Frank Lee eda30a058e [compatibility] fixed tensor parallel compatibility with torch 1.9 (#700)

2022-04-11 13:44:50 +08:00

..

colossalai_layer

[hotfix] Raise messages for indivisible batch sizes with tensor parallelism (#622)

2022-04-02 16:12:04 +08:00

moe

[zero] improve adaptability for not-shard parameters (#708)

2022-04-11 13:38:51 +08:00

parallel_1d

[model checkpoint] updated saving/loading for 1d layers (#594)

2022-04-01 16:51:52 +08:00

parallel_2d

[compatibility] fixed tensor parallel compatibility with torch 1.9 (#700)

2022-04-11 13:44:50 +08:00

parallel_2p5d

[compatibility] fixed tensor parallel compatibility with torch 1.9 (#700)

2022-04-11 13:44:50 +08:00

parallel_3d

[]Corrected 3d vocab parallel embedding (#707)

2022-04-11 10:17:55 +08:00

parallel_sequence

Refactored docstring to google style

2022-03-29 17:17:47 +08:00

utils

Refactored docstring to google style

2022-03-29 17:17:47 +08:00

vanilla

Refactored docstring to google style

2022-03-29 17:17:47 +08:00

wrapper

Refactored docstring to google style

2022-03-29 17:17:47 +08:00

__init__.py

[MOE] changed parallelmode to dist process group (#460)

2022-03-19 13:46:29 +08:00

base_layer.py

[model checkpoint] reworked unified layers for ease of save/load states (#593)

2022-04-01 16:49:56 +08:00