Uploaded March 2024 | Updated September 2026, 2 weeks ago

---

Learn about the key challenges in improving efficiency of LLM serving, and an overview of multiple techniques the team is developing to address this problem. This talk also discusses matformers, a technique to train one model but read-off 100s of smaller models, along with techniques to speed up decoding in LLMs.

Watch more Research@ Bangalore → https://goo.gle/3VSup7d
Google Research Channel → https://goo.gle/GoogleResearch

#GoogleResearch