sample_old.py

"""
采样代码
文件输入:
    prompt, 指定的输入文件夹路径, 制定的输出文件夹路径
文件输出:
    采样的图片, 存放于指定输出文件夹路径
- 指定prompt文件夹位置, 选手需要自行指定模型的地址以及其他微调参数的加载方式, 进行图片的生成并保存到指定地址, 此部分代码选手可以修改。
- 输入文件夹的内容组织方式和训练代码的输出一致
- sample的方法可以修改
- 生成过程中prompt可以修改, 但是指标测评时会按照给定的prompt进行测评。
"""

import os
import ml_collections
import torch
import random
import argparse
import utils
from libs.dpm_solver_pp import NoiseScheduleVP, DPM_Solver
import einops
import libs.autoencoder
import libs.clip
from torchvision.utils import save_image, make_grid
import numpy as np
import clip
import time
from libs.clip import FrozenCLIPEmbedder
import numpy as np
import json
from libs.uvit_multi_post_ln_v1 import UViT
from peft import inject_adapter_in_model, LoraConfig,get_peft_model
lora_config = LoraConfig(
   inference_mode=False, r=128, lora_alpha=90, lora_dropout=0.1,target_modules=["qkv","fc1","fc2","proj","to_out","to_q","to_k","to_v","text_embed","clip_img_embed"]
)

def get_model_size(model):
    """
    统计模型参数大小
    """
    para = sum([np.prod(list(p.size())) for p in model.parameters()])
    type_size = 4
    print('Model {} : params: {:4f}M'.format(model._get_name(), para * type_size / 1000 / 1000))
    return para

def stable_diffusion_beta_schedule(linear_start=0.00085, linear_end=0.0120, n_timestep=1000):
    _betas = (
        torch.linspace(linear_start ** 0.5, linear_end ** 0.5, n_timestep, dtype=torch.float64) ** 2
    )
    return _betas.numpy()


def prepare_contexts(config, clip_text_model, autoencoder):

    device = 'cuda' if torch.cuda.is_available() else 'cpu'

    contexts = torch.randn(config.n_samples, 77, config.clip_text_dim).to(device)
    img_contexts = torch.randn(config.n_samples, 2 * config.z_shape[0], config.z_shape[1], config.z_shape[2])
    clip_imgs = torch.randn(config.n_samples, 1, config.clip_img_dim)

    prompts = [ config.prompt ] * config.n_samples
    contexts = clip_text_model.encode(prompts)

    return contexts, img_contexts, clip_imgs


def unpreprocess(v):  # to B C H W and [0, 1]
    v = 0.5 * (v + 1.)
    v.clamp_(0., 1.)
    return v


def set_seed(seed: int):
    random.seed(seed)
    np.random.seed(seed)
    torch.manual_seed(seed)
    torch.cuda.manual_seed_all(seed)


def sample(prompt_index, prompt, config, nnet, clip_text_model, autoencoder, device):
    """
    using_prompt: if use prompt as file name
    """
    n_iter = config.n_iter
    if config.get('benchmark', False):
        torch.backends.cudnn.benchmark = True
        torch.backends.cudnn.deterministic = False
    config = ml_collections.FrozenConfigDict(config)

    _betas = stable_diffusion_beta_schedule()
    N = len(_betas)


    use_caption_decoder = config.text_dim < config.clip_text_dim or config.mode != 't2i'
    if use_caption_decoder:
        from libs.caption_decoder import CaptionDecoder
        caption_decoder = CaptionDecoder(device=device, **config.caption_decoder)
    else:
        caption_decoder = None

    empty_context = clip_text_model.encode([''])[0]

    def split(x):
        C, H, W = config.z_shape
        z_dim = C * H * W
        z, clip_img = x.split([z_dim, config.clip_img_dim], dim=1)
        z = einops.rearrange(z, 'B (C H W) -> B C H W', C=C, H=H, W=W)
        clip_img = einops.rearrange(clip_img, 'B (L D) -> B L D', L=1, D=config.clip_img_dim)
        return z, clip_img


    def combine(z, clip_img):
        z = einops.rearrange(z, 'B C H W -> B (C H W)')
        clip_img = einops.rearrange(clip_img, 'B L D -> B (L D)')
        return torch.concat([z, clip_img], dim=-1)


    def t2i_nnet(x, timesteps, text):  # text is the low dimension version of the text clip embedding
        """
        1. calculate the conditional model output
        2. calculate unconditional model output
            config.sample.t2i_cfg_mode == 'empty_token': using the original cfg with the empty string
            config.sample.t2i_cfg_mode == 'true_uncond: using the unconditional model learned by our method
        3. return linear combination of conditional output and unconditional output
        
