[Tracking] Improve Multimodal CI coverage · sgl-project/sglang#8496

2025-07-29T04:39:17.000Z

### Checklist - [ ] 1. If the issue you raised is not a feature but a question, please raise a discussion at https://github.com/sgl-project/sglang/discussions/new/choose Otherwise, it will be closed. - [ ] 2. Please use English, otherwise it will be closed. ### Motivation 1. All existing Multi-modal CI operations are executed on a single GPU, without considering the scenario of Tensor Parallelism (TP). There is a requirement to introduce test cases for TP-2/4 configurations. 2. The payload for VLM CI is relatively low. Stress tests are necessary, and simultaneously, it is crucial to investigate whether there are any memory leaks. 3. At present, the majority of the VLM CI is only to vision. It is essential to expand its scope to audio. 4. Welcome to add more. ### Related resources The test is under `test/srt/test_vision_openai_server_x` Related PRs: - https://github.com/sgl-project/sglang/pull/8428 - https://github.com/sgl-project/sglang/pull/7519

(3 commentaires) (5 réactions) (0 assignés)Python (6 216 forks)auto 404

Multi-modalcigood first issueperformance

Métriques du dépôt

Stars: (28 442 stars)
Métriques de merge PR: (Merge moyen 2j 1h) (1 000 PRs mergées en 30 j)

Description

Checklist

1. If the issue you raised is not a feature but a question, please raise a discussion at https://github.com/sgl-project/sglang/discussions/new/choose Otherwise, it will be closed.
2. Please use English, otherwise it will be closed.

Motivation

All existing Multi-modal CI operations are executed on a single GPU, without considering the scenario of Tensor Parallelism (TP). There is a requirement to introduce test cases for TP-2/4 configurations.
The payload for VLM CI is relatively low. Stress tests are necessary, and simultaneously, it is crucial to investigate whether there are any memory leaks.
At present, the majority of the VLM CI is only to vision. It is essential to expand its scope to audio.
Welcome to add more.

Related resources

The test is under test/srt/test_vision_openai_server_x

Related PRs:

Guide contributeur

Direction de recherche: Examinez les tests CI VLM existants sous test/srt/test vision openai server x, identifiez les lacunes pour les configurations TP, les tests de stress, les tests audio, et proposez ou implémentez des cas de test. Reportez vous aux PR associés #8428 et #7519.
Stack technique: python
Domaine: devops
Type d'issue: Test
Difficulté: 3
Temps estimé: Plus d'une semaine
Statut d'activité: Active
Clarté: Claire
Prérequis: Git
Accessibilité débutant: 65