Skip to content
AI HOT (Curated Pool)

Tencent Hunyuan Releases UniRL, a Unified Multimodal RL Infrastructure

腾讯混元发布UniRL:统一多模态强化学习基础设施

Tencent Hunyuan released UniRL, using one post-training loop to cover diffusion and flow-matching models, LLM/VLM systems, and unified multimodal models, while open-sourcing two algorithms, DRPO and Flow-DPPO.

Why it matters: Tencent Hunyuan open-sources UniRL, putting RL training for diffusion, flow matching, language and vision-language models into one post-training loop, with models and algorithms decoupled for free combination.

Read the original ↗Export Markdown