        'empty_token' 模式：在这种模式下，使用原始配置和空字符串来生成图像。这意味着生成的图像不受与之相关的文本信息的约束，生成的结果更加自由和多样化。

        'true_uncond' 模式：在这种模式下，使用通过我们方法学习到的无条件模型来生成图像。这意味着生成的图像不依赖于与之相关的文本信息，生成的结果更加无条件和独立。
        """
        z, clip_img = split(x)

        t_text = torch.zeros(timesteps.size(0), dtype=torch.int, device=device)

        z_out, clip_img_out, text_out = nnet(z, clip_img, text=text, t_img=timesteps, t_text=t_text,
                                             data_type=torch.zeros_like(t_text, device=device, dtype=torch.int) + config.data_type)
        x_out = combine(z_out, clip_img_out)

        if config.sample.scale == 0.:
            return x_out

        if config.sample.t2i_cfg_mode == 'empty_token':
            _empty_context = einops.repeat(empty_context, 'L D -> B L D', B=x.size(0))
            if use_caption_decoder:
                _empty_context = caption_decoder.encode_prefix(_empty_context)
            z_out_uncond, clip_img_out_uncond, text_out_uncond = nnet(z, clip_img, text=_empty_context, t_img=timesteps, t_text=t_text,
                                                                      data_type=torch.zeros_like(t_text, device=device, dtype=torch.int) + config.data_type)
            x_out_uncond = combine(z_out_uncond, clip_img_out_uncond)
        elif config.sample.t2i_cfg_mode == 'true_uncond':
            text_N = torch.randn_like(text)  # 3 other possible choices
            z_out_uncond, clip_img_out_uncond, text_out_uncond = nnet(z, clip_img, text=text_N, t_img=timesteps, t_text=torch.ones_like(timesteps) * N,
                                                                      data_type=torch.zeros_like(t_text, device=device, dtype=torch.int) + config.data_type)
            x_out_uncond = combine(z_out_uncond, clip_img_out_uncond)
        else:
            raise NotImplementedError

        return x_out + config.sample.scale * (x_out - x_out_uncond)



    @torch.cuda.amp.autocast()
    def encode(_batch):
        return autoencoder.encode(_batch)

    @torch.cuda.amp.autocast()
    def decode(_batch):
        return autoencoder.decode(_batch)


    contexts, img_contexts, clip_imgs = prepare_contexts(config, clip_text_model, autoencoder)
    contexts_low_dim = contexts if not use_caption_decoder else caption_decoder.encode_prefix(contexts)  # the low dimensional version of the contexts, which is the input to the nnet

    _n_samples = contexts_low_dim.size(0)


    def sample_fn(**kwargs):
        _z_init = torch.randn(_n_samples, *config.z_shape, device=device)
        _clip_img_init = torch.randn(_n_samples, 1, config.clip_img_dim, device=device)
        _x_init = combine(_z_init, _clip_img_init)

        noise_schedule = NoiseScheduleVP(schedule='discrete', betas=torch.tensor(_betas, device=device).float())

        def model_fn(x, t_continuous):
            t = t_continuous * N
            return t2i_nnet(x, t, **kwargs)

        dpm_solver = DPM_Solver(model_fn, noise_schedule, predict_x0=True, thresholding=False)
        with torch.no_grad(), torch.autocast(device_type="cuda" if "cuda" in str(device) else "cpu"):
            start_time = time.time()
            x = dpm_solver.sample(_x_init, steps=config.sample.sample_steps, eps=1. / N, T=1.)
            end_time = time.time()
            print(f'\ngenerate {_n_samples} samples with {config.sample.sample_steps} steps takes {end_time - start_time:.2f}s')

