Skip to content

ds-sequence-parallel(ulysses) for rope.#392

Open
inkcherry wants to merge 3 commits into
deepspeedai:mainfrom
inkcherry:rope-ds-sp
Open

ds-sequence-parallel(ulysses) for rope.#392
inkcherry wants to merge 3 commits into
deepspeedai:mainfrom
inkcherry:rope-ds-sp

Conversation

@inkcherry

@inkcherry inkcherry commented May 30, 2024

Copy link
Copy Markdown

get & calculate rope by using ds_sequence_parallelism

with the following test configurations:

  • ds_sp=1, gpus=1, apply rope
  • ds_sp=2, gpus=2, apply rope

When performing head parallelism between 2 all-to-all, I dumped the Q, K, and V values and obtained identical results.
dump code can be found at eedfe9c

Comment thread megatron/model/language_model.py Outdated
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants