arXiv subjects

Junhao Huang

Publications and source records attributed to Junhao Huang.

24 records · Page 2Linked to original sources

arXiv2025

Step-3 is Large yet Affordable: Model-system Co-design for Cost-effective Decoding

Large language models (LLMs) face low hardware efficiency during decoding, especially for long-context reasoning tasks. This paper introduces Step-3, a 321B-parameter VLM with hardware-aware model-system co-design optimized for minimizing decoding costs. Step-3 innovates in two key dimensions: (1) A novel Multi-Matrix Factorization Attention (MFA) mechanism that significantly reduces both KV cache size and computation while maintaining high attention expressiveness, and (2) Attention-FFN Disaggregation (AFD), a distributed inference system that decouples attention and Feed-Forward Network (FFN) layers into specialized subsystems. This co-design achieves unprecedented cost efficiency: Step-3 significantly reduces theoretical decoding costs compared with models like DeepSeek-V3 and Qwen3 MoE 235B, with the gains widening at longer context. Step-3 achieves low cost while activating 38B parameters per token (more than DeepSeek-V3 and Qwen3 MoE 235B), demonstrating that hardware-aligned attention arithmetic intensity, MoE sparsity, and AFD are critical to cost-effectiveness. We perform a head-to-head comparison with DeepSeek-V3 in its favorable scenarios. Our implementation on Hopper GPUs achieves a decoding throughput of up to 4,039 tokens per second per GPU under 50ms TPOT SLA (4K context, FP8, no MTP). It is higher than DeepSeek-V3's 2,324 in the same setup and sets a new Pareto frontier for LLM decoding.

cs.LG↗

arXiv2025

SongComposer: A Large Language Model for Lyric and Melody Generation in Song Composition

Creating lyrics and melodies for the vocal track in a symbolic format, known as song composition, demands expert musical knowledge of melody, an advanced understanding of lyrics, and precise alignment between them. Despite achievements in sub-tasks such as lyric generation, lyric-to-melody, and melody-to-lyric, etc, a unified model for song composition has not yet been achieved. In this paper, we introduce SongComposer, a pioneering step towards a unified song composition model that can readily create symbolic lyrics and melodies following instructions. SongComposer is a music-specialized large language model (LLM) that, for the first time, integrates the capability of simultaneously composing lyrics and melodies into LLMs by leveraging three key innovations: 1) a flexible tuple format for word-level alignment of lyrics and melodies, 2) an extended tokenizer vocabulary for song notes, with scalar initialization based on musical knowledge to capture rhythm, and 3) a multi-stage pipeline that captures musical structure, starting with motif-level melody patterns and progressing to phrase-level structure for improved coherence. Extensive experiments demonstrate that SongComposer outperforms advanced LLMs, including GPT-4, in tasks such as lyric-to-melody generation, melody-to-lyric generation, song continuation, and text-to-song creation. Moreover, we will release SongCompose, a large-scale dataset for training, containing paired lyrics and melodies in Chinese and English.

cs.SD↗

arXiv2024

Texture-Preserving Diffusion Models for High-Fidelity Virtual Try-On

Image-based virtual try-on is an increasingly important task for online shopping. It aims to synthesize images of a specific person wearing a specified garment. Diffusion model-based approaches have recently become popular, as they are excellent at image synthesis tasks. However, these approaches usually employ additional image encoders and rely on the cross-attention mechanism for texture transfer from the garment to the person image, which affects the try-on's efficiency and fidelity. To address these issues, we propose an Texture-Preserving Diffusion (TPD) model for virtual try-on, which enhances the fidelity of the results and introduces no additional image encoders. Accordingly, we make contributions from two aspects. First, we propose to concatenate the masked person and reference garment images along the spatial dimension and utilize the resulting image as the input for the diffusion model's denoising UNet. This enables the original self-attention layers contained in the diffusion model to achieve efficient and accurate texture transfer. Second, we propose a novel diffusion-based method that predicts a precise inpainting mask based on the person and reference garment images, further enhancing the reliability of the try-on results. In addition, we integrate mask prediction and image synthesis into a single compact model. The experimental results show that our approach can be applied to various try-on tasks, e.g., garment-to-person and person-to-person try-ons, and significantly outperforms state-of-the-art methods on popular VITON, VITON-HD databases.

