Jakiro: Boosting Speculative Decoding with Decoupled Multi-Head via MoE

Speculative decoding (SD) accelerates large language model inference by using a smaller draft model to predict multiple tokens, which are then verified in parallel by the larger target model. However, the limited capacity of the draft model often necessitates tree-based sampling to improve prediction accuracy, where multiple candidates are generated at each step. We identify a key limitation in this approach: the candidates at the same step are derived from the same representation, limiting diversity and reducing overall effectiveness. To address this, we propose Jakiro, leveraging Mixture of Experts (MoE), where independent experts generate diverse predictions, effectively decoupling correlations among candidates. Furthermore, we introduce a hybrid inference strategy, combining autoregressive decoding for initial tokens with parallel decoding for subsequent stages, and enhance the latter with contrastive mechanism in features to improve accuracy. Our method significantly boosts prediction accuracy and achieves higher inference speedups. Extensive experiments across diverse models validate the effectiveness and robustness of our approach, establishing a new SOTA in speculative decoding. Our codes are available at https://github.com/haiduo/Jakiro.

Nearly Lossless Adaptive Bit Switching

Model quantization is widely applied for compressing and accelerating deep neural networks (DNNs). However, conventional Quantization-Aware Training (QAT) focuses on training DNNs with uniform bit-width. The bit-width settings vary across different hardware and transmission demands, which induces considerable training and storage costs. Hence, the scheme of one-shot joint training multiple precisions is proposed to address this issue. Previous works either store a larger FP32 model to switch between different precision models for higher accuracy or store a smaller INT8 model but compromise accuracy due to using shared quantization parameters. In this paper, we introduce the Double Rounding quantization method, which fully utilizes the quantized representation range to accomplish nearly lossless bit-switching while reducing storage by using the highest integer precision instead of full precision. Furthermore, we observe a competitive interference among different precisions during one-shot joint training, primarily due to inconsistent gradients of quantization scales during backward propagation. To tackle this problem, we propose an Adaptive Learning Rate Scaling (ALRS) technique that dynamically adapts learning rates for various precisions to optimize the training process. Additionally, we extend our Double Rounding to one-shot mixed precision training and develop a Hessian-Aware Stochastic Bit-switching (HASB) strategy. Experimental results on the ImageNet-1K classification demonstrate that our methods have enough advantages to state-of-the-art one-shot joint QAT in both multi-precision and mixed-precision. We also validate the feasibility of our method on detection and segmentation tasks, as well as on LLMs task. Our codes are available at https://github.com/haiduo/Double-Rounding.

Grasp What You Want: Embodied Dexterous Grasping System Driven by Your Voice

In recent years, as robotics has advanced, human-robot collaboration has gained increasing importance. However, current robots struggle to fully and accurately interpret human intentions from voice commands alone. Traditional gripper and suction systems often fail to interact naturally with humans, lack advanced manipulation capabilities, and are not adaptable to diverse tasks, especially in unstructured environments. This paper introduces the Embodied Dexterous Grasping System (EDGS), designed to tackle object grasping in cluttered environments for human-robot interaction. We propose a novel approach to semantic-object alignment using a Vision-Language Model (VLM) that fuses voice commands and visual information, significantly enhancing the alignment of multi-dimensional attributes of target objects in complex scenarios. Inspired by human hand-object interactions, we develop a robust, precise, and efficient grasping strategy, incorporating principles like the thumb-object axis, multi-finger wrapping, and fingertip interaction with an object's contact mechanics. We also design experiments to assess Referring Expression Representation Enrichment (RERE) in referring expression segmentation, demonstrating that our system accurately detects and matches referring expressions. Extensive experiments confirm that EDGS can effectively handle complex grasping tasks, achieving stability and high success rates, highlighting its potential for further development in the field of Embodied AI.

Learning to Refuse: Towards Mitigating Privacy Risks in LLMs

Large language models (LLMs) exhibit remarkable capabilities in understanding and generating natural language. However, these models can inadvertently memorize private information, posing significant privacy risks. This study addresses the challenge of enabling LLMs to protect specific individuals' private data without the need for complete retraining. We propose \return, a Real-world pErsonal daTa UnleaRNing dataset, comprising 2,492 individuals from Wikipedia with associated QA pairs, to evaluate machine unlearning (MU) methods for protecting personal data in a realistic scenario. Additionally, we introduce the Name-Aware Unlearning Framework (NAUF) for Privacy Protection, which enables the model to learn which individuals' information should be protected without affecting its ability to answer questions related to other unrelated individuals. Our extensive experiments demonstrate that NAUF achieves a state-of-the-art average unlearning score, surpassing the best baseline method by 5.65 points, effectively protecting target individuals' personal data while maintaining the model's general capabilities.

Probing Language Models for Pre-training Data Detection

Large Language Models (LLMs) have shown their impressive capabilities, while also raising concerns about the data contamination problems due to privacy issues and leakage of benchmark datasets in the pre-training phase. Therefore, it is vital to detect the contamination by checking whether an LLM has been pre-trained on the target texts. Recent studies focus on the generated texts and compute perplexities, which are superficial features and not reliable. In this study, we propose to utilize the probing technique for pre-training data detection by examining the model's internal activations. Our method is simple and effective and leads to more trustworthy pre-training data detection. Additionally, we propose ArxivMIA, a new challenging benchmark comprising arxiv abstracts from Computer Science and Mathematics categories. Our experiments demonstrate that our method outperforms all baselines, and achieves state-of-the-art performance on both WikiMIA and ArxivMIA, with additional experiments confirming its efficacy (Our code and dataset are available at https://github.com/zhliu0106/probing-lm-data).

