Cover Image

FP4 Explore, BF16 Train: Diffusion Reinforcement Learning via Efficient Rollout Scaling

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Li, Yitong, Chen, Junsong, Xue, Shuchen, Zeren, Pengcuo, Fu, Siyuan, Yang, Dinghao, Tang, Yangyang, Bai, Junjie, Luo, Ping, Han, Song, Xie, Enze
Format:	Preprint
Published:	2026
Subjects:	Machine Learning Artificial Intelligence Computer Vision and Pattern Recognition
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Balancing Speed and Stability: The Trade-offs of FP8 vs. BF16 Training in LLMs
by: Fujii, Kazuki, et al.
Published: (2024)

SANA-Sprint: One-Step Diffusion with Continuous-Time Consistency Distillation
by: Chen, Junsong, et al.
Published: (2025)

Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding
by: Wu, Chengyue, et al.
Published: (2025)

Jet-RL: Enabling On-Policy FP8 Reinforcement Learning with Unified Training and Rollout Precision Flow
by: Xi, Haocheng, et al.
Published: (2026)

Accelerating Diffusion Sampling with Optimized Time Steps
by: Xue, Shuchen, et al.
Published: (2024)

Fast-dLLM v2: Efficient Block-Diffusion LLM
by: Wu, Chengyue, et al.
Published: (2025)

PixArt-Σ: Weak-to-Strong Training of Diffusion Transformer for 4K Text-to-Image Generation
by: Chen, Junsong, et al.
Published: (2024)

PixArt-$α$: Fast Training of Diffusion Transformer for Photorealistic Text-to-Image Synthesis
by: Chen, Junsong, et al.
Published: (2023)

SANA-WM: Efficient Minute-Scale World Modeling with Hybrid Linear Diffusion Transformer
by: Zhu, Haoyi, et al.
Published: (2026)

Defeating the Training-Inference Mismatch via FP16
by: Qi, Penghui, et al.
Published: (2025)

Learning Single Index Models with Diffusion Priors
by: Tang, Anqi, et al.
Published: (2025)

Data-regularized Reinforcement Learning for Diffusion Models at Scale
by: Ye, Haotian, et al.
Published: (2025)

SANA 1.5: Efficient Scaling of Training-Time and Inference-Time Compute in Linear Diffusion Transformer
by: Xie, Enze, et al.
Published: (2025)

FPSAttention: Training-Aware FP8 and Sparsity Co-Design for Fast Video Diffusion
by: Liu, Akide, et al.
Published: (2025)

DC-AE 1.5: Accelerating Diffusion Model Convergence with Structured Latent Space
by: Chen, Junyu, et al.
Published: (2025)

Deep Compression Autoencoder for Efficient High-Resolution Diffusion Models
by: Chen, Junyu, et al.
Published: (2024)

FP8-RL: A Practical and Stable Low-Precision Stack for LLM Reinforcement Learning
by: Qiu, Zhaopeng, et al.
Published: (2026)

Noise Consistency Training: A Native Approach for One-Step Generator in Learning Additional Controls
by: Luo, Yihong, et al.
Published: (2025)

Towards Fully FP8 GEMM LLM Training at Scale
by: Hernández-Cano, Alejandro, et al.
Published: (2025)

SGEMM-cube: Precision-Recovery FP32 GEMM Approximation on Ascend NPUs with FP16 Matrix Engines
by: Xue, Weicheng, et al.
Published: (2025)

FlashMLA-ETAP: Efficient Transpose Attention Pipeline for Accelerating MLA Inference on NVIDIA H20 GPUs
by: Dege, Pengcuo, et al.
Published: (2025)

Not All Rollouts are Useful: Down-Sampling Rollouts in LLM Reinforcement Learning
by: Xu, Yixuan Even, et al.
Published: (2025)

Lookahead Tree-Based Rollouts for Enhanced Trajectory-Level Exploration in Reinforcement Learning with Verifiable Rewards
by: Xing, Shangyu, et al.
Published: (2025)

$μ$nit Scaling: Simple and Scalable FP8 LLM Training
by: Narayan, Saaketh, et al.
Published: (2025)

Each Prompt Matters: Scaling Reinforcement Learning Without Wasting Rollouts on Hundred-Billion-Scale MoE
by: Zeng, Anxiang, et al.
Published: (2025)

PIXART-δ: Fast and Controllable Image Generation with Latent Consistency Models
by: Chen, Junsong, et al.
Published: (2024)

Leveraging Error Diversity in Group Rollouts for Reinforcement Learning
by: Liu, Wenpu, et al.
Published: (2026)

DiffusionRollout: Uncertainty-Aware Rollout Planning in Long-Horizon PDE Solving
by: Yoo, Seungwoo, et al.
Published: (2026)

DC-Gen: Post-Training Diffusion Acceleration with Deeply Compressed Latent Space
by: He, Wenkun, et al.
Published: (2025)

On Rollouts in Model-Based Reinforcement Learning
by: Frauenknecht, Bernd, et al.
Published: (2025)

SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformers
by: Xie, Enze, et al.
Published: (2024)

SANA-Streaming: Real-time Streaming Video Editing with Hybrid Diffusion Transformer
by: Zhao, Yuyang, et al.
Published: (2026)

MOSS: Efficient and Accurate FP8 LLM Training with Microscaling and Automatic Scaling
by: Zhang, Yu, et al.
Published: (2025)

GraphDancer: Training LLMs to Explore and Reason over Graphs via Two-Stage Curriculum Post-Training
by: Bai, Yuyang, et al.
Published: (2026)

Rollout-Training Co-Design for Efficient LLM-Based Multi-Agent Reinforcement Learning
by: Jiang, Zhida, et al.
Published: (2026)

Tape: A Cellular Automata Benchmark for Evaluating Rule-Shift Generalization in Reinforcement Learning
by: Pan, Enze
Published: (2026)

DyDiff: Long-Horizon Rollout via Dynamics Diffusion for Offline Reinforcement Learning
by: Zhao, Hanye, et al.
Published: (2024)

Guided Cooperation in Hierarchical Reinforcement Learning via Model-based Rollout
by: Wang, Haoran, et al.
Published: (2023)

Towards Efficient Pre-training: Exploring FP4 Precision in Large Language Models
by: Zhou, Jiecheng, et al.
Published: (2025)

Model-Driven Subspaces for Large-Scale Optimization with Local Approximation Strategy
by: He, Yitong, et al.
Published: (2025)