cs.CV↗

arXiv2024

Yet another Improvement of Plantard Arithmetic for Faster Kyber on Low-end 32-bit IoT Devices

This paper presents another improved version of Plantard arithmetic that could speed up Kyber implementations on two low-end 32-bit IoT platforms (ARM Cortex-M3 and RISC-V) without SIMD extensions. Specifically, we further enlarge the input range of the Plantard arithmetic without modifying its computation steps. After tailoring the Plantard arithmetic for Kyber's modulus, we show that the input range of the Plantard multiplication by a constant is at least 2.14 times larger than the original design in TCHES2022. Then, two optimization techniques for efficient Plantard arithmetic on Cortex-M3 and RISC-V are presented. We show that the Plantard arithmetic supersedes both Montgomery and Barrett arithmetic on low-end 32-bit platforms. With the enlarged input range and the efficient implementation of the Plantard arithmetic on these platforms, we propose various optimization strategies for NTT/INTT. We minimize or entirely eliminate the modular reduction of coefficients in NTT/INTT by taking advantage of the larger input range of the proposed Plantard arithmetic on low-end 32-bit platforms. Furthermore, we propose two memory optimization strategies that reduce 23.50% to 28.31% stack usage for the speed-version Kyber implementation when compared to its counterpart on Cortex-M4. The proposed optimizations make the speed-version implementation more feasible on low-end IoT devices. Thanks to the aforementioned optimizations, our NTT/INTT implementation shows considerable speedups compared to the state-of-the-art work. Overall, we demonstrate the applicability of the speed-version Kyber implementation on memory-constrained IoT platforms and set new speed records for Kyber on these platforms.

cs.CY↗

arXiv2023

Probability-based Global Cross-modal Upsampling for Pansharpening

Pansharpening is an essential preprocessing step for remote sensing image processing. Although deep learning (DL) approaches performed well on this task, current upsampling methods used in these approaches only utilize the local information of each pixel in the low-resolution multispectral (LRMS) image while neglecting to exploit its global information as well as the cross-modal information of the guiding panchromatic (PAN) image, which limits their performance improvement. To address this issue, this paper develops a novel probability-based global cross-modal upsampling (PGCU) method for pan-sharpening. Precisely, we first formulate the PGCU method from a probabilistic perspective and then design an efficient network module to implement it by fully utilizing the information mentioned above while simultaneously considering the channel specificity. The PGCU module consists of three blocks, i.e., information extraction (IE), distribution and expectation estimation (DEE), and fine adjustment (FA). Extensive experiments verify the superiority of the PGCU method compared with other popular upsampling methods. Additionally, experiments also show that the PGCU module can help improve the performance of existing SOTA deep learning pansharpening methods. The codes are available at https://github.com/Zeyu-Zhu/PGCU.

cs.CV↗

arXiv2022

abess: A Fast Best Subset Selection Library in Python and R

We introduce a new library named abess that implements a unified framework of best-subset selection for solving diverse machine learning problems, e.g., linear regression, classification, and principal component analysis. Particularly, the abess certifiably gets the optimal solution within polynomial times with high probability under the linear model. Our efficient implementation allows abess to attain the solution of best-subset selection problems as fast as or even 20x faster than existing competing variable (model) selection toolboxes. Furthermore, it supports common variants like best group subset selection and $\ell_2$ regularized best-subset selection. The core of the library is programmed in C++. For ease of use, a Python library is designed for conveniently integrating with scikit-learn, and it can be installed from the Python library Index. In addition, a user-friendly R library is available at the Comprehensive R Archive Network. The source code is available at: https://github.com/abess-team/abess.

stat.ML↗