stochasticai/x-stable-diffusion
A compilation of acceleration techniques and benchmarks for achieving sub-second latency Stable Diffusion inference using TensorRT, AITemplate, FlashAttention, and nvFuser.
Velocity · 7d
+0.4
★ / day
Trend
→steady
star history
This project compiles various optimization techniques for accelerating Stable Diffusion model inference, covering AITemplate, TensorRT, FlashAttention, and nvFuser. It includes comprehensive benchmarks on different GPUs (A100, T4) and sample images to compare quality vs speed tradeoffs. The repository provides a CLI tool called stochasticx for easy local deployment of optimized inference.