0.2.3

What's Changed

• [Bugfix] add mask for offs_n when storing dk & dv by @StrongSpoon in #33

New Contributors

• @StrongSpoon made their first contribution in #33

Full Changelog: 0.2.2...0.2.3

0.2.2

address https://github.com/lucidrains/native-sparse-attention-pytorch…

0.2.1

fix maximum tracking in triton

0.2.0

What's Changed

• compress_block_sliding_stride by @Mr-Grin in #25

Full Changelog: 0.1.27...0.2.0

0.1.27

What's Changed

• Small change so token embeddings aren't looked up for past tokens during inference by @Pasewark in #24

Full Changelog: 0.1.26...0.1.27

0.1.26

Full Changelog: 0.1.25...0.1.26

0.1.25

make it right for selection block sizes > 16

0.1.24

What's Changed

• Added CompressTransformer, a simple transformer to compress the blocks by @Pasewark in #21

New Contributors

• @Pasewark made their first contribution in #21

Full Changelog: 0.1.23...0.1.24

0.1.23

Full Changelog: 01.22...0.1.23

0.1.21

seq parallel for backwards nsa