Skip to content

Fix replicated K/V all-reduce to run once per accumulation window - #9064

Draft
VenusTZZ wants to merge 6 commits into
deepspeedai:masterfrom
VenusTZZ:fix-gptbigcode-autotp-defer-kv
Draft

VenusTZZ wants to merge 6 commits into
deepspeedai:masterfrom
VenusTZZ:fix-gptbigcode-autotp-defer-kv