BBuf/sglang-mixtral-quark-int4fp8-moe-optimization
skills/model-optimization/sglang/sglang-mixtral-quark-int4fp8-moe-optimization/SKILL.md
PR-backed optimization manual for Mixtral-8x7B with SGLang's AMD-only quark_int4fp8_moe online MoE quantization. Use when an engineer needs to audit or extend Mixtral AMD quantization, online INT4-to-FP8 MoE loading, AITER fused-MoE execution, or the registered GSM8K regression test.
$ install --global
skillsauthnpx skillsauth add BBuf/AI-Infra-Auto-Driven-SKILLS sglang-mixtral-quark-int4fp8-moe-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, 2:37 AM255.0s2 files scanned
SKILL.md
- name:
- sglang-mixtral-quark-int4fp8-moe-optimization
- description:
- PR-backed optimization manual for Mixtral-8x7B with SGLang's AMD-only quark_int4fp8_moe online MoE quantization. Use when an engineer needs to audit or extend Mixtral AMD quantization, online INT4-to-FP8 MoE loading, AITER fused-MoE execution, or the registered GSM8K regression test.
SGLang Mixtral Quark INT4-FP8 MoE Optimization
Overview
quark_int4fp8_moe is an AMD-only online quantization path used by MoE checkpoints such as mistralai/Mixtral-8x7B-Instruct-v0.1. It loads high-precision MoE expert weights, quantizes them to packed INT4, stores scales, and executes with FP8-style MoE math on ROCm.
Current evidence snapshot:
- SGLang
origin/main:bca3dd958on2026-04-24 - Quantization config:
python/sglang/srt/layers/quantization/quark_int4fp8_moe.py - Test:
test/registered/quant/test_int4fp8_moe.py - Docs:
docs_new/docs/advanced_features/quantization.mdx - Diff-reviewed PRs: #7392, #17116, #23455
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.
Before You Change Anything
Capture:
- model id:
mistralai/Mixtral-8x7B-Instruct-v0.1 - hardware: AMD CDNA3/CDNA4 only
- launch flags:
--quantization quark_int4fp8_moe,--attention-backend triton,--tp 2 - test suite:
stage-b-test-2-gpu-large-amd - GSM8K threshold: score greater than
0.56
Core Principles
- This path is online quantization, not a pre-quantized checkpoint loader.
- It is HIP-only; non-AMD execution should fail loudly.
- MoE expert weights are sharded before online quantization when checkpoints are not pre-sharded.
- Packed INT4 weights and FP8/int4 scales are runtime contract fields on
FusedMoE. - Direct-file test execution matters because #23455 restored the test after missing
__main__handling.
Validation Lanes
- Launch Mixtral with
--quantization quark_int4fp8_moe --attention-backend triton --tp 2. - Run
test/registered/quant/test_int4fp8_moe.pyon AMD and verify GSM8K score above0.56. - Check that quantization refuses non-HIP devices.
- Inspect online quantization progress and memory behavior during weight load.
References
references/pr-history.md: diff-reviewed Mixtral/quark INT4-FP8 MoE PR cards.
Related Skills
BBuf/sglang-humanize-review
development
VerifiedTrustedCommunity
Perform SGLang code review in the style of human maintainers by consulting the full non-agent PR review episode corpus from project start through the latest refresh (June 2026), including inline review threads, top-level PR comments, review submissions, original multilingual text, and multi-round discussions. Use when reviewing SGLang PRs, diffs, patches, or local changes for correctness, tests, performance, GPU/runtime risks, API compatibility, and maintainability.
531SKILL.mdUpdated May 21, 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.
531SKILL.mdUpdated May 17, 2026
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