What's Changed

• feat: Simpler assistant mask calculation (#56)
• Prompted models (#55)
• Remove default values in trajectories (#53)
• Allow logging of non-trainable models, log full trajectories (#52)
• More flexible TrainableModel (#51)

Full Changelog: v0.1.23...v0.1.24

What's Changed

No pull requests were merged in this release.

Full Changelog: v0.1.22...v0.1.23

What's Changed

• Small tweaks (#50)

Full Changelog: v0.1.21...v0.1.22

What's Changed

• feat: Prisoners dilemma example (#49)
• Update 2048 example to use 3B model with longer context on T4 (#48)
• refactor: Remove Unsloth fork dependencies (#47)

Full Changelog: v0.1.20...v0.1.21

What's Changed

• fix: Patch MultiStepModelRunner for Unsloth compatibility (#46)

Full Changelog: v0.1.19...v0.1.20

What's Changed

• refactor: Transition to new dev module structure + be able to train on longer sequences in memory constrained environments (#45)
• Small tweaks (#44)

Full Changelog: v0.1.18...v0.1.19

What's Changed

No pull requests were merged in this release.

Full Changelog: v0.1.17...v0.1.18

What's Changed

No pull requests were merged in this release.

Full Changelog: v0.1.16...v0.1.17

What's Changed

No pull requests were merged in this release.

Full Changelog: v0.1.15...v0.1.16

What's Changed

No pull requests were merged in this release.

Full Changelog: v0.1.14...v0.1.15