SGLang Summit Workshop by AMD: Scalable agentic RL on AMD Instinct GPUs with Miles

A SGLang Summit ticket is required to attend the event and get approved on Luma. Both Full and Expo passes are accepted.

Ticket 🎫 https://luma.com/sglang-summit

The room seats 45, and seating is first come, first served, so registering doesn't guarantee a seat. Please arrive early to get a spot.Reinforcement learning for coding agents is as much a systems problem as an algorithmic one. Each training sample is a full multi-turn episode, with dozens of shell commands run in an isolated sandbox, and a reward only if the repository's tests pass. We walk through where the time goes, how to scale sandboxes, how weight sync and train–rollout consistency affect training, and the practical fixes that mattered most. This intermediate-to-advanced, hands-on workshop shows how to run scalable agentic RL on AMD GPUs with Miles, SGLang, Megatron-LM and Harbor. Participants will run multi-turn agentic RL on mixed datasets such as SWE-bench and Terminal-Bench.

Session Details

Duration: 90 minutes

Format: Hands-on. AMD provides remote GPU access through Jupyter notebooks, so there's nothing to install. Everything runs in your browser.

Level: Intermediate to advanced. Not designed for beginners.

What to bring: Your own laptop

Before You Arrive

You'll need an AMD Developer Program account to access the GPUs during the workshop. We'll email registered attendees the sign-up link ahead of time.

Speakers

Liz Li, Principal AI Engineer, AMD

Sree Rohith Pulipaka, Sr. Software Engineer AI, AMD