vllm_omni.diffusion.vllm_config ¶

Build native vLLM configuration views for the Omni diffusion runtime.

configure_diffusion_vllm_config ¶

configure_diffusion_vllm_config(
    vllm_config: VllmConfig, od_config: OmniDiffusionConfig
) -> VllmConfig

Populate native model, parallel, cache, and Scheduler configuration.

create_base_diffusion_vllm_config(
    device: device, od_config: OmniDiffusionConfig
) -> VllmConfig

Create the native vLLM 0.27 config used by diffusion.

create_diffusion_vllm_config(
    device: device, od_config: OmniDiffusionConfig
) -> VllmConfig

Build the same fully populated native config in Engine and Workers.

resolve_diffusion_max_model_len(
    od_config: OmniDiffusionConfig,
) -> int

Resolve the native KV admission bound for a diffusion stage.