Ruizhe Wang

Houwen Peng, Kan Wu, Yixuan Wei et al.

In this paper, we explore FP8 low-bit data formats for efficient training of large language models (LLMs). Our key insight is that most variables, such as gradients and optimizer states, in LLM training can employ low-precision data formats without compromising model accuracy and requiring no changes to hyper-parameters. Specifically, we propose a new FP8 automatic mixed-precision framework for training LLMs. This framework offers three levels of FP8 utilization to streamline mixed-precision and distributed parallel training for LLMs. It gradually incorporates 8-bit gradients, optimizer states, and distributed learning in an incremental manner. Experiment results show that, during the training of GPT-175B model on H100 GPU platform, our FP8 mixed-precision training framework not only achieved a remarkable 39% reduction in real memory usage but also ran 75% faster than the widely adopted BF16 framework (i.e., Megatron-LM), surpassing the speed of Nvidia Transformer Engine by 37%. This largely reduces the training costs for large foundation models. Furthermore, our FP8 mixed-precision training methodology is generic. It can be seamlessly applied to other tasks such as LLM instruction tuning and reinforcement learning with human feedback, offering savings in fine-tuning expenses. Our FP8 low-precision training framework is open-sourced at {https://github.com/Azure/MS-AMP}{aka.ms/MS.AMP}.

Lei Qu, Lianhai Ren, Peng Cheng et al.

An increasing variety of AI accelerators is being considered for large-scale training. However, enabling large-scale training on early-life AI accelerators faces three core challenges: frequent system disruptions and undefined failure modes that undermine reliability; numerical errors and training instabilities that threaten correctness and convergence; and the complexity of parallelism optimization combined with unpredictable local noise that degrades efficiency. To address these challenges, SIGMA is an open-source training stack designed to improve the reliability, stability, and efficiency of large-scale distributed training on early-life AI hardware. The core of this initiative is the LUCIA TRAINING PLATFORM (LTP), the system optimized for clusters with early-life AI accelerators. Since its launch in March 2025, LTP has significantly enhanced training reliability and operational productivity. Over the past five months, it has achieved an impressive 94.45% effective cluster accelerator utilization, while also substantially reducing node recycling and job-recovery times. Building on the foundation of LTP, the LUCIA TRAINING FRAMEWORK (LTF) successfully trained SIGMA-MOE, a 200B MoE model, using 2,048 AI accelerators. This effort delivered remarkable stability and efficiency outcomes, achieving 21.08% MFU, state-of-the-art downstream accuracy, and encountering only one stability incident over a 75-day period. Together, these advances establish SIGMA, which not only tackles the critical challenges of large-scale training but also establishes a new benchmark for AI infrastructure and platform innovation, offering a robust, cost-effective alternative to prevailing established accelerator stacks and significantly advancing AI capabilities and scalability. The source code of SIGMA is available at https://github.com/microsoft/LuciaTrainingPlatform.

Qingguo Hu, Zhenghao Lin, Ziyue Yang et al.

Mixture-of-Experts (MoE) has emerged as a promising paradigm for foundation models due to its efficient and powerful scalability. In this work, we present Sigma-MoE-Tiny, an MoE language model that achieves the highest sparsity compared to existing open-source models. Sigma-MoE-Tiny employs fine-grained expert segmentation with up to 96 experts per layer, while activating only one expert for each token, resulting in 20B total parameters with just 0.5B activated. The major challenge introduced by such extreme sparsity lies in expert load balancing. We find that the widely-used load balancing loss tends to become ineffective in the lower layers under this setting. To address this issue, we propose a progressive sparsification schedule aiming to balance expert utilization and training stability. Sigma-MoE-Tiny is pre-trained on a diverse and high-quality corpus, followed by post-training to further unlock its capabilities. The entire training process remains remarkably stable, with no occurrence of irrecoverable loss spikes. Comprehensive evaluations reveal that, despite activating only 0.5B parameters, Sigma-MoE-Tiny achieves top-tier performance among counterparts of comparable or significantly larger scale. In addition, we provide an in-depth discussion of load balancing in highly sparse MoE models, offering insights for advancing sparsity in future MoE architectures. Project page: https://qghuxmu.github.io/Sigma-MoE-Tiny Code: https://github.com/microsoft/ltp-megatron-lm

Xinxin Zhuo, Mengyuan Niu, Ruizhe Wang et al.

