bghira/SimpleTuner

Based on the README, here's a technical summary: --- Supports unified fine-tuning across image, video, and audio diffusion models via LoRA, LyCORIS, and full-rank training with advanced memory optimizations including gradient checkpointing, parameter/optimizer state sharding (DeepSpeed, FSDP2), and context parallel attention. Provides embedding caching to disk, aspect bucketing for variable resolutions, and concept sliders for slider-based LoRA targeting with positive/negative sampling. Integrates with Hugging Face Accelerate and supports distributed S3 training, quantization (int8/fp8/nf4), and flow matching architectures across 20+ model architectures from SDXL to Flux to video models. ---

2,782 stars and 3,236 monthly downloads. Actively maintained with 41 commits in the last 30 days. Available on PyPI.