Hongsheng Liu f1abb13958 [Diffusion] Register runtime cache for direct DLO H2D ...

Avoid recurrent mmap-to-pinned packing in no-AllGather DLO by registering transform-complete runtime-cache mappings under an explicit per-worker budget. Copy mapped tensors directly into the rotating device buffers while preserving bounded-staging fallback and ordered shutdown cleanup.

Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>

2026-08-17 11:50:32 +08:00

..

cpu_offload.md

[Docs] Split diffusion attention and CPU offload guides (#6075)

2026-08-15 14:41:09 +08:00

distributed_layerwise_offload.md

[Diffusion] Register runtime cache for direct DLO H2D

2026-08-17 11:50:32 +08:00

layerwise_offload.md

[Docs] Split diffusion attention and CPU offload guides (#6075)

2026-08-15 14:41:09 +08:00