What are 'Wardrobes' in this context?

Wardrobes are deployment configurations that act as a 'closet' of specialized LoRA adapters, allowing an agent to switch between different domain expertise instantly.

What model does this skill train?

This skill specifically targets the Gemma 270M model, creating lightweight LoRA adapters for domain-specific tasks while keeping the base model frozen.

Can this skill be used for autonomous deployment?

Yes, it generates an execution-ready deployment configuration that enables autonomous agents to deploy and verify new model versions without manual intervention.

How is the training data validated?

The skill performs automated schema validation on JSON datasets and calculates semantic similarity between generated outputs and validation sets to ensure a minimum fidelity threshold of 0.90.

Do I need a high-end GPU to use the Gemma Domain Trainer?

While GPU acceleration is supported, the skill is designed to be CPU-friendly, allowing for fast training of small adapters (~10MB) in approximately 5-10 minutes.

Gemma Domain Trainer

Name: Gemma Domain Trainer
Author: Foundup

byFoundup

•

Data Science & ML

Automates the fine-tuning of Gemma 270M models using LoRA adapters for domain-specific tasks and autonomous operations.

The Gemma Domain Trainer is a specialized skill designed to orchestrate the creation of domain-specific AI experts within the Claude Code environment. By utilizing LoRA (Low-Rank Adaptation), it fine-tunes the compact Gemma 270M model on instruction-tuning datasets extracted from project knowledge bases. This skill manages the entire lifecycle of model improvement—from data formatting and hyperparameter tuning to validation and deployment configuration—allowing developers to swap 'model wardrobes' for different tasks like autonomous cleanup, project scoring, or issue triage without the overhead of reloading base models.

Key Features

01Generation of 'Wardrobe' configs for seamless model swapping

02Intelligent hyperparameter configuration based on domain complexity

034 GitHub stars

04Validation monitoring with semantic similarity and accuracy scoring

05Automated LoRA adapter training for Gemma 270M models

06CPU-friendly training orchestration for localized model specialization

Use Cases

01Deploying domain-expert adapters for recursive multi-agent workflows

02Fine-tuning lightweight models on proprietary project documentation

03Creating specialized models for autonomous task prioritization and scoring

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add foundup/foundups-agent gemma_domain_trainer_prototype

For use in Claude.ai and ChatGPT

Key Features

01Generation of 'Wardrobe' configs for seamless model swapping

02Intelligent hyperparameter configuration based on domain complexity

034 GitHub stars

04Validation monitoring with semantic similarity and accuracy scoring

05Automated LoRA adapter training for Gemma 270M models

06CPU-friendly training orchestration for localized model specialization

Use Cases

01Deploying domain-expert adapters for recursive multi-agent workflows

02Fine-tuning lightweight models on proprietary project documentation

03Creating specialized models for autonomous task prioritization and scoring

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add foundup/foundups-agent gemma_domain_trainer_prototype

For use in Claude.ai and ChatGPT