NVIDIA NeMo Automodel enables distributed diffusion model fine-tuning with Hugging Face Diffusers

NVIDIA and Hugging Face have integrated NVIDIA NeMo Automodel with the 🤗 Diffusers library to provide production-grade, distributed training for open-source diffusion models. This collaboration allows users to fine-tune any Diffusers-format model on the Hugging Face Hub without requiring checkpoint conversion or model code rewrites.

• The integration supports full fine-tuning and LoRA-style parameter-efficient fine-tuning (PEFT) for flow-matching models like FLUX.1-dev and Wan 2.1.
• NeMo Automodel utilizes PyTorch DTensor-native parallelism, including FSDP2, tensor, context, and pipeline parallelisms, to scale training from one GPU to hundreds.
• The system employs latent-space training with pre-encoded VAE outputs and multiresolution bucketed dataloaders to accelerate throughput.
• Pretrained weights load directly into DiffusionPipeline for inference, ensuring downstream tools like quantization and custom samplers remain compatible.

This approach enables researchers and builders to scale diffusion model adaptation efficiently while maintaining compatibility with the existing Hugging Face ecosystem.