Kangaroo: Lossless Self-Speculative Decoding via Double Early Exiting

Speculative decoding has demonstrated its effectiveness in accelerating the inference of large language models while maintaining a consistent sampling distribution. However, the conventional approach of training a separate draft model to achieve a satisfactory token acceptance rate can be costly. Drawing inspiration from early exiting, we propose a novel self-speculative decoding framework \emph{Kangaroo}, which uses a fixed shallow sub-network as a self-draft model, with the remaining layers serving as the larger target model. We train a lightweight and efficient adapter module on top of the sub-network to bridge the gap between the sub-network and the full model's representation ability. It is noteworthy that the inference latency of the self-draft model may no longer be negligible compared to the large model, necessitating strategies to increase the token acceptance rate while minimizing the drafting steps of the small model. To address this challenge, we introduce an additional early exiting mechanism for generating draft tokens. Specifically, we halt the small model's subsequent prediction during the drafting phase once the confidence level for the current token falls below a certain threshold. Extensive experiments on the Spec-Bench demonstrate the effectiveness of Kangaroo. Under single-sequence verification, Kangaroo achieves speedups up to $1.68\times$ on Spec-Bench, outperforming Medusa-1 with 88.7\% fewer additional parameters (67M compared to 591M). The code for Kangaroo is available at https://github.com/Equationliu/Kangaroo.

GhostNetV3: Exploring the Training Strategies for Compact Models

Compact neural networks are specially designed for applications on edge devices with faster inference speed yet modest performance. However, training strategies of compact models are borrowed from that of conventional models at present, which ignores their difference in model capacity and thus may impede the performance of compact models. In this paper, by systematically investigating the impact of different training ingredients, we introduce a strong training strategy for compact models. We find that the appropriate designs of re-parameterization and knowledge distillation are crucial for training high-performance compact models, while some commonly used data augmentations for training conventional models, such as Mixup and CutMix, lead to worse performance. Our experiments on ImageNet-1K dataset demonstrate that our specialized training strategy for compact models is applicable to various architectures, including GhostNetV2, MobileNetV2 and ShuffleNetV2. Specifically, equipped with our strategy, GhostNetV3 1.3$\times$ achieves a top-1 accuracy of 79.1% with only 269M FLOPs and a latency of 14.46ms on mobile devices, surpassing its ordinarily trained counterpart by a large margin. Moreover, our observation can also be extended to object detection scenarios. PyTorch code and checkpoints can be found at https://github.com/huawei-noah/Efficient-AI-Backbones/tree/master/ghostnetv3_pytorch.

DiffusionDialog: A Diffusion Model for Diverse Dialog Generation with Latent Space

In real-life conversations, the content is diverse, and there exists the one-to-many problem that requires diverse generation. Previous studies attempted to introduce discrete or Gaussian-based continuous latent variables to address the one-to-many problem, but the diversity is limited. Recently, diffusion models have made breakthroughs in computer vision, and some attempts have been made in natural language processing. In this paper, we propose DiffusionDialog, a novel approach to enhance the diversity of dialogue generation with the help of diffusion model. In our approach, we introduce continuous latent variables into the diffusion model. The problem of using latent variables in the dialog task is how to build both an effective prior of the latent space and an inferring process to obtain the proper latent given the context. By combining the encoder and latent-based diffusion model, we encode the response's latent representation in a continuous space as the prior, instead of fixed Gaussian distribution or simply discrete ones. We then infer the latent by denoising step by step with the diffusion model. The experimental results show that our model greatly enhances the diversity of dialog responses while maintaining coherence. Furthermore, in further analysis, we find that our diffusion model achieves high inference efficiency, which is the main challenge of applying diffusion models in natural language processing.

Controllable and Diverse Data Augmentation with Large Language Model for Low-Resource Open-Domain Dialogue Generation

Data augmentation (DA) is crucial to mitigate model training instability and over-fitting problems in low-resource open-domain dialogue generation. However, traditional DA methods often neglect semantic data diversity, restricting the overall quality. Recently, large language models (LLM) have been used for DA to generate diversified dialogues. However, they have limited controllability and tend to generate dialogues with a distribution shift compared to the seed dialogues. To maximize the augmentation diversity and address the controllability problem, we propose \textbf{S}ummary-based \textbf{D}ialogue \textbf{A}ugmentation with LLM (SDA). Our approach enhances the controllability of LLM by using dialogue summaries as a planning tool. Based on summaries, SDA can generate high-quality and diverse dialogue data even with a small seed dataset. To evaluate the efficacy of data augmentation methods for open-domain dialogue, we designed a clustering-based metric to characterize the semantic diversity of the augmented dialogue data. The experimental results show that SDA can augment high-quality and semantically diverse dialogues given a small seed dataset and an LLM, and the augmented data can boost the performance of open-domain dialogue models.

A Similarity Alignment Model for Video Copy Segment Matching

With the development of multimedia technology, Video Copy Detection has been a crucial problem for social media platforms. Meta AI hold Video Similarity Challenge on CVPR 2023 to push the technology forward. In this report, we share our winner solutions on Matching Track. We propose a Similarity Alignment Model(SAM) for video copy segment matching. Our SAM exhibits superior performance compared to other competitors, with a 0.108 / 0.144 absolute improvement over the second-place competitor in Phase 1 / Phase 2. Code is available at https://github.com/FeipengMa6/VSC22-Submission/tree/main/VSC22-Matching-Track-1st.