#6262nd of 68 in Training and fine-tuningFor voice, speech and music
stable-audio-tools
Training and inference code for diffusion-based audio generation models
- Stars
- 3.9k
- License
- MIT
- Last commit
- May 2026
- Language
- Python
3,651 on 25 Apr 20263,915 on 14 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
PyTorch Lightning codebase from Stability AI for training, fine-tuning and running audio generation models, configured through JSON model and dataset files. It supports autoencoder, unconditional and conditional diffusion, inpainting, diffusion autoencoder and LM model types. A basic Gradio interface can load pretrained checkpoints such as stable-audio-open-1.0 from Hugging Face.
Who it is for: Researchers and engineers training or fine-tuning audio generation models
Strengths
- Training, fine-tuning and inference in one repo, driven by JSON configs
- Multi-GPU and multi-node training via PyTorch Lightning, with DeepSpeed ZeRO Stage 2 option
- Datasets from local audio directories or WebDataset on Amazon S3
- Unwrap script strips training state to produce smaller inference checkpoints
Weaknesses
- Training logs to Weights & Biases, which needs an account and login
- Gradio UI is described as basic, meant for testing models
- Stable Audio Open weights require accepting terms on Hugging Face
- Needs PyTorch 2.5+; Flash Attention must be installed separately
What it needs
- Needs PyTorch 2.5+, PyTorch Lightning, Weights & Biases (training), Gradio (UI), Flash Attention (recommended), uv
- Models: stabilityai/stable-audio-open-1.0, autoencoder, diffusion_uncond, diffusion_cond, diffusion_cond_inpaint
Also in Training and fine-tuning
See all 68| Rank | Project | Score |
|---|---|---|
| 1 | peftPython library for LoRA and other parameter-efficient fine-tuning of Transformers models | 89 out of 100 |
| 2 | unslothDesktop and web app to run and fine-tune local models | 88 out of 100 |
| 3 | trlPost-training library with SFT, DPO, GRPO and KTO trainers | 88 out of 100 |
| #4 | deepspeedPyTorch library for distributed training of very large models with ZeRO | 87 out of 100 |
| #5 | megatron-lmDistributed training library for large transformer models on NVIDIA GPUs | 86 out of 100 |