# tide

**Repository Path**: coxine/tide

## Basic Information

- **Project Name**: tide
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-09-20
- **Last Updated**: 2026-09-20

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# TIDE: Reducing VLM Serving Tail Latency with Selective Vision Encoding Offload

## Overview

Modern VLM serving introduces an additional vision encoding stage before language model inference. Under heterogeneous and bursty multimodal workloads, this stage may become a bottleneck and lead to increased queueing delay and tail latency.

TIDE explores an edge-cloud collaborative serving architecture that dynamically determines where vision encoding should be executed based on runtime system conditions.

## Acknowledgement

Thanks for the support of SJTU Kunpeng & Ascend Center of Excellence.

## Repository Status

> **Work in Progress**

This repository is currently under active development.