed by a diverse set of experts in parallel, and seamlessly reintegrated into the original token form. The multi-head mechanism enables the model to collectively attend to information from various representation spaces within different experts, while significantly enhances expert activation, thus deepens context understanding and alleviate overfitting. Moreover, our MH-MoE is straightforward to implement and decouples from other SMoE optimization methods, making it easy to integrate with other SMoE models for enhanced performance. Extensive experimental results across three tasks: English-focused language modeling, Multi-lingual language modeling and Masked multi-modality modeling tasks, demonstrate the effectiveness of MH-MoE.

cs.CL↗

arXiv2024

The Era of 1-bit LLMs: All Large Language Models are in 1.58 Bits

Recent research, such as BitNet, is paving the way for a new era of 1-bit Large Language Models (LLMs). In this work, we introduce a 1-bit LLM variant, namely BitNet b1.58, in which every single parameter (or weight) of the LLM is ternary {-1, 0, 1}. It matches the full-precision (i.e., FP16 or BF16) Transformer LLM with the same model size and training tokens in terms of both perplexity and end-task performance, while being significantly more cost-effective in terms of latency, memory, throughput, and energy consumption. More profoundly, the 1.58-bit LLM defines a new scaling law and recipe for training new generations of LLMs that are both high-performance and cost-effective. Furthermore, it enables a new computation paradigm and opens the door for designing specific hardware optimized for 1-bit LLMs.

cs.CL↗

arXiv2023

When an Image is Worth 1,024 x 1,024 Words: A Case Study in Computational Pathology

This technical report presents LongViT, a vision Transformer that can process gigapixel images in an end-to-end manner. Specifically, we split the gigapixel image into a sequence of millions of patches and project them linearly into embeddings. LongNet is then employed to model the extremely long sequence, generating representations that capture both short-range and long-range dependencies. The linear computation complexity of LongNet, along with its distributed algorithm, enables us to overcome the constraints of both computation and memory. We apply LongViT in the field of computational pathology, aiming for cancer diagnosis and prognosis within gigapixel whole-slide images. Experimental results demonstrate that LongViT effectively encodes gigapixel images and outperforms previous state-of-the-art methods on cancer subtyping and survival prediction. Code and models will be available at https://aka.ms/LongViT.

cs.CV↗

arXiv2023

Single-shot deterministic complex amplitude imaging with a single-layer metalens

Conventional imaging systems can only capture light intensity. Meanwhile, the lost phase information may be critical for a variety of applications such as label-free microscopy and optical metrology. Existing phase retrieval techniques typically require a bulky setup, multi-frame measurements, or prior information of the target scene. Here, we proposed an extremely compact system for complex amplitude imaging, leveraging the extreme versatility of a single-layer metalens to generate spatially-multiplexed and polarization-phase-shifted point spread functions. Combining the metalens with a polarization camera, the system can simultaneously record four polarization shearing interference patterns along both in-plane directions, thus allowing the deterministic reconstruction of the complex amplitude light field in a single shot. Using an incoherent light-emitting diode as the illumination, we experimentally demonstrated speckle-noise-free complex amplitude imaging for both static and moving objects with tailored magnification ratio and field-of-view. The miniaturized and robust system may open the door for complex amplitude imaging in portable devices for point-of-care applications.

physics.optics↗