laksjdjf

import gradio as gr
import json
import requests
import argparse
from dataclasses import dataclass

############### utils ###############
BAN_TOKENS = ["<|END_OF_TURN_TOKEN|>"] # command -r 用の回避トークン

parser = argparse.ArgumentParser()

laksjdjf

from comfy.samplers import KSAMPLER
import torch
from comfy.k_diffusion.sampling import default_noise_sampler, to_d
from tqdm.auto import trange

@torch.no_grad()
def sampler_tcd(model, x, sigmas, extra_args=None, callback=None, disable=None, noise_sampler=None, gamma=None):
    extra_args = {} if extra_args is None else extra_args
    noise_sampler = default_noise_sampler(x) if noise_sampler is None else noise_sampler
    s_in = x.new_ones([x.shape[0]])

laksjdjf

# https://github.com/huggingface/transformers/blob/838b87abe231fd70be5132088d0dee72a7bb8d62/src/transformers/models/opt/modeling_opt.py#L147
"""
model = AutoModelForCausalLM.from_pretrained("p1atdev/dart-v1-sft")
apply_hook(model)
"""


import torch
import torch.nn as nn
def forward_hooker(self):

laksjdjf

# https://huggingface.co/shadowlilac/aesthetic-shadow-v2

from transformers import pipeline
import torch
from PIL import Image
from comfy.ldm.modules.attention import optimized_attention
    
def optimized_forward(self):
    def forward(hidden_states, head_mask = None, output_attentions = False):
        query = self.query(hidden_states)

laksjdjf

def make_unet_conversion_map():
    unet_conversion_map_layer = []
    # unet
    # https://github.com/kohya-ss/sd-scripts/blob/2d7389185c021bc527b414563c245c5489d6328a/library/sdxl_model_util.py#L293
    for i in range(3):  # num_blocks is 3 in sdxl
        # loop over downblocks/upblocks
        for j in range(2):
            # loop over resnets/attentions for downblocks
            hf_down_res_prefix = f"down_blocks.{i}.resnets.{j}."
            sd_down_res_prefix = f"input_blocks.{3*i + j + 1}.0."

laksjdjf

import torch

class VisualStylePrompting:
    @classmethod
    def INPUT_TYPES(s):
        return {"required": { 
            "model": ("MODEL",),
            "reference": ("LATENT",),
            "depth": ("INT", {"default": 0, "min": -1, "max": 12}),
            "batch_size": ("INT", {"default": 1, "min": 1, "max": 64}),

laksjdjf

# ref: ScaleCrafter https://github.com/YingqingHe/ScaleCrafter

import math
import comfy.ops
import torch.nn.functional as F
ops = comfy.ops.disable_weight_init

class ScaleCrafter:
    @classmethod
    def INPUT_TYPES(s):

laksjdjf

# https://arxiv.org/abs/2310.07702

import comfy.ops
ops = comfy.ops.disable_weight_init

class DilateConv:
    @classmethod
    def INPUT_TYPES(s):
        return {
            "required": {

laksjdjf

===========================================================================================================================================================
Layer (type (var_name))                                 Input Shape               Output Shape              Param #                   Kernel Shape
===========================================================================================================================================================
StableCascadeUnet (StableCascadeUnet)                   --                        [1, 4, 256, 256]          --                        3
├─Linear (clip_txt_pooled_mapper)                       [1, 1, 1280]              [1, 1, 5120]              6,558,720                 --
├─LayerNorm (clip_norm)                                 [1, 4, 1280]              [1, 4, 1280]              --                        --
├─Sequential (embedding)                                [1, 4, 256, 256]          [1, 320, 128, 128]        --                        --
│    └─PixelUnshuf

laksjdjf

===========================================================================================================================================================
Layer (type (var_name))                                 Input Shape               Output Shape              Param #                   Kernel Shape
===========================================================================================================================================================
StableCascadeUnet (StableCascadeUnet)                   [2, 16, 24, 24]           [2, 16, 24, 24]           8,923,136                 3
├─Linear (clip_txt_pooled_mapper)                       [2, 77, 1280]             [2, 77, 8192]             10,493,952                --
├─LayerNorm (clip_norm)                                 [2, 308, 2048]            [2, 308, 2048]            --                        --
├─Sequential (embedding)                                [2, 16, 24, 24]           [2, 2048, 24, 24]         --                        --
│    └─PixelUnshuf