BBuf/vllm-deepseek-v3-r1-optimization
skills/model-optimization/vllm/vllm-deepseek-v3-r1-optimization/SKILL.md
PR-backed optimization manual for DeepSeek V3 / R1 in vLLM. Use when an engineer needs to audit, debug, extend, or document DeepSeek V3 and DeepSeek R1 MLA, MoE, packed-module loading, LoRA, MTP/Eagle, and quantized ROCm/CUDA validation paths.
$ install --global
skillsauthnpx skillsauth add BBuf/AI-Infra-Auto-Driven-SKILLS vllm-deepseek-v3-r1-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, 2:42 AM389.1s1 file scanned
SKILL.md
- name:
- vllm-deepseek-v3-r1-optimization
- description:
- PR-backed optimization manual for DeepSeek V3 / R1 in vLLM. Use when an engineer needs to audit, debug, extend, or document DeepSeek V3 and DeepSeek R1 MLA, MoE, packed-module loading, LoRA, MTP/Eagle, and quantized ROCm/CUDA validation paths.
vLLM DeepSeek V3 / R1 Optimization
Overview
This skill covers DeepSeek V3 and DeepSeek R1 MLA, MoE, packed-module loading, LoRA, MTP/Eagle, and quantized ROCm/CUDA validation paths.
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/deepseek-v3-r1/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/deepseek_v2.pyvllm/vllm/model_executor/models/deepseek_eagle.pyvllm/vllm/model_executor/models/deepseek_eagle3.pyvllm/vllm/model_executor/models/deepseek_mtp.py
Current Main Summary
DeepseekV2ForCausalLM/DeepseekV3ForCausalLMremain the shared runtime for V3 and R1.- The highest-risk regressions cluster around packed module mapping, quantized MLA/MoE weight loading, LoRA, and MTP draft paths.
- R1 validation should split BF16, FP8/ModelOpt, and compressed-tensors or ROCm lanes.
Key Landed PRs
- #22352
Add missingpacked_modules_mappingtoDeepseekV2ForCausalLM``: Fixed quantized and packed-weight loading for DeepSeek V2/V3/R1 style checkpoints. - #23971
Add LoRA support for DeepSeek models (V2, V3, R1-0528): Enabled adapter injection on the DeepSeek family rather than only base dense models. - #29545
Fix DeepSeek R1 MTP weight loading: Hardened R1 NextN / MTP draft loading after launch failures on draft weights. - #36247
Fix compressed-tensors quantization failure for DeepSeek-R1 on MI300x: Closed a production ROCm gap for compressed-tensors DeepSeek-R1 deployment.
Open Optimization Items
- Re-run PR search before claiming new support beyond the checked mainline commit.
Validation Lanes
- BF16 DeepSeek-V3 / R1 serve plus generation sanity.
- FP8 or compressed-tensors startup on Hopper/Blackwell and MI300X.
- MTP / Eagle draft-model acceptance tests.
- LoRA adapter load and unload on the same checkpoint.
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