model-fine-tuning

Overview

Fine-tune pre-trained models for specific tasks. Covers LoRA/QLoRA for efficient training, dataset preparation, evaluation, and deployment of custom models.

Capabilities

• Fine-tune LLMs with LoRA and QLoRA (low-rank adaptation)
• Prepare datasets in instruction/chat format
• Use Hugging Face Transformers + PEFT for training
• Evaluate fine-tuned models with benchmarks
• Merge LoRA adapters back into base models
• Deploy fine-tuned models via vLLM or Ollama

When to Use

Trigger phrases:

• "model fine tuning"

• "Fine-tune LLMs and ML models — LoRA, QLoRA, PEFT, Hugging Face"

• Need a model specialized for a specific domain (legal, medical, code)

• Want better performance on specific tasks than general models

• Have domain-specific data that improves with training

• Need to reduce model size while maintaining quality

• Building a product that needs a custom AI model

When NOT to Use

• Task is outside your authorization scope
• You need to implement controls (use implementing-* skills)
• Task is about analysis, not action (use analyzing-* skills)
• You don't have access to target systems
• Task requires compliance expertise (consult professionals)
• Task is about defense, not offense (use defensive skills)

Pseudo Code

# Example workflow for this skill
def execute(input_data):
    # Step 1: Validate input
    if not input_data:
        raise ValueError("Input data is required")

    # Step 2: Process core logic
    result = process(input_data)

    # Step 3: Validate output
    validate_output(result)

    return result

Dataset Preparation (Hugging Face Format)

from datasets import Dataset

# Instruction format
data = [
    {"instruction": "Summarize this text", "input": "Long article...", "output": "Summary..."},
    {"instruction": "Translate to French", "input": "Hello world", "output": "Bonjour le monde"},
]

# Chat format (for chat models)
data = [
    {"messages": [
        {"role": "system", "content": "You are a legal assistant."},
        {"role": "user", "content": "What is a contract?"},
        {"role": "assistant", "content": "A contract is a legally binding agreement..."}
    ]},
]

dataset = Dataset.from_list(data)
dataset.push_to_hub("username/my-dataset")