Xinydev updated the memory management metrics documentation for the pinned host allocator in the pytorch/pytorch repository, focusing on improving observability and developer clarity. Working primarily in Python and leveraging expertise in CUDA and memory management, Xinydev aligned the documentation with the current implementation by clarifying the semantics of allocation counters and their relationship to memory events. The update addressed ambiguities in telemetry by detailing fields such as allocations, allocated bytes, and active requests, and included explicit notes on rounding and peak metrics. This work enhanced the accuracy and usability of memory management telemetry without introducing any runtime changes to the codebase.