skills.rest
musical-basics
musical-basicsCOMMUNITY
Explore all skills by this author →
GitHub

axolotl

Guide Axolotl fine-tuning with YAML configs, adapters, and distributed training.

Updated Apr 11, 2026
Search Tags:#llm#fine-tuning#lora#fsdp#qlora#deepspeed#axolotl
One-click install
npx skills add https://github.com/musical-basics/hermes-build-2 --skill axolotl-musical-basics
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: axolotl
Source: https://github.com/musical-basics/hermes-build-2/tree/main/skills/mlops/training/axolotl
Command: npx skills add https://github.com/musical-basics/hermes-build-2 --skill axolotl-musical-basics

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Provides expert, operational guidance to configure and run large language model fine-tuning workflows with Axolotl, removing guesswork around YAML configs, distributed training, adapters, and common failure modes so engineers and MLOps practitioners can run reliable training jobs.

Core Features & Use Cases

  • Comprehensive Config Patterns: YAML examples and recommended settings for FSDP, DeepSpeed, mixed precision (fp16/bf16/fp8), and performance tuning.
  • Adapter & Quantization Support: Instructions for LoRA/QLoRA, PEFT integration, adapter merging, and post-training quantization.
  • RL/RLHF & Preference Methods: Guidance for DPO, KTO, ORPO, GRPO workflows and dataset formats for preference-based training.
  • Debugging & Multi-GPU Workflows: Tips for NCCL, sequence/sequence-parallel sizing, sampler strategies, monkeypatches, and common environment fixes.
  • Use Case Example: Prepare an Axolotl YAML to fine-tune a 7B model with QLoRA and FSDP across 8 GPUs, using a pretokenized dataset and LoRA adapters for fast iteration.

Quick Start

Run a LoRA/QLoRA fine-tuning job using an Axolotl YAML config that specifies the base model, dataset, adapter settings, and the distributed launcher.

Frequently Asked Questions about axolotl

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

FAQPage Schema
How do I configure a YAML file for LLM fine-tuning with QLoRA and FSDP?▼

LLM fine-tuning with QLoRA and FSDP requires specifying base model paths, pretokenized datasets, adapter settings, and the FSDP distributed launcher in your YAML config. You can apply recommended patterns for mixed precision and sequence parallel sizing to optimize multi-GPU training.

What is the difference between DPO, KTO, ORPO, and GRPO for preference-based LLM training?▼

DPO, KTO, ORPO, and GRPO are preference methods for LLM training that require specific dataset formats. You configure these preference-based training workflows in your YAML config to align model outputs without standard supervised fine-tuning.

Can I use DeepSpeed for multi-GPU training and mixed precision?▼

Yes, you can use DeepSpeed for multi-GPU training and mixed precision. The framework provides specific YAML configuration patterns and performance tuning guidance to set up fp16, bf16, or fp8 training across multiple GPUs reliably.

How do I merge LoRA adapters and apply post-training quantization?▼

To merge LoRA adapters and apply post-training quantization, you use the PEFT integration tools. This workflow combines trained adapter weights with the base model and prepares the quantized model for efficient inference deployment.

Why does my distributed training fail with NCCL errors during fine-tuning?▼

Distributed training NCCL errors during fine-tuning usually indicate environment misconfiguration. You can apply specific monkeypatches, sampler strategies, and common environment fixes to resolve these distributed training failure modes.

Does Axolotl support multimodal training workflows?▼

Yes, Axolotl supports multimodal training workflows. You can configure your YAML to train models on datasets that include images and text, applying the same adapter settings, mixed precision, and multi-GPU optimizations used in standard LLM fine-tuning.

Related Skills

Explore More Software Engineering Skills→
x-TheFoxx-TheFox

axolotl

Fine-tune large language models with YAML-driven LoRA, DPO, and GRPO configurations.

Community
Advanced
camelranchentertainmentcamelranchentertainment
1

fine-tuning-expert

Fine-tune language models with LoRA/QLoRA adapters on JSONL datasets.

Community
Advanced
yo-stevenyo-steven

mlm-bridge-training

Correlate Megatron-LM and Megatron Bridge training runs by matching configs and diagnosing loss divergence.

Community
Advanced
Supporter09Supporter09

fine-tuning-with-trl

Fine-tune Hugging Face Transformers models with TRL for instruction alignment and preference optimization.

Community
Advanced