ConceptAI & ML
Gradient accumulation steps in DDP
The number of gradient accumulation steps adjusted to account for the total batch size, micro-batch size, sequence length, and world size.
Know first
Named as a prerequisite by the lectures that define this. The number is how many say so. A grey one has no page yet.
DDP rank1Gradient accumulation1
Lecturers
1Connections
2The lectures that define it or assume it, and the concepts next to it.