Skip to content

Clamp capacity to num_tokens in MoE gating drop branches (complete #5353) - #8155

Open
ebarkhordar wants to merge 3 commits into
deepspeedai:masterfrom
ebarkhordar:fix/topkgating-probs-capacity
Open

Clamp capacity to num_tokens in MoE gating drop branches (complete #5353)#8155
ebarkhordar wants to merge 3 commits into
deepspeedai:masterfrom
ebarkhordar:fix/topkgating-probs-capacity