Search results

Sage Attention with WAN FP8 model (or FP8 quantization) causes black output. So I'm stuck either doing FP16 with Sage but maxing VRAM or using FP8 but getting no Sage speedup

Status: Open.

#221 In thu-ml/SageAttention;

· slikvik55 opened on Aug 1, 2025

sage attention error inference

Status: Open.

#220 In thu-ml/SageAttention;

· Passenger12138 opened on Aug 1, 2025

Install SageAttention 2.2.0 error: RuntimeError: CUDA 12.8 or higher is required for compute capability 12.0.

Status: Open.

#219 In thu-ml/SageAttention;

· driqeks opened on Jul 30, 2025

import sageattn met undefined symbol error

Status: Open.

#216 In thu-ml/SageAttention;

· motoight opened on Jul 23, 2025

Do you have plan to support VLLM 、sglang or other LLM framework?

Status: Open.

#214 In thu-ml/SageAttention;

· zhang22222 opened on Jul 21, 2025

Q Smoothing Mentioned in SageAttention2 Paper

Status: Open.

#213 In thu-ml/SageAttention;

· ColinZ22 opened on Jul 18, 2025

The detected CUDA version (12.0) mismatches the version that was used to compile PyTorch (11.8). Please make sure to use the same CUDA versions.

Status: Open.

#212 In thu-ml/SageAttention;

· gateway opened on Jul 15, 2025

sageattention-1 论文中的CogvideoX的测试

Status: Open.

#211 In thu-ml/SageAttention;

· Romaosir opened on Jul 15, 2025

SageAttentionV3

Status: Open.

#210 In thu-ml/SageAttention;

· Bezdarnost opened on Jul 9, 2025

Compiling from main branch results in 2.1.1 version?

Status: Open.

#209 In thu-ml/SageAttention;

· SamTyurenkov opened on Jul 9, 2025

How to use NVFP4 gemm

Status: Open.

#208 In thu-ml/SageAttention;

· nanzh-19 opened on Jul 9, 2025

sageattention2.2 can't be used on ubuntu20.04?

Status: Open.

#206 In thu-ml/SageAttention;

· fangg2000 opened on Jul 6, 2025