Skip to content

vllm_omni.diffusion.models.ltx2.ltx2_phase_adapter ¶

Construction and execution of a fixed LTX refinement-phase adapter.

Checkpoint key parsing remains in :mod:ltx2_adapter_parser; this module resolves the declared artifact and makes the existing Transformer linears adapter-capable while retaining their quantization and tensor-parallel paths.

logger `module-attribute` ¶

logger = init_logger(__name__)

LTXPhaseAdapterRuntime ¶

Install and operate one fixed phase adapter on an existing transformer.

manifest `instance-attribute` ¶

manifest = manifest

transformer `instance-attribute` ¶

transformer = transformer

activate ¶

activate(adapter_slot: str | None) -> None

base_parameter_name ¶

base_parameter_name(name: str) -> str

finalize ¶

finalize() -> None

Stream rank-local A/B buffers after base weight processing completes.

build_ltx_phase_adapter ¶

build_ltx_phase_adapter(
    pipeline: Any,
) -> LTXPhaseAdapterRuntime | None

Build the fixed adapter required by a multi-phase LTX recipe.