Commit Graph
14 Commits
Author SHA1 Message Date
Hongsheng Liu 16642dc2e8 [Diffusion] Make DLO host weight cache opt-in
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
2026-08-19 11:23:36 +08:00
Hongsheng Liu 010b3c1d74 [Diffusion] Make no-AllGather host weight cache automatic
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
2026-08-19 11:23:36 +08:00
Hongsheng Liu 3bb7547a0f [Diffusion] Rename runtime cache to host weight cache
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
2026-08-19 11:23:36 +08:00
Hongsheng Liu 1ccdc2e80e [Docs] Separate DLO transfer mode design
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
2026-08-19 11:23:16 +08:00
Hongsheng Liu 8f93a74951 [Docs] Remove stale DLO validation results
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
2026-08-19 11:22:52 +08:00
Hongsheng Liu 4acb3b1387 [Docs] Add DLO component architecture
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
2026-08-19 11:21:59 +08:00
Hongsheng Liu 6751c1f404 [Diffusion] Address runtime cache review feedback
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
2026-08-19 11:21:59 +08:00
Hongsheng Liu 23ab14a15b [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>
2026-08-19 11:21:28 +08:00
Hongsheng Liu 67f15e73cd [Diffusion] Document runtime cache hardware tradeoffs
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
2026-08-19 11:21:28 +08:00
Hongsheng Liu 59549968cf [Diffusion] Tighten runtime cache compatibility guards
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
2026-08-19 11:21:28 +08:00
Hongsheng Liu d30ccdc315 [Diffusion] Add node-local runtime weight cache for DLO
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
2026-08-19 11:21:01 +08:00
Hongsheng Liu 284e05c88b [Core] Support online FP8 with DLO AllGather (#6279)
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
2026-08-18 23:48:07 +08:00
Hongsheng Liu 9d2bb23ff6 [Diffusion] Add loader-owned host-weight plans for DLO (TP=1) (#6213)
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
2026-08-16 11:02:38 +08:00
Hongsheng Liu 1260ceaae6 [Docs] Split diffusion attention and CPU offload guides (#6075)
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
2026-08-15 14:41:09 +08:00