model-training-expert

Configure, execute, and evaluate AI model training workflows with LoRA adapters.

Updated Dec 30, 2025

Search Tags:#llm #evaluation #adapter #model-training #synthetic-data #hmoe #vast-ai

Ready to run

npx skills add https://github.com/scawful/afs_scawful --skill model-training-expert

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: model-training-expert
Source: https://github.com/scawful/afs_scawful/tree/main/skills/model-training-expert
Command: npx skills add https://github.com/scawful/afs_scawful --skill model-training-expert

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve?

This Skill provides end-to-end guidance for configuring, executing, and evaluating AI model training workflows, with emphasis on recursive data generation, MoE management, and tool orchestration to accelerate research and production pipelines.

Core Features & Use Cases

Hierarchical MoE (H-MoE) setup and adapter strategies using LoRA to enable domain-specific specialization and hot-swapping without reloading base models.
Synthetic Data Evolution (SDE) workflows from draft generation to verification, correction, and dataset expansion, including teacher-model-based failure analysis.
Agentic Evaluation (AgE) workflows leveraging Agahnim and HAFS, with sandboxed emulation (Mesen2) and multi-modal ingestion guidance for technical papers to support tooling and experiments.

Quick Start

Provide an end-to-end training objective and initiate a full setup for a 7B LLM using LoRA adapters on Vast.ai.

Frequently Asked Questions about model-training-expert

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I configure LoRA adapters for a 7B LLM training workflow?▼

To configure LoRA adapters for a 7B LLM training workflow, provide an end-to-end training objective to initiate a full setup. This enforces modular backbone choices and hot-swapping without reloading base models.

What is synthetic data evolution and how does it improve model training?▼

Synthetic data evolution improves model training through workflows spanning draft generation to verification, correction, and dataset expansion. It leverages teacher-model-based failure analysis to recursively generate and refine training data.

How do I set up a Hierarchical MoE for domain-specific model specialization?▼

Setting up a Hierarchical MoE for domain-specific specialization involves configuring adapter strategies using LoRA. This approach enables hot-swapping specialized modules without reloading the base model.

Can I run agentic evaluation workflows in a sandboxed environment?▼

Yes, you can run agentic evaluation workflows in a sandboxed environment using Mesen2 emulation. This leverages Agahnim and HAFS, providing multi-modal ingestion guidance for technical papers to support tooling and experiments.

Does this model training workflow support deployment on Vast.ai?▼

Yes, the model training workflow supports deployment on Vast.ai. You can initiate a full setup for a 7B LLM using LoRA adapters directly on the platform to accelerate research pipelines.

What is the best way to evaluate AI models using agentic workflows?▼

The best way to evaluate AI models using agentic workflows is through AgE leveraging Agahnim and HAFS. This provides sandboxed emulation and multi-modal ingestion guidance for technical papers to validate tooling.