Module tp

Module tp 

Source
Expand description

K3 tensor-parallel storage plan, shared by host loading and device binding. Config head counts are rank-local; expert and dense widths remain full.

Structs§

TpBytePlan
Byte ranges from one full checkpoint tensor that form this rank’s tensor. Ranges preserve row-major packed nibbles and their matching E8M0 groups.

Enums§

TpAxis

Functions§

plan_tensor_bytes
Plan checkpoint storage BEFORE allocating a device buffer.
tensor_plan