You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This issue acted as a PR tracker to Intel customer support related PRs. The purpose is to get understanding of what each PR does and how important are they compared to other customer support related PRs. This also help us to aware of merged PRs and PRs progress.
This issue acted as a PR tracker to Intel customer support related PRs. The purpose is to get understanding of what each PR does and how important are they compared to other customer support related PRs. This also help us to aware of merged PRs and PRs progress.
Under review
Already merged
o MoE
o Ulysess
o AutoTP
o Accelerator Graph
o ZeRO
o Others
non_reentrant_checkpoint
fix requires_grad of input must be true for activation checkpoint layer in pipeline train. #4224deepspeed.comm
instead oftorch.distributed
#5225torch.nan_to_num
replace numpy wrapper one #5877The text was updated successfully, but these errors were encountered: