RLinf + SGLang runs Cosmos3 at 3.33x end-to-end eval throughput for robots
ying11231 · x · 2026-09-21
LMSYS's new blog details RLinf, an open-source embodied AI framework, now supporting Cosmos3 from fine-tuning to robot evaluation with 3.33x end-to-end evaluation throughput via SGLang. SGLang batches inference for 128 parallel environments on 8 GPUs covering 500 LIBERO-10 episodes, while RLinf overlaps CPU simulation with GPU inference to cut stage waiting.
More from Embodied
- 3D-printed RC glider gear replicates jet-style dead-point locking mechanism — TinfoilTricorn · 2026-09-21
- Ex-Tesla/Rivian Manufacturing Exec: Robotics Funding Shifts From Demos to Factory Floors — Sethwinterroth · 2026-09-21
- China is doing about 30 times more agricultural drone work than the US — yogthos · 2026-09-21
- Scobleizer hears robot upgrade lands Tuesday morning, wouldn't rule out a Monday ship — Scobleizer · 2026-09-21
- Two Amazon drones hit the same crane: SAA failure raises hard reliability questions — TinfoilTricorn · 2026-09-21
- HUMAIN's Horizon Ultra AI PC hits enterprise availability on Snapdragon X2 Elite — Codeblix_Ltd · 2026-09-21