Commit graph

8 commits

Author SHA1 Message Date
Zangwei Zheng
52aeb3769f format and some fix 2024-03-30 12:02:48 +08:00
Hongxin Liu
8d5278b99a [feature] impl cached pos embedding (#1)
* [feature] impl cached pos embedding

* [feature] update pos emb

* [feature] update pos emb
2024-03-28 17:30:47 +08:00
Frank Lee
14526479fc added sp for inference (#76) 2024-03-17 11:00:23 +08:00
Zheng Zangwei (Alex Zheng)
d851a85535 format (#69) 2024-03-15 22:16:20 +08:00
Frank Lee
c4c5d64e49 migrate some new files (#64) 2024-03-15 21:49:38 +08:00
Frank Lee
2f87a9af75
removed old files (#63) 2024-03-15 21:48:36 +08:00
Hongxin Liu
91275b2b5e
[feature] impl fastseq-style seq parallel (#21)
* [feature] add fastseq-style sp attn

* [feature] add overlap fastseq

* [test] add test for self attn

* [feature] update dit model to fit fastseq

* [polish] refactor attn

* [feature] update train & benchmark script

* [polish] update benchmark script

* [polish] update benchmark script
2024-03-01 17:31:59 +08:00
Hongxin Liu
97c089daec
[feature] impl ulysses-style seq parallel (#20)
* [feature] add ulysses style sp attn

* [test] add sp attn test

* [feature] add zero sp plugin

* [hotfix] fix sp backward

* [test] add test for dit model
2024-03-01 14:42:06 +08:00