Micro-scale street-level economic assessment is fundamental for precision spatial resource allocation. While Street View Imagery (SVI) advances urban sensing, existing approaches remain semantically superficial and overlook brand hierarchy heterogeneity and structural recession. To address this, we propose a visual-semantic and field-based spatiotemporal framework, operationalized via the Street Economic Vitality Index (SEVI). Our approach integrates physical and semantic streetscape parsing through instance segmentation of signboards, glass interfaces, and storefront closures. A dual-stage VLM-LLM pipeline standardizes signage into global hierarchies to quantify a spatially smoothed brand premium index. To overcome static SVI limitations, we introduce a temporal lag design using Location-Based Services (LBS) data to capture realized demand. Combined with a category-weighted Gaussian spillover model, we construct a three-dimensional diagnostic system covering Commercial Activity, Spatial Utilization, and Physical Environment. Experiments based on time-lagged geographically weighted regression across eight tidal periods in Nanjing reveal quasi-causal spatiotemporal heterogeneity. Street vibrancy arises from interactions between hierarchical brand clustering and mall-induced externalities. High-quality interfaces show peak attraction during midday and evening, while structural recession produces a lagged nighttime repulsion effect. The framework offers evidence-based support for precision spatial governance.

Ruizhe Wang, Chunliang Hua, Tomakayev Shingys et al.

The photorealistic reconstruction and rendering of architectural scenes have extensive applications in industries such as film, games, and transportation. It also plays an important role in urban planning, architectural design, and the city's promotion, especially in protecting historical and cultural relics. The 3D Gaussian Splatting, due to better performance over NeRF, has become a mainstream technology in 3D reconstruction. Its only input is a set of images but it relies heavily on geometric parameters computed by the SfM process. At the same time, there is an existing abundance of raw 3D models, that could inform the structural perception of certain buildings but cannot be applied. In this paper, we propose a straightforward method to harness these raw 3D models to guide 3D Gaussians in capturing the basic shape of the building and improve the visual quality of textures and details when photos are captured non-systematically. This exploration opens up new possibilities for improving the effectiveness of 3D reconstruction techniques in the field of architectural design.

Ruizhe Wang, Yeyun Gong, Xiao Liu et al.

The growing computational demands of training large language models (LLMs) necessitate more efficient methods. Quantized training presents a promising solution by enabling low-bit arithmetic operations to reduce these costs. While FP8 precision has demonstrated feasibility, leveraging FP4 remains a challenge due to significant quantization errors and limited representational capacity. This work introduces the first FP4 training framework for LLMs, addressing these challenges with two key innovations: a differentiable quantization estimator for precise weight updates and an outlier clamping and compensation strategy to prevent activation collapse. To ensure stability, the framework integrates a mixed-precision training scheme and vector-wise quantization. Experimental results demonstrate that our FP4 framework achieves accuracy comparable to BF16 and FP8, with minimal degradation, scaling effectively to 13B-parameter LLMs trained on up to 100B tokens. With the emergence of next-generation hardware supporting FP4, our framework sets a foundation for efficient ultra-low precision training.

Ruizhe Wang, Yucheng Ding, Xiao Liu et al.

