[Perf]: CFG parallel abstraction (#851)

Signed-off-by: Didan Deng <33117903+wtomin@users.noreply.github.com>
Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>
This commit is contained in:
Didan DengandHongsheng Liu authored and GitHub committed 2026-01-30 16:23:44 +00:00
1 parent 8ad39570b8
commit 23cf48d7bb
31 files changed
+1506 -814

No files matched your search

+6 -1
View File
@@ -225,7 +225,12 @@ class OmniServeCommand(CLISubcommand):
help="Scheduler flow_shift for video models (e.g., 5.0 for 720p, 12.0 for 480p).",
)
omni_config_group.add_argument(
"--cfg-parallel-size", type=int, default=1, help="Number of GPUs for CFG parallel computation"
"--cfg-parallel-size",
type=int,
default=1,
choices=[1, 2],
help="Number of devices for CFG parallel computation for diffusion models. "
"Equivalent to setting DiffusionParallelConfig.cfg_parallel_size.",
)
return serve_parser