Agent skill

finetune-llm

LLM fine-tuning 教練式引導工作流程 v2。 核心功能：主動探索使用者痛點、引導明確目標、多任務管理、資料來源追蹤、完整版本 lineage。 支援：LoRA/QLoRA/DoRA 微調、SFT/ORPO/DPO 對齊、資料準備、Benchmark 評估、HuggingFace 部署。 特色：教練式引導、可重現的資料管線、多任務版本追蹤。 觸發詞：「訓練模型」「fine-tune」「微調」「LoRA」「建立新任務」「改善模型」「優化準確率」「資料管線」「任務管理」

View SKILL.md on GitHub Repository
Stars 0
Forks 0

Install this agent skill to your Project

npx add-skill https://github.com/p988744/nlp-skills/tree/main/skills/finetune-llm

SKILL.md

Fine-tune LLM v2 - 教練式引導工作流程

自動引導使用者完成 LLM 訓練專案，從痛點探索到模型部署。

核心理念

v2 採用「教練式引導」設計：

  • 前期激勵：主動探索使用者痛點和目標
  • 主動提問：引導釐清需求而非假設
  • 決策支援：根據資源和目標推薦最佳方案
  • 完整追蹤：資料來源、配置、模型全程可重現

快速開始

啟動教練引導

「我想訓練一個模型」
「幫我分析這個任務該怎麼做」
→ 觸發 goal-clarifier agent，主動引導釐清目標

管理多個任務

「列出所有任務」
「比較 entity-sentiment v1 和 v2」
→ 使用 /nlp-skills:tasks 指令

配置資料來源

「資料從 PostgreSQL 來」
「用 GPT 生成訓練資料」
→ 觸發 data-source-advisor agent

組件架構

Skills (4 個專精領域)

Skill 職責
llm-coach 教練式引導主入口
llm-knowledge 獨立知識庫
task-manager 多任務管理
data-pipeline 資料管線配置

Commands (7 個快捷指令)

指令 功能
/nlp-skills:coach 啟動教練式對話
/nlp-skills:tasks 列出所有任務狀態
/nlp-skills:new-task 建立新任務
/nlp-skills:data-source 配置資料來源
/nlp-skills:generate 生成專案結構
/nlp-skills:evaluate 執行評估分析
/nlp-skills:deploy 部署模型

Agents (4 個自主助手)

Agent 觸發時機 功能
goal-clarifier 偵測模糊需求 主動引導釐清目標
data-source-advisor 詢問資料來源 協助配置資料管線
problem-diagnoser 效能問題 自動診斷推薦改善
result-analyzer 訓練/評估後 分析結果決策建議

Hooks (2 個事件處理)

Hook 事件 動作
data-validation 資料更新後 自動驗證格式分佈
version-tracking 訓練完成 記錄完整 lineage

專案結構

每個任務是完全獨立的自包含專案：

{任務名稱}/
├── task.yaml               # 任務定義
├── data_source.yaml        # 資料來源配置（可重現）
├── versions/               # 版本追蹤（完整 lineage）
│   ├── v1/
│   │   ├── config.yaml     # 訓練配置快照
│   │   ├── data_snapshot.json  # 資料版本資訊
│   │   ├── results.json    # 評估結果
│   │   └── model_info.json # 模型資訊
│   └── v2/
├── data/
│   ├── raw/                # 原始資料
│   ├── train.jsonl
│   ├── valid.jsonl
│   └── test.jsonl
├── scripts/                # 執行腳本
│   ├── 01_regenerate_data.py   # 重新生成資料
│   ├── 02_validate_data.py
│   ├── 03_convert_format.py
│   ├── 04_train.py
│   ├── 05_evaluate.py
│   └── 06_upload_hf.py
├── configs/
├── models/
├── benchmarks/
└── docs/

資料來源配置

v2 的核心特色是可重現的資料管線：

yaml
# data_source.yaml
sources:
  - type: database
    connection: postgresql://user:pass@host/db
    query: "SELECT text, label FROM annotations"
    snapshot_date: 2026-01-06

  - type: api
    endpoint: https://api.example.com/data
    params:
      limit: 1000

  - type: web_scrape
    urls: ["https://..."]
    keywords: ["金融", "股票"]

  - type: llm_generated
    prompt_template: |
      生成 {count} 筆金融情感分析訓練資料...
    model: gpt-4o
    count: 500

regeneration:
  script: scripts/01_regenerate_data.py
  last_run: 2026-01-06T10:30:00

版本追蹤

完整的 lineage 追蹤每次迭代：

yaml
# versions/v2/lineage.yaml
version: v2
created: 2026-01-06T14:00:00
parent: v1

data:
  source_hash: abc123
  train_count: 500
  valid_count: 100
  test_count: 100

config:
  base_model: Qwen/Qwen3-4B
  method: sft
  lora_r: 64
  epochs: 6

results:
  macro_f1: 0.815
  accuracy: 0.82

changes:
  - "增加 LoRA rank 32 → 64"
  - "新增中立樣本 200 筆"

環境需求

本地開發

  • Python: uv run python
  • 模型測試: Ollama

遠端訓練

支援混合模式：

  • 本地 GPU
  • 遠端 SSH 伺服器
  • 雲端服務（AWS, GCP, RunPod）

相關文件

  • skills/llm-coach/SKILL.md - 教練引導
  • skills/llm-knowledge/SKILL.md - 知識庫
  • skills/task-manager/SKILL.md - 任務管理
  • skills/data-pipeline/SKILL.md - 資料管線

v2.0.0 - 教練式引導設計

Recommended Agent Skills

Expand your agent's capabilities with these related and highly-rated skills.

p988744/nlp-skills

data-pipeline

This skill should be used when the user asks to "configure data source", "data from database", "fetch data from API", "scrape web data", "generate training data with LLM", "regenerate data", "data pipeline", "where does data come from", or needs to set up reproducible data collection. Provides data source configuration, reproducibility tracking, and data regeneration capabilities.

0 0
Explore
p988744/nlp-skills

task-manager

This skill should be used when the user asks to "list tasks", "show all tasks", "compare versions", "switch to task X", "task status", "version history", "what tasks do I have", or needs to manage multiple training tasks. Provides multi-task management, version tracking, and performance comparison capabilities.

0 0
Explore
p988744/nlp-skills

writing-plans

This skill should be used when the user asks to "write a plan", "create training plan", "plan my fine-tuning", "make a plan for model training", or when starting a new training task that requires multiple steps. Creates detailed implementation plans with bite-sized tasks for LLM fine-tuning workflows.

0 0
Explore
p988744/nlp-skills

executing-plans

This skill should be used when the user asks to "execute plan", "run the plan", "start executing", "continue plan execution", or has a plan file ready to execute. Manages phased implementation with batch execution and checkpoint reviews.

0 0
Explore
p988744/nlp-skills

llm-coach

This skill should be used when the user asks to "train a model", "fine-tune", "build NLP model", "create training task", "optimize model performance", "improve accuracy", "what model should I use", or expresses vague training needs like "I want to do sentiment analysis" or "help me with NER". Provides coaching-style guidance to clarify goals, diagnose pain points, and recommend optimal training approaches.

0 0
Explore
p988744/nlp-skills

llm-knowledge

This skill should be used when the user asks "what is LoRA", "compare models", "which model is best for Chinese", "SFT vs DPO", "how to handle overfitting", "class imbalance solution", "model architecture", "training method comparison", or needs reference information about LLM fine-tuning. Provides structured knowledge base for models, methods, architectures, and troubleshooting.

0 0
Explore

Didn't find tool you were looking for?

Be as detailed as possible for better results