All issues

Search results

Is it faster than flash_attn

Status: Open.

#34 In lucidrains/native-sparse-attention-pytorch;

· poryfly opened on Oct 22, 2025

Speed benchmarks

Status: Open.

#30 In lucidrains/native-sparse-attention-pytorch;

· isayoften opened on Jun 9, 2025

why set USE_SPARSE_ATTN = False USE_TRITON_NSA = False is faster ;with NSA the speed is 2.44it/s，without NSA the speed is 6.00it/s

Status: Open.

#29 In lucidrains/native-sparse-attention-pytorch;

· zhaoyingjun opened on May 27, 2025

How can we replace 'attend‘ function when compressing KV cache with flash-attn impls?

Status: Open.

#28 In lucidrains/native-sparse-attention-pytorch;

· wxthu opened on May 16, 2025

When calling the native_sparse_attend() function on an GTX 3090 GPU, it runs normally when dim_head=64, but when dim_head=128, the following error occurs:

Status: Open.

#26 In lucidrains/native-sparse-attention-pytorch;

· yyjFish opened on Mar 26, 2025

Can this api support V100 GPU?

Status: Open.

#11 In lucidrains/native-sparse-attention-pytorch;

· epiphanyer opened on Mar 5, 2025

AttributeError: 'NoneType' object has no attribute 'start'

Status: Open.

#10 In lucidrains/native-sparse-attention-pytorch;

· Gavin-jay opened on Mar 3, 2025

how to install triton-nightly

Status: Open.

#9 In lucidrains/native-sparse-attention-pytorch;

· flomerit opened on Feb 28, 2025

How can we conduct tests by integrating native-sparse-attention-pytorch like sglang or vllm ?

Status: Open.

#7 In lucidrains/native-sparse-attention-pytorch;

· BigCousin-z opened on Feb 26, 2025

请教block selection处的理解

Status: Open.

#3 In lucidrains/native-sparse-attention-pytorch;

· cafeii opened on Feb 20, 2025

todo

Status: Open.

#1 In lucidrains/native-sparse-attention-pytorch;

· lucidrains opened on Feb 19, 2025