Intended Audience
This document describes how to use the Python APIs of the LLM-DataDist component in the Unilateral communication library to implement point-to-point data transmission and build a disaggregated architecture for LLM inference, thereby improving the throughput performance of the LLM inference system.
To better understand this document, you are expected to:
- Have basic understanding of LLM structure and concepts.
- Be familiar with the LLM inference process.
- Be proficient in Python programming.
- Be experienced in LLM inference frameworks such as vLLM.
Parent topic: Learning Wizard