sglang-qwen-vlm-omni-asr-optimization

Optimize Qwen multimodal pipelines in SGLang using PR-diff-driven workflows and runtime references.

721|65|Updated Apr 1, 2026

Search Tags:#optimization #playbook #qwen #asr #vlm #omni #diff-audit

Ready to run

npx skills add https://github.com/BBuf/AI-Infra-Auto-Driven-SKILLS --skill sglang-qwen-vlm-omni-asr-optimization

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: sglang-qwen-vlm-omni-asr-optimization
Source: https://github.com/BBuf/AI-Infra-Auto-Driven-SKILLS/tree/main/skills/model-optimization/sglang/sglang-qwen-vlm-omni-asr-optimization
Command: npx skills add https://github.com/BBuf/AI-Infra-Auto-Driven-SKILLS --skill sglang-qwen-vlm-omni-asr-optimization

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

PR-diff-backed optimization manual for Qwen2.5-VL, Qwen3-VL, Qwen3-VL-MoE, Qwen3-Omni, Qwen3-ASR, and Qwen3-5 multimodal paths in SGLang. Use when Codex needs to audit, debug, extend, or document multimodal processors, ViT DP/PP/chunk/cache, mRoPE, DeepStack, EAGLE3, LoRA, audio encoder, streaming ASR, encoder disaggregation, AMD/NPU/CPU support, or Qwen VLM cookbook deployment recipes.

Core Features & Use Cases

PR-diff-driven guidance: channels best-practice paths for Qwen multimodal optimization, including VLM, Omni, ASR, and DP/PP/encoder considerations.
Reference-driven workflow: integrates PR history and playbook references to reproduce and validate changes on real deployments.
Comprehensive scope: covers Qwen2.5-VL, Qwen3-VL, Qwen3-VL-MoE, Qwen3-Omni, Qwen3-ASR, and related cross-family recipes.

Quick Start

Review the PR history in references/pr-history.md, then consult references/playbook.md and begin following the Qwen optimization workflow.

Frequently Asked Questions about sglang-qwen-vlm-omni-asr-optimization

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I optimize Qwen VLM pipelines in SGLang using PR diffs?▼

You can optimize Qwen VLM pipelines in SGLang by following a PR-diff-driven framework that audits multimodal processors, ViT DP/PP chunking, mRoPE, and deployment cookbook recipes across Qwen2.5-VL and Qwen3-VL models.

What is the best way to debug Qwen3-Omni streaming ASR and audio encoder issues?▼

The best way to debug Qwen3-Omni streaming ASR and audio encoder issues is using a reference-driven workflow that integrates PR history and playbooks to reproduce and validate runtime tuning changes on real multimodal deployments.

Does this optimization playbook cover Qwen3-VL-MoE and DeepStack deployments?▼

Yes, the optimization playbook covers Qwen3-VL-MoE and DeepStack deployments. It provides structured guidance for extending and documenting multimodal processors, EAGLE3, LoRA, and encoder disaggregation across Qwen3 family models.

How do I update Qwen VLM cookbook deployment recipes for AMD or NPU environments?▼

You can update Qwen VLM cookbook deployment recipes for AMD or NPU environments by reviewing PR history to capture diffs and runtime files, then applying the playbook's safety considerations and explicit dependencies for hardware-specific support.

Why do I need PR history references for Qwen multimodal runtime tuning?▼

PR history references are needed for Qwen multimodal runtime tuning because they provide a structured, reference-driven workflow to reproduce and validate changes, ensuring safety and accuracy when modifying complex paths like mRoPE and ViT caching.