axolotl-ai-cloud/axolotl

Go ahead and axolotl questions

What it solves

Axolotl simplifies the complex process of post-training and fine-tuning large language models (LLMs). It provides a unified framework that removes the need to write custom training scripts for different models or techniques, allowing users to move from dataset preprocessing to inference using a single configuration file.

How it works

Users define their training parameters—such as the model, dataset, and fine-tuning method—in a YAML configuration file. Axolotl then handles the underlying orchestration, integrating various performance kernels and distributed training libraries to execute the training run across single or multiple GPUs and nodes.

Who it’s for

It is designed for AI researchers and developers who want to fine-tune state-of-the-art LLMs and vision-language models (VLMs) without managing the low-level boilerplate of training loops and hardware optimization.

Highlights

  • Broad Model Support: Compatible with LLaMA, Mistral, Mixtral, Qwen, and many other models from the Hugging Face Hub.
  • Diverse Training Methods: Supports full fine-tuning, LoRA, QLoRA, Preference Tuning (DPO, ORPO), RL (GRPO), and Reward Modelling.
  • Multimodal Capabilities: Ability to fine-tune vision-language models (VLMs) and audio models.
  • Advanced Optimizations: Includes Flash Attention 2/3/4, Sequence Parallelism, and support for FSDP and DeepSpeed for multi-GPU/multi-node scaling.
  • Flexible Data Loading: Supports datasets from local storage, Hugging Face, and various cloud providers (S3, Azure, GCP, OCI).