vllm_omni.diffusion.config ¶

Global diffusion config — mirrors vLLM's set_current_vllm_config pattern.

Set once during model initialisation so that every Attention layer can read the OmniDiffusionConfig without coupling __init__ to the runtime ForwardContext.

get_current_diffusion_config ¶

get_current_diffusion_config() -> OmniDiffusionConfig

Return the current diffusion config or raise.

get_current_diffusion_config_or_none ¶

get_current_diffusion_config_or_none() -> (
    OmniDiffusionConfig | None
)

Return the current diffusion config, or None if not set.

set_current_diffusion_config ¶

set_current_diffusion_config(config: OmniDiffusionConfig)

Context manager that sets the global diffusion config for model init.