Agent Skills › R6410418/Jackrong-llm-finetuning-guide › qwen-mtp-gguf-release

qwen-mtp-gguf-release

GitHub

Qwen MTP GGUF发布工作流的仓库级封装，用于准备、验证、转换、量化和冒烟测试模型工件。通过预检和安全门控确保流程合规，仅在授权后上传至Hugging Face，防止误发布。

.agents/skills/qwen-mtp-gguf-release/SKILL.md R6410418/Jackrong-llm-finetuning-guide

Trigger Scenarios

需要将Qwen系列模型转换为GGUF格式执行模型量化或转换流水线对生成的GGUF工件进行冒烟测试在受控环境下发布模型到Hugging Face

Install

npx skills add R6410418/Jackrong-llm-finetuning-guide --skill qwen-mtp-gguf-release -g -y

More Options

Non-standard path

npx skills add https://github.com/R6410418/Jackrong-llm-finetuning-guide/tree/main/.agents/skills/qwen-mtp-gguf-release -g -y

Use without installing

npx skills use R6410418/Jackrong-llm-finetuning-guide@qwen-mtp-gguf-release

指定 Agent (Claude Code)

npx skills add R6410418/Jackrong-llm-finetuning-guide --skill qwen-mtp-gguf-release -a claude-code -g -y

安装 repo 全部 skill

npx skills add R6410418/Jackrong-llm-finetuning-guide --all -g -y

预览 repo 内 skill

npx skills add R6410418/Jackrong-llm-finetuning-guide --list

SKILL.md

Frontmatter

{
    "name": "qwen-mtp-gguf-release",
    "description": "Repository-level wrapper for the canonical Qwen MTP or nextn GGUF release workflow. Use when Codex needs to prepare, validate, convert, quantize, smoke-test, or optionally release Qwen-family GGUF artifacts from this repository while preserving the canonical implementation under qwen-mtp-gguf\/."
}

Qwen MTP GGUF Release

Use this as a lightweight repository wrapper. The canonical implementation lives under qwen-mtp-gguf/.

Required Reading

Read these files before running or changing the workflow:

qwen-mtp-gguf/SKILL.md
qwen-mtp-gguf/README.md

Load the reference files inside qwen-mtp-gguf/references/ only when the task touches sizing, extraction, conversion, troubleshooting, or agent packaging.

Operating Rules

Reuse qwen-mtp-gguf/scripts/ rather than copying conversion logic.
Run the canonical pipeline with --preflight-only before any conversion or model download.
Treat disk shortages, model-config mismatches, missing tools, and missing token access as blockers.
Keep examples public-safe with placeholders for model IDs, local paths, output repos, and tokens.
Keep Hugging Face upload disabled unless the user explicitly authorizes upload for that run.
Keep cleanup disabled unless confirmed uploads are verified.
Never publish or upload artifacts merely because GitHub sync was approved.

Canonical Flow

Inspect requested target model, MTP source model, output mode, quant formats, and available machine resources.
Run the canonical preflight command from qwen-mtp-gguf/SKILL.md.
Review preflight_report.md and stop on hard blockers.
Run the full canonical pipeline only after the user accepts the launch plan and safety gates.
Smoke-test outputs before any upload.
Upload only when separately authorized, then clean up only after successful upload confirmation.