axolotl-ai-cloud/axolotl

Health v2 · maintenance only

Adoption not part of the score

12408 stars · 1412 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

Axolotl is a free, open-source, config-driven framework for fine-tuning large language models, supporting SFT, preference learning (DPO/KTO/ORPO), RL (GRPO), and reward modeling via YAML configs. It supports LoRA, QLoRA, and full fine-tuning across a wide range of HuggingFace model architectures, including multimodal and MoE models, with distributed training via FSDP and DeepSpeed.

Use cases

• fine-tune an LLM with LoRA on my own instruction dataset
• run QLoRA fine-tuning of a large model on a single GPU
• train a model with DPO using preference pairs
• fine-tune a vision-language model on image-text data
• do GRPO reinforcement learning with a reward function
• merge LoRA adapters into a full model checkpoint
• fine-tune a MoE model with reduced VRAM
• chat interactively with a fine-tuned model for testing

When to choose

• you want a YAML-config-driven fine-tuning workflow with broad model support
• you need LoRA/QLoRA/full fine-tuning plus SFT, DPO, KTO, ORPO, or GRPO in one tool
• you train on NVIDIA or AMD GPUs, including multi-GPU and cloud setups
• you want cutting-edge features like MoE LoRA, expert parallelism, or BitNet fine-tuning

When to avoid

• you only need inference or serving of existing models without training
• you have no GPU and cannot use cloud GPU environments
• you need a no-code GUI trainer rather than config files and a CLI
• you train non-LLM models like image classifiers from scratch

Facets

framework · maturity active

llm-training machine-learning deep-learning cli large-language-models machine-learning deep-learning artificial-intelligence python cloud fine-tuning lora qlora sft dpo grpo huggingface deepspeed fsdp moe multimodal yaml-config linux docker gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="axolotl-ai-cloud/axolotl")