Releases: lucidrains/native-sparse-attention-pytorch
Releases · lucidrains/native-sparse-attention-pytorch
0.1.9
more guards
0.1.8
allow for fine selection block sizes greater than 16, finishing the t…
0.1.7
more guards
0.1.6
some guards
0.1.5
fix some issues, validate it works on a100
0.1.4
keep chipping away at it
0.1.2
fix no selected kv during inference
0.1.1
allow for different compress to fine block sizes during inference, an…
0.1.0
fix a tricky bug with lse, lse is rounded to 128, but padding needs t…
0.0.78
fix flex sliding window attn