        _z, _clip_img = split(x)
        return _z, _clip_img


    samples = None    
    for i in range(n_iter):
        _z, _clip_img = sample_fn(text=contexts_low_dim)  # conditioned on the text embedding
        new_samples = unpreprocess(decode(_z))
        if samples is None:
            samples = new_samples
        else:
            samples = torch.vstack((samples, new_samples))

    os.makedirs(config.output_path, exist_ok=True)
    for idx, sample in enumerate(samples):
        save_path = os.path.join(config.output_path, f'{prompt_index}-{prompt}-{idx:03}.jpg')
        save_image(sample, save_path)
        

    print(f'\nGPU memory usage: {torch.cuda.max_memory_reserved() / 1024 ** 3:.2f} GB')
    print(f'\nresults are saved in {os.path.join(config.output_path)} :)')

def compare_and_print_models(standard_model, model, mapping_dict={}):
    total = 0
    origin_dict = dict(standard_model.named_parameters())
    compare_dict = dict(model.named_parameters())
    device = "cuda"
    diff_parameters = {}

    # Compare parameters
    for param_name, origin_param in origin_dict.items():
        compare_param_name = mapping_dict.get(param_name, param_name)
        if compare_param_name not in compare_dict:
            continue

        origin_p = origin_param.data.to(device)
        compare_p = compare_dict[compare_param_name].data.to(device)

        if origin_p.shape != compare_p.shape or (origin_p - compare_p).norm() != 0:
            model_part = "origin" if param_name not in mapping_dict else "compare"
            diff_parameters[param_name] = {
                "origin_size": origin_p.numel(),
                "compare_size": compare_p.numel(),
                "model_part": model_part,
            }

    # Print different parameters
    print("Different parameters:")
    for param_name, param_info in diff_parameters.items():
        print(f"Parameter: {param_name}, Origin Size: {param_info['origin_size']}, Compare Size: {param_info['compare_size']}, Model Part: {param_info['model_part']}")
        total += param_info['compare_size']
    print(f"total changed:{total}")

# def compare_model(standard_model:torch.nn.Module, model:torch.nn.Module, mapping_dict= {}):
#     """
#     compare model parameters based on paramter name
#     for parameters with same name(common key), directly compare the paramter conetent
#     all other parameters will be regarded as differ paramters, except keys in mapping_dict.
#     mapping_dict is a python dict class, with keys as a subset of `origin_only_keys` and values as a subset of `compare_only_keys`.
    
#     """
#     device = "cuda"
#     origin_dict = dict(standard_model.named_parameters())
#     origin_keys = set(origin_dict.keys())
#     compare_dict = dict(model.named_parameters())
#     compare_keys = set(compare_dict.keys())
    
#     origin_only_keys = origin_keys - compare_keys
#     compare_only_keys = compare_keys - origin_keys
#     common_keys = set.intersection(origin_keys, compare_keys)
    
    
#     diff_paramters = 0
#     # compare parameters of common keys
#     for k in common_keys:
#         origin_p = origin_dict[k].to(device)
#         compare_p = compare_dict[k].to(device)
#         if origin_p.shape != compare_p.shape:
#             diff_paramters += origin_p.numel() + compare_p.numel()
#         elif (origin_p - compare_p).norm() != 0:
#             diff_paramters += origin_p.numel()
    
#     print(f"common diff: {diff_paramters}")
    
#     mapping_keys = set(mapping_dict.keys())
#     assert set.issubset(mapping_keys, origin_only_keys)
#     assert set.issubset(set(mapping_dict.values()), compare_only_keys)
    
#     for k in mapping_keys:
#         origin_p = origin_dict[k]
#         compare_p = compare_dict[mapping_keys[k]]
#         if origin_p.shape != compare_p.shape:
#             diff_paramters += origin_p.numel() + compare_p.numel()
#         elif (origin_p - compare_p).norm() != 0:
#             diff_paramters += origin_p.numel()
#     # all keys left are counted
#     extra_origin_keys = origin_only_keys - mapping_keys
#     extra_compare_keys = compare_only_keys - set(mapping_dict.values())
    
#     for k in extra_compare_keys:
#         diff_paramters += compare_dict[k]
    
#     for k in extra_origin_keys:
#         diff_paramters += origin_dict[k]    
    
#     print("diff parameters:", diff_paramters)
    
#     return diff_paramters

def get_args():
    parser = argparse.ArgumentParser()
    parser.add_argument("--restore_path", type=str, default="models/uvit_v1.pth", help="nnet path to resume")
    parser.add_argument("--prompt_path", type=str, default="eval_prompts/boy1.json", help="file contains prompts")
    parser.add_argument("--output_path", type=str, default="outputs/boy1", help="output dir for generated images")
    parser.add_argument("--weight_dir", type=str, default="model_output/girl2", help="output dir for weights of text encoder")
     
    return parser.parse_args()


def main(argv=None):
    # config args
    from configs.sample_config import get_config
    set_seed(42)
    config = get_config()
    args = get_args()
    