The rapidly increasing computational cost of pretraining Large Language Models necessitates more efficient approaches. Numerous computational costs have been invested in existing well-trained checkpoints, but many of them remain underutilized due to engineering constraints or limited model capacity. To efficiently reuse this "sunk" cost, we propose to recycle pretrained checkpoints by expanding their parameter counts and continuing training. We propose orthogonal growth method well-suited for converged Mixture-of-Experts model: interpositional layer copying for depth growth and expert duplication with injected noise for width growth. To determine the optimal timing for such growth across checkpoints sequences, we perform comprehensive scaling experiments revealing that the final accuracy has a strong positive correlation with the amount of sunk cost, indicating that greater prior investment leads to better performance. We scale our approach to models with 70B parameters and over 1T training tokens, achieving 10.66% accuracy gain over training from scratch under the same additional compute budget. Our checkpoint recycling approach establishes a foundation for economically efficient large language model pretraining.

Yuzhe Ma, Jon Sharp, Ruizhe Wang et al.

Kalman Filter (KF) is widely used in various domains to perform sequential learning or variable estimation. In the context of autonomous vehicles, KF constitutes the core component of many Advanced Driver Assistance Systems (ADAS), such as Forward Collision Warning (FCW). It tracks the states (distance, velocity etc.) of relevant traffic objects based on sensor measurements. The tracking output of KF is often fed into downstream logic to produce alerts, which will then be used by human drivers to make driving decisions in near-collision scenarios. In this paper, we study adversarial attacks on KF as part of the more complex machine-human hybrid system of Forward Collision Warning. Our attack goal is to negatively affect human braking decisions by causing KF to output incorrect state estimations that lead to false or delayed alerts. We accomplish this by sequentially manipulating measure ments fed into the KF, and propose a novel Model Predictive Control (MPC) approach to compute the optimal manipulation. Via experiments conducted in a simulated driving environment, we show that the attacker is able to successfully change FCW alert signals through planned manipulation over measurements prior to the desired target time. These results demonstrate that our attack can stealthily mislead a distracted human driver and cause vehicle collisions.

Yunang Chen, Amrita Roy Chowdhury, Ruizhe Wang et al.

Trigger-action platforms (TAPs) allow users to connect independent web-based or IoT services to achieve useful automation. They provide a simple interface that helps end-users create trigger-compute-action rules that pass data between disparate Internet services. Unfortunately, TAPs introduce a large-scale security risk: if they are compromised, attackers will gain access to sensitive data for millions of users. To avoid this risk, we propose eTAP, a privacy-enhancing trigger-action platform that executes trigger-compute-action rules without accessing users' private data in plaintext or learning anything about the results of the computation. We use garbled circuits as a primitive, and leverage the unique structure of trigger-compute-action rules to make them practical. We formally state and prove the security guarantees of our protocols. We prototyped eTAP, which supports the most commonly used operations on popular commercial TAPs like IFTTT and Zapier. Specifically, it supports Boolean, arithmetic, and string operations on private trigger data and can run 100% of the top-500 rules of IFTTT users and 93.4% of all publicly-available rules on Zapier. Based on ten existing rules that exercise a wide variety of operations, we show that eTAP has a modest performance impact: on average rule execution latency increases by 70 ms (55%) and throughput reduces by 59%.

Xudong Liu, Ruizhe Wang, Chih-Fan Chen et al.

Facial appearance plays an important role in our social lives. Subjective perception of women's beauty depends on various face-related (e.g., skin, shape, hair) and environmental (e.g., makeup, lighting, angle) factors. Similar to cosmetic surgery in the physical world, virtual face beautification is an emerging field with many open issues to be addressed. Inspired by the latest advances in style-based synthesis and face beauty prediction, we propose a novel framework of face beautification. For a given reference face with a high beauty score, our GAN-based architecture is capable of translating an inquiry face into a sequence of beautified face images with referenced beauty style and targeted beauty score values. To achieve this objective, we propose to integrate both style-based beauty representation (extracted from the reference face) and beauty score prediction (trained on SCUT-FBP database) into the process of beautification. Unlike makeup transfer, our approach targets at many-to-many (instead of one-to-one) translation where multiple outputs can be defined by either different references or varying beauty scores. Extensive experimental results are reported to demonstrate the effectiveness and flexibility of the proposed face beautification framework.