---
name: "axolotl-llm-finetuning-config"
description: "Fine-tune open-source language models efficiently with Axolotl YAML configurations."
metadata:
  author: "OpenAccess AI Collective"
  version: "1.0.0"
  runtime_compatibility:
    - "codex"
    - "claude-code"
    - "cursor"
    - "windsurf"
triggers:
  - "fine tune model with axolotl"
  - "configure qlora axolotl yaml"
  - "train llama 3 with axolotl"
permissions:
  fs_read: true
  fs_write: "workspace"
  shell_exec: false
  net_access: false
---

# Axolotl LoRA & QLoRA LLM Fine-Tuning Pipeline

Orchestrates LLM fine-tuning pipelines using Axolotl: LoRA, QLoRA, full parameter tuning, FlashAttention-2, DeepSpeed ZeRO-2/3, multi-GPU FSDP, and Weights & Biases telemetry.

## When to Invoke
Invoke this skill whenever you need to:
- fine tune model with axolotl
- configure qlora axolotl yaml
- train llama 3 with axolotl

## Execution SOP & Rules
1. Define training hyperparameters in a single declarative YAML configuration file.
2. Use adapter: qlora and load_in_4bit: true to train large models on single GPUs.
3. Enable flash_attention: true to drastically cut memory usage and boost training speed.
4. Export merged 16-bit model weights using python -m axolotl.cli.merge_lora config.yml.

## Verification Checklist
- Run local linting, syntax validation, and type-checks before reporting completion.
- Ensure all created or modified files adhere to workspace boundaries and avoid uncommitted secrets.
- Verify tests or build commands exit cleanly with return code 0.