Add type annotation UnetWrapperFunction

This commit is contained in:
huchenlei 2024-05-20 15:02:36 -04:00
parent 1900e5119f
commit 58b9cc0502

View File

@ -3,10 +3,36 @@ import copy
import inspect
import logging
import uuid
from typing import Callable, Protocol, TypedDict, Optional
import comfy.utils
import comfy.model_management
class UnetApplyFunction(Protocol):
"""Function signature protocol on comfy.model_base.BaseModel.apply_model"""
def __call__(self, x: torch.Tensor, t: torch.Tensor, **kwargs) -> torch.Tensor:
pass
class UnetApplyConds(TypedDict):
"""Optional conditions for unet apply function."""
c_concat: Optional[torch.Tensor]
c_crossattn: Optional[torch.Tensor]
control: Optional[torch.Tensor]
transformer_options: Optional[dict]
class UnetParams(TypedDict):
input: torch.Tensor
timestep: torch.Tensor
c: UnetApplyConds
cond_or_uncond: torch.Tensor
UnetWrapperFunction = Callable[[UnetApplyFunction, UnetParams], torch.Tensor]
def apply_weight_decompose(dora_scale, weight):
weight_norm = (
weight.transpose(0, 1)
@ -117,7 +143,7 @@ class ModelPatcher:
if disable_cfg1_optimization:
self.model_options["disable_cfg1_optimization"] = True
def set_model_unet_function_wrapper(self, unet_wrapper_function):
def set_model_unet_function_wrapper(self, unet_wrapper_function: UnetWrapperFunction):
self.model_options["model_function_wrapper"] = unet_wrapper_function
def set_model_denoise_mask_function(self, denoise_mask_function):