takuma104

import torch
import sys
from safetensors.torch import load_file

if __name__ == "__main__":
    filename = sys.argv[1]
    state_dict = load_file(filename)
    for key, value in state_dict.items():
        if "lora_down" in key:
            lora_name = key.split(".")[0]

takuma104

import math

import numpy as np
import safetensors
import torch
import torch.nn as nn
from PIL import Image

from diffusers import DiffusionPipeline, DPMSolverMultistepScheduler, StableDiffusionPipeline
from diffusers.utils import _get_model_file, DIFFUSERS_CACHE

takuma104

import torch
import json
from diffusers import StableDiffusionPipeline, DPMSolverMultistepScheduler
from diffusers.models.attention import Attention
from diffusers.models.attention_processor import (
    AttnProcessor2_0,
    XFormersAttnProcessor,
    LoRAAttnProcessor2_0,
    LoRAXFormersAttnProcessor,
)

takuma104

takuma104

import json
import struct
import sys
from pprint import pprint

def parse_safetensors_header(file_path):
    # modified from https://huggingface.co/docs/safetensors/metadata_parsing
    with open(file_path, 'rb') as f:
        header_bytes = f.read(8)
        # Interpret the bytes as a little-endian unsigned 64-bit integer

takuma104

import torch
import json
from diffusers import StableDiffusionPipeline

def print_memory_usage(width, height, batch, xformers, with_lora):
    def on_off(cond):
        return 'ON' if cond else 'OFF'
    mem_bytes = torch.cuda.max_memory_allocated()
    mem_MB = int(mem_bytes/(10**6))
    dict = {'width':width, 'height':height, 'batch':batch, 'xformers':on_off(xformers), 

takuma104

import torch
from diffusers import StableDiffusionPipeline, DPMSolverMultistepScheduler

def create_pipeline():
    pipe = StableDiffusionPipeline.from_pretrained(
        "gsdf/Counterfeit-V2.5", torch_dtype=torch.float16, safety_checker=None
    ).to("cuda")
    pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config, use_karras_sigmas=True)
    pipe.enable_xformers_memory_efficient_attention()
    return pipe

takuma104

import torch
from diffusers import StableDiffusionPipeline, DPMSolverMultistepScheduler

def create_pipeline():
    pipe = StableDiffusionPipeline.from_pretrained(
        "gsdf/Counterfeit-V2.5", torch_dtype=torch.float16, safety_checker=None
    ).to("cuda")
    pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config, use_karras_sigmas=True)
    pipe.enable_xformers_memory_efficient_attention()
    return pipe

takuma104

import torch
import safetensors.torch
import sys

def dump_keys(parent, suffix=''):
    for k in sorted(parent.keys()):
        if isinstance(parent[k], torch.Tensor):
            print(f'{suffix}{k} {list(parent[k].shape)} mean={torch.mean(parent[k]):.3g} std={torch.std(parent[k]):.3g}')
        else:
            dump_keys(parent[k], f'{suffix}{k}.')

takuma104