[Diffusion] Add loader-owned host-weight plans for DLO
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
This commit is contained in:
1 parent
d1e230c95b
commit
e2796b6168
17 files changed
+1717
-588
No files matched your search
@@ -540,13 +540,6 @@ class MiniMaxH3Pipeline(
|
||||
encoder_block_attrs={"text_encoder": ("vision.blocks", "text_model.layers")},
|
||||
on_demand_component_paths=frozenset({"text_encoder", "video_vae", "audio_vae"}),
|
||||
)
|
||||
# H3's regular loader performs checkpoint-layout conversions (grouped QKV
|
||||
# and fused MLP). Keep it on that path until mmap can run the same loader
|
||||
# callbacks for every affected parameter.
|
||||
_supports_mmap_loading: ClassVar[bool] = False
|
||||
# TODO(offload): Re-enable after the generic rank-local mmap path can run
|
||||
# this model's grouped-QKV reorder and fused-MLP packing before TP sharding.
|
||||
# Do not bypass the regular loader until that equivalence is tested.
|
||||
_PROFILER_TARGETS: ClassVar[list[str]] = [
|
||||
"_prepare_reference_videos",
|
||||
"encode_prompt",
|
||||
|
||||
Reference in new issue
Block a user