# Changelog

## [0.3.4] - 2026-08-23

### Added

- Add giant model summary command [gitea #46](https://git.larsbogner.de/lars/giant/issues/46)

- Add per-stage init_from/freeze [gitea #42](https://git.larsbogner.de/lars/giant/issues/42)

- Add bf16 autocast to the training loop [gitea #47](https://git.larsbogner.de/lars/giant/issues/47)

- Add class-balanced secondary particle-type loss [gitea #44](https://git.larsbogner.de/lars/giant/issues/44)


### Changed

- Implement stage2_model.stage1_context = "sampled" [gitea #41](https://git.larsbogner.de/lars/giant/issues/41)

- Bump patch version to 0.3.3

- Offset event_id across multi-shard reference reads in giant analyze [gitea #22](https://git.larsbogner.de/lars/giant/issues/22)

- Auto-bump patch version, tag, and update changelog on merge to master [gitea #50](https://git.larsbogner.de/lars/giant/issues/50)

- Document CI_TOKEN's write:repository scope requirement [gitea #50](https://git.larsbogner.de/lars/giant/issues/50)

# Changelog