    # config.n_iter = 6
    # config.n_samples = 9
    
    config.output_path = args.output_path
    config.nnet_path = os.path.join(args.restore_path, "final.ckpt",'nnet.pth')
    device = "cuda"

    # init models
    nnet = UViT(**config.nnet)
    print(f'load nnet from {config.nnet_path}')
    nnet = get_peft_model(nnet,lora_config)

    nnet.load_state_dict(torch.load(config.nnet_path, map_location='cpu'), True)
    # for name,params in nnet.named_parameters():
    #     print(name)
    
    autoencoder = libs.autoencoder.get_model(**config.autoencoder)
    clip_text_model = FrozenCLIPEmbedder(version=config.clip_text_model, device=device)
    clip_text_model.to(device)
    clip_text_model.load_textual_inversion(args.weight_dir, token = "<new1>" , weight_name="<new1>.bin")
    
    # for name,params in nnet.named_parameters():
    #     if 'lora' in name:
    #         print(name,params)


    # nnet_mapping_dict = {"pos_embed", "pos_embed_token", "patch_embed.proj.weight", "patch_embed.proj.bias", "text_embed.weight", "text_embed.bias", "text_out.weight", "text_out.bias", "clip_img_embed.weight", "clip_img_embed.bias", "clip_img_out.weight", "clip_img_out.bias", "in_blocks.0.norm2.weight", "in_blocks.0.norm2.bias", "in_blocks.0.attn.qkv.weight", "in_blocks.0.attn.proj.weight", "in_blocks.0.attn.proj.bias", "in_blocks.0.norm3.weight", "in_blocks.0.norm3.bias", "in_blocks.0.mlp.fc1.weight", "in_blocks.0.mlp.fc1.bias", "in_blocks.0.mlp.fc2.weight", "in_blocks.0.mlp.fc2.bias", "in_blocks.1.norm2.weight", "in_blocks.1.norm2.bias", "in_blocks.1.attn.qkv.weight", "in_blocks.1.attn.proj.weight", "in_blocks.1.attn.proj.bias", "in_blocks.1.norm3.weight", "in_blocks.1.norm3.bias", "in_blocks.1.mlp.fc1.weight", "in_blocks.1.mlp.fc1.bias", "in_blocks.1.mlp.fc2.weight", "in_blocks.1.mlp.fc2.bias", "in_blocks.2.norm2.weight", "in_blocks.2.norm2.bias", "in_blocks.2.attn.qkv.weight", "in_blocks.2.attn.proj.weight", "in_blocks.2.attn.proj.bias", "in_blocks.2.norm3.weight", "in_blocks.2.norm3.bias", "in_blocks.2.mlp.fc1.weight", "in_blocks.2.mlp.fc1.bias", "in_blocks.2.mlp.fc2.weight", "in_blocks.2.mlp.fc2.bias", "in_blocks.3.norm2.weight", "in_blocks.3.norm2.bias", "in_blocks.3.attn.qkv.weight", "in_blocks.3.attn.proj.weight", "in_blocks.3.attn.proj.bias", "in_blocks.3.norm3.weight", "in_blocks.3.norm3.bias", "in_blocks.3.mlp.fc1.weight", "in_blocks.3.mlp.fc1.bias", "in_blocks.3.mlp.fc2.weight", "in_blocks.3.mlp.fc2.bias", "in_blocks.4.norm2.weight", "in_blocks.4.norm2.bias", "in_blocks.4.attn.qkv.weight", "in_blocks.4.attn.proj.weight", "in_blocks.4.attn.proj.bias", "in_blocks.4.norm3.weight", "in_blocks.4.norm3.bias", "in_blocks.4.mlp.fc1.weight", "in_blocks.4.mlp.fc1.bias", "in_blocks.4.mlp.fc2.weight", "in_blocks.4.mlp.fc2.bias", "in_blocks.5.norm2.weight", "in_blocks.5.norm2.bias", "in_blocks.5.attn.qkv.weight", "in_blocks.5.attn.proj.weight", "in_blocks.5.attn.proj.bias", "in_blocks.5.norm3.weight", "in_blocks.5.norm3.bias", "in_blocks.5.mlp.fc1.weight", "in_blocks.5.mlp.fc1.bias", "in_blocks.5.mlp.fc2.weight", "in_blocks.5.mlp.fc2.bias", "in_blocks.6.norm2.weight", "in_blocks.6.norm2.bias", "in_blocks.6.attn.qkv.weight", "in_blocks.6.attn.proj.weight", "in_blocks.6.attn.proj.bias", "in_blocks.6.norm3.weight", "in_blocks.6.norm3.bias", "in_blocks.6.mlp.fc1.weight", "in_blocks.6.mlp.fc1.bias", "in_blocks.6.mlp.fc2.weight", "in_blocks.6.mlp.fc2.bias", "in_blocks.7.norm2.weight", "in_blocks.7.norm2.bias", "in_blocks.7.attn.qkv.weight", "in_blocks.7.attn.proj.weight", "in_blocks.7.attn.proj.bias", "in_blocks.7.norm3.weight", "in_blocks.7.norm3.bias", "in_blocks.7.mlp.fc1.weight", "in_blocks.7.mlp.fc1.bias", "in_blocks.7.mlp.fc2.weight", "in_blocks.7.mlp.fc2.bias", "in_blocks.8.norm2.weight", "in_blocks.8.norm2.bias", "in_blocks.8.attn.qkv.weight", "in_blocks.8.attn.proj.weight", "in_blocks.8.attn.proj.bias", "in_blocks.8.norm3.weight", "in_blocks.8.norm3.bias", "in_blocks.8.mlp.fc1.weight", "in_blocks.8.mlp.fc1.bias", "in_blocks.8.mlp.fc2.weight", "in_blocks.8.mlp.fc2.bias", "in_blocks.9.norm2.weight", "in_blocks.9.norm2.bias", "in_blocks.9.attn.qkv.weight", "in_blocks.9.attn.proj.weight", "in_blocks.9.attn.proj.bias", "in_blocks.9.norm3.weight", "in_blocks.9.norm3.bias", "in_blocks.9.mlp.fc1.weight", "in_blocks.9.mlp.fc1.bias", "in_blocks.9.mlp.fc2.weight", "in_blocks.9.mlp.fc2.bias", "in_blocks.10.norm2.weight", "in_blocks.10.norm2.bias", "in_blocks.10.attn.qkv.weight", "in_blocks.10.attn.proj.weight", "in_blocks.10.attn.proj.bias", "in_blocks.10.norm3.weight", "in_blocks.10.norm3.bias", "in_blocks.10.mlp.fc1.weight", "in_blocks.10.mlp.fc1.bias", "in_blocks.10.mlp.fc2.weight", "in_blocks.10.mlp.fc2.bias", "in_blocks.11.norm2.weight", "in_blocks.11.norm2.bias", "in_blocks.11.attn.qkv.weight", "in_blocks.11.attn.proj.weight", "in_blocks.11.attn.proj.bias", "in_blocks.11.norm3.weight", "in_blocks.11.norm3.bias", "in_blocks.11.mlp.fc1.weight", "in_blocks.11.mlp.fc1.bias", "in_blocks.11.mlp.fc2.weight", "in_blocks.11.mlp.fc2.bias", "in_blocks.12.norm2.weight", "in_blocks.12.norm2.bias", "in_blocks.12.attn.qkv.weight", "in_blocks.12.attn.proj.weight", "in_blocks.12.attn.proj.bias", "in_blocks.12.norm3.weight", "in_blocks.12.norm3.bias", "in_blocks.12.mlp.fc1.weight", "in_blocks.12.mlp.fc1.bias", "in_blocks.12.mlp.fc2.weight", "in_blocks.12.mlp.fc2.bias", "in_blocks.13.norm2.weight", "in_blocks.13.norm2.bias", "in_blocks.13.attn.qkv.weight", "in_blocks.13.attn.proj.weight", "in_blocks.13.attn.proj.bias", "in_blocks.13.norm3.weight", "in_blocks.13.norm3.bias", "in_blocks.13.mlp.fc1.weight", "in_blocks.13.mlp.fc1.bias", "in_blocks.13.mlp.fc2.weight", "in_blocks.