Skip to content

vllm_omni.diffusion.model_loader.host_weights.producers

Diffusion Host Weight Runtime producers.

Modules:

Name Description
final_layout_bf16

Concrete BF16-with-preserved-FP32 final-layout artifact policy and producer.

FINAL_LAYOUT_BF16_MANIFEST_SCHEMA module-attribute

FINAL_LAYOUT_BF16_MANIFEST_SCHEMA = (
    "diffusion-final-layout-bf16-manifest-v1"
)

FINAL_LAYOUT_BF16_POLICY module-attribute

FINAL_LAYOUT_BF16_POLICY = FinalLayoutBF16Policy()

FINAL_LAYOUT_BF16_PRODUCER_ID module-attribute

FINAL_LAYOUT_BF16_PRODUCER_ID = (
    "vllm-omni.diffusion.final-layout-bf16"
)

FINAL_LAYOUT_BF16_REPRESENTATION module-attribute

FINAL_LAYOUT_BF16_REPRESENTATION = (
    "diffusion-final-layout-bf16"
)

FINAL_LAYOUT_BF16_SPEC module-attribute

FINAL_LAYOUT_BF16_SPEC = FinalLayoutArtifactSpec(
    representation=WeightRepresentation(
        name=FINAL_LAYOUT_BF16_REPRESENTATION,
        dtype=str(torch.bfloat16),
        metadata=CanonicalJson.from_value(
            {
                "format": FINAL_LAYOUT_BF16_REPRESENTATION,
                "mixed_precision_policy": "bf16-with-preserved-fp32",
            }
        ),
    ),
    producer=ProducerIdentity(
        producer_id=FINAL_LAYOUT_BF16_PRODUCER_ID,
        version=FINAL_LAYOUT_BF16_VERSION,
        implementation_fingerprint=implementation_abi_fingerprint(
            _BF16_IMPLEMENTATION_ABI
        ),
        manifest_schema=FINAL_LAYOUT_BF16_MANIFEST_SCHEMA,
        restorer_schema=FINAL_LAYOUT_TENSOR_RESTORER_SCHEMA,
    ),
    implementation_abi=_BF16_IMPLEMENTATION_ABI,
)

FINAL_LAYOUT_BF16_VERSION module-attribute

FINAL_LAYOUT_BF16_VERSION = '1'

FinalLayoutBF16Policy

Concrete compatibility and metadata policy for finalized BF16 tensors.

spec property

build_format_metadata

build_format_metadata(
    *,
    component_names: tuple[str, ...],
    tensor_contract_digest: str,
    tensor_count: int,
) -> CanonicalJson

tensor_role

tensor_role(
    name: str, tensor: Tensor, kind: TensorKind
) -> str

validate_collection

validate_collection(
    targets: Sequence[RuntimeTensorTarget],
) -> None

validate_format_metadata

validate_format_metadata(
    metadata: CanonicalJson,
    *,
    component_names: tuple[str, ...],
    tensor_contract_digest: str,
    tensor_count: int,
) -> None

validate_request

validate_request(request: FinalLayoutRequest) -> None

validate_target

validate_target(target: RuntimeTensorTarget) -> None

FinalLayoutBF16Producer

Publish finalized BF16-policy tensors through a store-scoped writer.

spec property

produce

produce(writer: ArtifactWriter) -> ProductionMetadata