sgl-project/sglang

[Feature] Support more multi-modal input for VLM

Open

#5,964 opened on May 2, 2025

View on GitHub
 (5 comments) (5 reactions) (1 assignee)Python (6,216 forks)auto 404
Multi-modalfeaturegood first issuehelp wanted

Repository metrics

Stars
 (28,442 stars)
PR merge metrics
 (Avg merge 2d 1h) (1,000 merged PRs in 30d)

Description

Motivation

The current endpoint only supports image data input, limiting its flexibility for diverse VLM use cases. We need additional input formats, particularly for RL applications: (Could be split into multiple PRs)

  • Pre-computed Image Embeddings
  • Pixel Values
  • Pixel Value Range Parameters (min_pixel/max_pixel) for qwen-vl

Welcome to propose more.

Benefits

  1. Enhanced flexibility for RL workflows
  2. Reduced preprocessing overhead
  3. Better integration with existing pipelines

Contributor guide

Research direction
Examine how pre computed image embeddings are currently implemented (already done). For pixel values, investigate the VLM model's forward pass and how image inputs are processed. Look for existing functions that convert pixel values to embeddings. Check the qwen vl implementation for min pixel/max pixel parameters.
Tech stack
python
Domain
backend
Issue type
Feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-2 days
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
Active
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
Clear
Prerequisites
GitPython
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
40