0.0.34

oops

0.0.33

make the fine flex block mask also aware of gqa

0.0.31

oops, think they pair up the query heads to kv heads in gqa differently

0.0.30

use enable_gqa for flex attention for the sliding windows branch

0.0.29

wire up flex fine selected attention and make sure it runs

0.0.28

Full Changelog: 0.0.27...0.0.28

0.0.27

Full Changelog: 0.0.26...0.0.27

0.0.26

Full Changelog: 0.0.24...0.0.26

0.0.25

last commit for the day, should be ready for experiments tomorrow

0.0.24

complete the fine attention masking with flex attention, not wired up