Unified framework for efficiently fine-tuning 100+ LLMs and VLMs with a zero-code CLI and Web UI.
75.6K9.3KPythonApache-2.0Updated 22 hours ago
axolotl-ai-cloud/axolotl
A free, open-source framework for fine-tuning and post-training large language models, configured through YAML files.
- Stars
- 12,592
- Forks
- 1,637
- Open issues
- 253
- Last commit
- 6 hours ago
What is it?
Axolotl is a framework for fine-tuning and post-training large language models. It covers the full pipeline from dataset preprocessing and training through evaluation, quantization and inference, and a single YAML configuration file drives each stage.
It supports full fine-tuning, LoRA, QLoRA, GPTQ, QAT and preference or reinforcement learning methods such as DPO, KTO, ORPO and GRPO. Training covers many model families from the Hugging Face Hub, including vision-language and audio models, and datasets can be loaded from local disk, Hugging Face, or S3, Azure, GCP and OCI storage.
For performance, it integrates options such as multipacking, Flash Attention, Liger Kernel, FSDP2 and DeepSpeed multi-GPU training, and multi-node training through Torchrun or Ray. Docker images and PyPI packages are provided for cloud and local setups.
Key features
- 01Single YAML config covering preprocessing, training, evaluation, quantization and inference
- 02Supports full fine-tuning, LoRA, QLoRA, GPTQ and QAT
- 03Preference tuning (DPO, KTO, ORPO) and RL (GRPO) methods
- 04Multimodal fine-tuning for vision-language and audio models
- 05Multi-GPU training with FSDP2 and DeepSpeed, multi-node with Torchrun or Ray
- 06Loads datasets from local storage, Hugging Face and major cloud object stores
Use cases
- Fine-tuning an open-weight LLM on a custom instruction or chat dataset with LoRA
- Running preference optimization or GRPO on a model using a reproducible YAML config
- Scaling a training job across multiple GPUs or nodes on cloud infrastructure
Topics
FAQ
- What is axolotl?
- A free, open-source framework for fine-tuning and post-training large language models, configured through YAML files. Axolotl is a framework for fine-tuning and post-training large language models. It covers the full pipeline from dataset preprocessing and training through evaluation, quantization and inference, and a single YAML configuration file drives each stage.
- Is axolotl free to use? What license does it have?
- axolotl is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
- Is axolotl still maintained?
- Actively maintained. The last commit to axolotl was on Oct 20, 2026.
- What can I use axolotl for?
- Fine-tuning an open-weight LLM on a custom instruction or chat dataset with LoRA Running preference optimization or GRPO on a model using a reproducible YAML config Scaling a training job across multiple GPUs or nodes on cloud infrastructure
- Where can I get axolotl?
- The source code and installation instructions are on GitHub: https://github.com/axolotl-ai-cloud/axolotl
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
CLI tool to fine-tune and post-train LLMs from a single YAML config, with options for training on low-VRAM consumer GPUs.
9.4K1.5KPythonApache-2.0Updated yesterday
Python library that runs large language models layer by layer so 70B-class models can infer on a single 4GB GPU, without quantization.
39.5K4KJupyter NotebookApache-2.0Updated yesterday
A from-scratch PyTorch walkthrough and codebase for training a transformer LLM, covering pretraining, SFT, reward modeling, DPO/PPO/GRPO and generation.
12.3K1.8KPythonMITUpdated 18 days ago
A bilingual collection of 100+ original diagrams explaining LLM, RL, RLHF, DPO, GRPO and related large-model algorithms.
5K478PythonOtherUpdated last month
A deep learning optimization library for making distributed training and inference of large models easier and more efficient.
43.2K5KPythonApache-2.0Updated yesterday