Skip to content

vllm_omni.diffusion.models.ltx2.ops.swiglu ¶

Bit-exact workspace-reusing SwiGLU for the LTX diffusion VAE decoder.

logger `module-attribute` ¶

logger = logging.getLogger(__name__)

try_swiglu_tiled_exact ¶

try_swiglu_tiled_exact(
    hidden_states: Tensor,
    gate_weight: Tensor,
    up_weight: Tensor,
    down_weight: Tensor,
) -> Tensor | None

Run workspace-reusing tiled SwiGLU, or expose the original module.