arXiv subjects

Kangrui Du

Publications and source records attributed to Kangrui Du.

4 recordsLinked to original sources

arXiv2026

Miles v0.1: Production-Level Post-Training

We present Miles v0.1, a full-stack, production-ready system for frontier post-training. Building upon the clean design of slime, Miles designs each stage of the reinforcement-learning (RL) training loop around a single principle: components should be verified, clean, and customizable. With accuracy, efficiency, reliability, and scalability as first-class goals, Miles aims to make frontier-scale RL accessible to researchers and enterprises alike. This report walks through the system end to end: rollout engines built on SGLang, a trainer with a choice of two backends (NVIDIA Megatron-LM and PyTorch FSDP), and three weight-synchronization transports for different deployment topologies. Beyond full-parameter RL, Miles also supports LoRA RL, on-policy distillation, supervised fine-tuning, and true-on-policy rollout-training alignment, and extends the same architecture to diffusion models. We close with an end-to-end case study: fully asynchronous agentic RL on a GLM-5.2 744B-A40B model over terminal-use coding tasks, running on 64 NVIDIA GB300 GPUs with a median step time of 263 seconds over the first 30 measured steps. Miles is open-sourced at https://github.com/radixark/miles, with the project website at https://miles.radixark.com.

cs.LG↗

arXiv2025

APRIL: Active Partial Rollouts in Reinforcement Learning to Tame Long-tail Generation

Reinforcement learning (RL) has become a cornerstone in advancing large-scale pre-trained language models (LLMs). Successive generations, including GPT-o series, DeepSeek-R1, Kimi-K1.5, Grok 4, and GLM-4.5, have relied on large-scale RL training to enhance reasoning and coding capabilities. To meet the community's growing RL needs, numerous RL frameworks have been proposed. However, RL training remains computationally expensive, with rollout generation accounting for more than 90% of total runtime. In addition, its efficiency is often constrained by the long-tail distribution of rollout response lengths, where a few lengthy responses stall entire batches, leaving GPUs idle and underutilized. As model and rollout sizes continue to grow, this bottleneck increasingly limits scalability. To address this challenge, we propose Active Partial Rollouts in Reinforcement Learning (APRIL), which mitigates long-tail inefficiency. In the rollout phase, APRIL over-provisions rollout requests, terminates once the target number of responses is reached, and recycles incomplete responses for continuation in future steps. This strategy ensures that no rollouts are discarded while substantially reducing GPU idle time. Experiments show that APRIL improves rollout throughput by 22.5% on average (at most 44%) across commonly used RL algorithms (GRPO, DAPO, GSPO), accelerates convergence, and achieves 2.1% on average(at most 8%) higher final accuracy across tasks. Moreover, APRIL is both framework and hardware agnostic, already integrated into the slime RL framework, and deployable on NVIDIA and AMD GPUs alike. Taken together, this work unifies system-level and algorithmic considerations in proposing APRIL, with the aim of advancing RL training efficiency and inspiring further optimizations in RL systems. Our codebase is available at https://github.com/RLsys-Foundation/APRIL

cs.LG↗

arXiv2025

Temporal Flexibility in Spiking Neural Networks: Towards Generalization Across Time Steps and Deployment Friendliness

Spiking Neural Networks (SNNs), models inspired by neural mechanisms in the brain, allow for energy-efficient implementation on neuromorphic hardware. However, SNNs trained with current direct training approaches are constrained to a specific time step. This "temporal inflexibility" 1) hinders SNNs' deployment on time-step-free fully event-driven chips and 2) prevents energy-performance balance based on dynamic inference time steps. In this study, we first explore the feasibility of training SNNs that generalize across different time steps. We then introduce Mixed Time-step Training (MTT), a novel method that improves the temporal flexibility of SNNs, making SNNs adaptive to diverse temporal structures. During each iteration of MTT, random time steps are assigned to different SNN stages, with spikes transmitted between stages via communication modules. After training, the weights are deployed and evaluated on both time-stepped and fully event-driven platforms. Experimental results show that models trained by MTT gain remarkable temporal flexibility, friendliness for both event-driven and clock-driven deployment (nearly lossless on N-MNIST and 10.1% higher than standard methods on CIFAR10-DVS), enhanced network generalization, and near SOTA performance. To the best of our knowledge, this is the first work to report the results of large-scale SNN deployment on fully event-driven scenarios.

cs.LG↗

arXiv2024

CMViM: Contrastive Masked Vim Autoencoder for 3D Multi-modal Representation Learning for AD classification

Alzheimer's disease (AD) is an incurable neurodegenerative condition leading to cognitive and functional deterioration. Given the lack of a cure, prompt and precise AD diagnosis is vital, a complex process dependent on multiple factors and multi-modal data. While successful efforts have been made to integrate multi-modal representation learning into medical datasets, scant attention has been given to 3D medical images. In this paper, we propose Contrastive Masked Vim Autoencoder (CMViM), the first efficient representation learning method tailored for 3D multi-modal data. Our proposed framework is built on a masked Vim autoencoder to learn a unified multi-modal representation and long-dependencies contained in 3D medical images. We also introduce an intra-modal contrastive learning module to enhance the capability of the multi-modal Vim encoder for modeling the discriminative features in the same modality, and an inter-modal contrastive learning module to alleviate misaligned representation among modalities. Our framework consists of two main steps: 1) incorporate the Vision Mamba (Vim) into the mask autoencoder to reconstruct 3D masked multi-modal data efficiently. 2) align the multi-modal representations with contrastive learning mechanisms from both intra-modal and inter-modal aspects. Our framework is pre-trained and validated ADNI2 dataset and validated on the downstream task for AD classification. The proposed CMViM yields 2.7\% AUC performance improvement compared with other state-of-the-art methods.

cs.CV↗