SGLang Team Open-Sources Miles v0.1 Production Post-Training Framework

The SGLang team open-sourced Miles v0.1, a full-stack production post-training framework whose paper and GitHub repo are public. It features one-command Docker setup, trains a 744B model on 64 GPUs at 263 seconds per step, and supports LoRA and diffusion models.

2026-09-13 ~ 2026-09-13 · 2 related posts

• SGLang Team Open-Sources Miles v0.1: 744B Model RL on 64 GPUs at 263s/Step — aigclink · 2026-09-13
• Miles v0.1 Paper: Full-Stack Production Post-Training with LoRA and Diffusion Support — aigclink · 2026-09-13