We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
Hugging Face Blog·18d ago·5 min read
Fine-tune video and image models at scale with NVIDIA NeMo Automodel and 🤗 Diffusers
TL;DR
NVIDIA and Hugging Face have collaborated to create NeMo Automodel, an open-source library for fine-tuning diffusion models at scale. This tool allows seamless integration with models from the 🤗Diffusers library without requiring checkpoint conversions.
✦ Why It Matters
Engineers can implement NeMo Automodel today to enhance their diffusion model training workflows without extensive modifications.
Key Takeaways
- Utilize NeMo Automodel for seamless model fine-tuning.
- Access a variety of diffusion models directly from Hugging Face.
- Choose between full fine-tuning and LoRA for efficiency.
- Leverage scalable training configurations for large models.
How It Works
NeMo Automodel integrates with the 🤗 Diffusers library, allowing users to load and fine-tune models directly from the Hugging Face Hub. It employs flow-matching as the training objective and supports various parallelism configurations, enabling efficient training across multiple GPUs.
The library's design allows for easy adaptation to new models with minimal code changes, ensuring that users can quickly implement the latest diffusion models.