SageAttention3: Microscaling FP4 Attention for Inference and An Exploration of 8-Bit Training

The efficiency of attention is important due to its quadratic time complexity. We enhance the efficiency of attention through two key contributions: First, we leverage the new $\text{FP4}$ Tensor Cores in Blackwell GPUs to accelerate attention computation. Our implementation achieves $\text{1038}$ $\text{TOPS}$ on $\text{RTX5090}$, which is a $\text{5}\times$ speedup over the fastest FlashAttention on $\text{RTX5090}$. Experiments show that our $\text{FP4}$ attention can accelerate inference of various models in a plug-and-play way.

Second, we pioneer low-bit attention to training tasks. Existing low-bit attention works like FlashAttention3 and SageAttention focus only on inference. However, the efficiency of training large models is also important. To explore whether low-bit attention can be effectively applied to training tasks, we design an accurate and efficient $\text{8-bit}$ attention for both forward and backward propagation. Experiments indicate that $\text{8-bit}$ attention achieves lossless performance in fine-tuning tasks but exhibits slower convergence in pretraining tasks.

The code is available at https://github.com/thu-ml/SageAttention.