Github
Discord

We make fine-tuning

:accessible
>scalable
=fun

Axolotl is an Open Source tool to make fine-tuning AI models friendly, fast and fun - without sacrificing functionality or scale.

Learn more

Supported LLMs

HuggingFace
Cerebras
XGen
Qwen
RWKV
Gemma
MS Phi
Mistral AI
llama
Eleuther AI
Falcon
MPT

Our Open Source community makes it easy and fun to learn and get started with model fine-tuning.

{friendly}
User icon

If I’ve never done any fine-tuning before, I can learn about it from the Axolotl community – with over 100 contributors and 500+ active Discord members, we have answers to every kind of question.

Sliders

There are pre-existing configurations (in YAML) provided in the OSS repo that makes it easy to start training a model without having to tweak any settings.

Speech bubbles

When I want to experiment with fine-tuning the latest models, I know I can ask in the Axolotl Discord, and Axolotl also integrates nicely with Runpod, Latitude, Modal, and Jarvislabs, making it super easy for me to get started without configuring hardware.

Use the latest techniques first, and fine-tune models fast.

<fast>
Speedometer

When new techniques are discovered in the quickly moving research world of AI models and fine-tuning, they hit Axolotl very quickly. From FSDP+qLoRA fixes, LoRA+, Multipack, PEFT/LoRA, you can rely on Axolotl for quick integration of high performance fine-tuning techniques

Flash

Axolotl’s amazing community implementations mean that you can leverage the latest technology to quickly and efficiently fine-tune models – sometimes 3-5x faster than elsewhere.

Nodes

Supports wide range of model architectures supported by Huggingface transformers as well as the ability to compose techniques quickly with new model architectures.

Gen AI platforms, AI practitioners, researchers, and enterprises all trust Axolotl for fine-tuning at scale.

_powerful
CLI

You can run Axolotl anywhere – in your own cloud, on premise, with your own Docker and k8s setup, whatever you need. Enterprises use Axolotl on their own terms with massive results.

Data nodes

Bring your own data – Your data doesn’t have to be uploaded across the public internet to other finetuning platforms such as OpenAI, easing concerns around compliance and data governance.

Tech stack

Axolotl is operated at scale and trusted by research companies, model builders, Gen AI platforms, and AI practitioners like Teknium/Nous Research, Jeremy Howard, Modal, Replicate, OpenPipe.

Open source

[heroes]
@friends
/enablers

Axolotl is an Open Source tool to make finetuning AI models friendly and fast, without sacrificing functionality or scale.

Top contributors

All contributors

Sponsors

Modal
Become a sponsor

Axolotl cookbook

Get started with our pre made recipes and notebooks

Stay updated by subscribing to our newsletter "Tuned"