v0.2.5
[embedding] rename FreqAwareEmbedding -> CachedEmbedding (#1699)
[zero] add zero wrappers (#2523)
[gemini] add fake_release_chunk for keep-gathered chunk in the inference mode (#2671)
[hotfix] fix lightning error (#2529)
[ddp] ColoDDP uses bucket all-reduce (#1177)
[gemini] update the gpt example (#2527)