[Perf][Diffusion] Optimize HSDP startup via Rank-0 shared weight loading and accelerated LoRA delta computation (#7005)

Signed-off-by: samithuang <285365963@qq.com>
This commit is contained in:
Samit authored and GitHub committed 2026-09-11 08:28:42 +00:00
1 parent bad50980c9
commit 96360ccb00
13 files changed
+637 -16

No files matched your search

+7
View File
@@ -726,6 +726,13 @@ class OmniServeCommand(CLISubcommand):
default=True,
help="Disable multi-threaded safetensors loading (default: enabled with 4 threads).",
)
omni_config_group.add_argument(
"--enable-broadcast-weight-load",
action="store_true",
dest="enable_broadcast_weight_load",
default=False,
help="Enable Rank-0 shared weight broadcast across workers for HSDP (default: disabled).",
)
omni_config_group.add_argument(
"--num-weight-load-threads",
type=int,