# DigitalOcean's MOP agent runs SGLang trials to find serving recipes

> Original title: Find the serving recipe that fits the workload.

- Source: LMSYS Org
- Published: 2026-11-09T01:58:47.000Z
- Bad Influence: https://badinfluence.news/items/gjl8d2l52vkdebs0ohkgce5g4
- Original: https://x.com/lmsysorg/status/2119610623675990386

## Summary

DigitalOcean's Balaji Varadarajan describes MOP, a self-learning agent that runs SGLang trials and diagnoses bottlenecks to optimize model serving. MOP uses bandit-driven exploration to choose the next experiment for each workload.

## Post

Find the serving recipe that fits the workload.

Balaji Varadarajan from @digitalocean on MOP, a self-learning agent that runs SGLang trials, diagnoses bottlenecks and uses bandit-driven exploration to choose the next experiment.

![](https://badinfluence.news/api/img-proxy?u=https%3A%2F%2Fpbs.twimg.com%2Fmedia%2FHWpZXtPaYAA2s6u.jpg&mode=thumb&exp=1794528000&sig=9eeb01308b454647)