Blog

Explore RLinf engineering reports, system optimizations, and model integrations.

Comparison with VeRL

Benchmark setup, configuration, and performance comparisons.

Comparison with VeRL

Online RL for Code Completion

Connect code completion feedback to online policy training.

A First Look at the “Last Mile” of Agent Deployment: Cursor Online Reinforcement Learning

BEHAVIOR System Optimization

Accelerate environment interaction for household manipulation tasks.

Accelerating the “ImageNet Moment” of Embodied AI: RLinf Brings a 25× System Optimization to BEHAVIOR

Cosmos3 with SGLang

Fine-tune Cosmos3 and evaluate it with parallel inference.

RLinf x SGLang: Integrating Cosmos3 from Fine-Tuning to Efficient Parallel Evaluation

WorldLoop and WoVR

Use a training-free VLM reward for world-model RL.

RLinf × WorldLoop: Extending WoVR with a Training-Free VLM Reward

WoVR on Moore Threads S5000

MUSA adaptation, accuracy comparisons, and eight- and sixteen-GPU performance.

RLinf × MUSA: Reproducing WoVR on Moore Threads S5000