1.7.1

Full Changelog: 1.6.3...1.7.1

1.6.3

add learned residual scaling for main transformer path

1.6.2

update for tweaks to laser

1.6.1

laser attention appears to work well

1.6.0

Full Changelog: 1.5.1...1.6.0

1.5.2

it does not work

1.5.1

restrict direction loss to random span masked region

1.5.0

add the direction loss for flow matching, claimed to accelerate train…

1.4.1

fix an issue with g2p and pydantic, and also add value residual learn…

1.4.0

fix an issue with g2p and pydantic, and also add value residual learn…