BBuf/vllm-qwen3-next-optimization
skills/model-optimization/vllm/vllm-qwen3-next-optimization/SKILL.md
PR-backed optimization manual for Qwen3-Next in vLLM. Use when an engineer needs to audit, debug, extend, or document Qwen3-Next GDN attention, MTP, packed module naming, PP, and cross-hardware tuned MoE configuration in vLLM.
$ install --global
skillsauthnpx skillsauth add BBuf/AI-Infra-Auto-Driven-SKILLS vllm-qwen3-next-optimizationInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
Security Scan Results
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
3
Scanners Passed
9
Scanners in report
Clean
TrivyContainer and dependency vulnerability scanner
95%
Clean
SemgrepStatic code analysis for vulnerabilities
95%
Clean
mcp-scan (Snyk)Model Context Protocol security validation
95%
Skipped
Snyk (dep)Open source security scanning
50%
Skipped
Socket.devSupply chain security analysis
50%
Skipped
VirusTotalMulti-engine malware detection
50%
Skipped
CrowdStrikeAdvanced threat intelligence
50%
Skipped
OSV-ScannerOpen Source Vulnerability database check
50%
Skipped
OWASP Dep-Check
50%
Last scanned: May 2, 2026, 2:43 AM180.7s2 files scanned
SKILL.md
- name:
- vllm-qwen3-next-optimization
- description:
- PR-backed optimization manual for Qwen3-Next in vLLM. Use when an engineer needs to audit, debug, extend, or document Qwen3-Next GDN attention, MTP, packed module naming, PP, and cross-hardware tuned MoE configuration in vLLM.
vLLM Qwen3-Next Optimization
Overview
This skill covers Qwen3-Next GDN attention, MTP, packed module naming, PP, and cross-hardware tuned MoE configuration in vLLM.
Evidence snapshot:
- vLLM mainline checked around
0f7be0f2f76814f80f9091220a5fbbb53912ad00 - Support status: supported on current mainline
- Canonical PR notes:
references/pr-history.md - History mirrors:
model-pr-optimization-history/vllm/qwen3-next/README.zh.mdandREADME.en.md
Non-Negotiable Evidence Rule
Use skills/model-optimization/model-pr-diff-dossier/SKILL.md as the production bar.
Every PR cited for this family must be based on diff reading, not only PR titles.
Runtime Surfaces
vllm/vllm/model_executor/models/qwen3_next.pyvllm/vllm/model_executor/models/qwen3_next_mtp.py
Current Main Summary
- Qwen3-Next is its own runtime family because of Gated DeltaNet attention and its MTP path.
- The practical risks are PP, MTP varlen handling, quantized shared-expert naming, and GDN-specific CUDA graph bugs.
Key Landed PRs
- #24709
Fix Qwen3-Next PP: Corrected pipeline-parallel execution on Qwen3-Next. - #24957
Fix the varlen issue in qwen3-next MTP implementation: Removed a concrete MTP correctness bug on variable-length batches. - #24960
Add prefixes to shared_expert in qwen3-next: Fixed ignored-parameter and quantized weight loading for shared experts. - #25743
Fix cuda graph capture bug in GDN metadata and a stride bug: Stabilized GDN execution under CUDA graphs. - #31722
Speed-up of GDN attention decode part: Improved decode throughput on the GDN attention path. - #33657
Initial support for GDN attention on Qwen3-next/Qwen3.5 (XPU): Extended the family beyond CUDA with XPU GDN coverage.
Open Optimization Items
- Re-run PR search before claiming new support beyond the checked mainline commit.
Validation Lanes
- Qwen3-Next PP and MTP correctness.
- GDN decode throughput / CUDA graph capture.
- Quantized shared-expert loading.
- XPU GDN fallback if relevant.
Related Skills
BBuf/vllm-sota-humanize-loop
development
VerifiedTrustedCommunity
Run an autonomous Humanize-governed vLLM SOTA performance loop for one LLM model: first perform the fixed fair vLLM/SGLang/TensorRT-LLM deployment search and benchmark, then start one RLCR loop that repeatedly decides the gap, profiles the current bottleneck, runs layer/kernel pipeline analysis, patches vLLM code, optionally uses ncu-report-skill for kernel evidence, and revalidates until vLLM matches or beats the best observed framework under the same workload and SLA.
423SKILL.mdUpdated May 27, 2026
BBuf/llm-pipeline-analysis
devops
VerifiedTrustedCommunity
Inspect LLM torch profiler traces at forward-pass, layer, and kernel level. Use when you need layer timings, anchor-kernel boundaries, representative kernel flows, or Perfetto time ranges.
423SKILL.mdUpdated May 21, 2026
BBuf/sglang-sota-humanize-loop
development
VerifiedTrustedCommunity
Run an autonomous Humanize-governed SGLang SOTA performance loop for one LLM model: first perform the fixed fair SGLang/vLLM/TensorRT-LLM deployment search and benchmark, then start one RLCR loop that repeatedly decides the gap, profiles the current bottleneck, runs layer/kernel pipeline analysis, patches SGLang code, optionally uses ncu-report-skill for kernel evidence, and revalidates until SGLang matches or beats the best observed framework under the same workload and SLA.
423SKILL.mdUpdated May 17, 2026
BBuf/model-pr-history-knowledge
documentation
VerifiedTrustedCommunity
Use when an SGLang, vLLM, or TensorRT-LLM serving/model optimization task needs prior model-family PR evidence. Query and read the PR-driven history docs under model-pr-optimization-history before choosing source paths, fast paths, kernel/fusion ideas, regression risks, or validation lanes.
423SKILL.mdUpdated May 17, 2026