13.mlp.fc2.bias", "in_blocks.14.norm2.weight", "in_blocks.14.norm2.bias", "in_blocks.14.attn.qkv.weight", "in_blocks.14.attn.proj.weight", "in_blocks.14.attn.proj.bias", "in_blocks.14.norm3.weight", "in_blocks.14.norm3.bias", "in_blocks.14.mlp.fc1.weight", "in_blocks.14.mlp.fc1.bias", "in_blocks.14.mlp.fc2.weight", "in_blocks.14.mlp.fc2.bias", "mid_block.norm2.weight", "mid_block.norm2.bias", "mid_block.attn.qkv.weight", "mid_block.attn.proj.weight", "mid_block.attn.proj.bias", "mid_block.norm3.weight", "mid_block.norm3.bias", "mid_block.mlp.fc1.weight", "mid_block.mlp.fc1.bias", "mid_block.mlp.fc2.weight", "mid_block.mlp.fc2.bias", "out_blocks.0.norm1.weight", "out_blocks.0.norm1.bias", "out_blocks.0.norm2.weight", "out_blocks.0.norm2.bias", "out_blocks.0.attn.qkv.weight", "out_blocks.0.attn.proj.weight", "out_blocks.0.attn.proj.bias", "out_blocks.0.norm3.weight", "out_blocks.0.norm3.bias", "out_blocks.0.mlp.fc1.weight", "out_blocks.0.mlp.fc1.bias", "out_blocks.0.mlp.fc2.weight", "out_blocks.0.mlp.fc2.bias", "out_blocks.0.skip_linear.weight", "out_blocks.0.skip_linear.bias", "out_blocks.1.norm1.weight", "out_blocks.1.norm1.bias", "out_blocks.1.norm2.weight", "out_blocks.1.norm2.bias", "out_blocks.1.attn.qkv.weight", "out_blocks.1.attn.proj.weight", "out_blocks.1.attn.proj.bias", "out_blocks.1.norm3.weight", "out_blocks.1.norm3.bias", "out_blocks.1.mlp.fc1.weight", "out_blocks.1.mlp.fc1.bias", "out_blocks.1.mlp.fc2.weight", "out_blocks.1.mlp.fc2.bias", "out_blocks.1.skip_linear.weight", "out_blocks.1.skip_linear.bias", "out_blocks.2.norm1.weight", "out_blocks.2.norm1.bias", "out_blocks.2.norm2.weight", "out_blocks.2.norm2.bias", "out_blocks.2.attn.qkv.weight", "out_blocks.2.attn.proj.weight", "out_blocks.2.attn.proj.bias", "out_blocks.2.norm3.weight", "out_blocks.2.norm3.bias", "out_blocks.2.mlp.fc1.weight", "out_blocks.2.mlp.fc1.bias", "out_blocks.2.mlp.fc2.weight", "out_blocks.2.mlp.fc2.bias", "out_blocks.2.skip_linear.weight", "out_blocks.2.skip_linear.bias", "out_blocks.3.norm1.weight", "out_blocks.3.norm1.bias", "out_blocks.3.norm2.weight", "out_blocks.3.norm2.bias", "out_blocks.3.attn.qkv.weight", "out_blocks.3.attn.proj.weight", "out_blocks.3.attn.proj.bias", "out_blocks.3.norm3.weight", "out_blocks.3.norm3.bias", "out_blocks.3.mlp.fc1.weight", "out_blocks.3.mlp.fc1.bias", "out_blocks.3.mlp.fc2.weight", "out_blocks.3.mlp.fc2.bias", "out_blocks.3.skip_linear.weight", "out_blocks.3.skip_linear.bias", "out_blocks.4.norm1.weight", "out_blocks.4.norm1.bias", "out_blocks.4.norm2.weight", "out_blocks.4.norm2.bias", "out_blocks.4.attn.qkv.weight", "out_blocks.4.attn.proj.weight", "out_blocks.4.attn.proj.bias", "out_blocks.4.norm3.weight", "out_blocks.4.norm3.bias", "out_blocks.4.mlp.fc1.weight", "out_blocks.4.mlp.fc1.bias", "out_blocks.4.mlp.fc2.weight", "out_blocks.4.mlp.fc2.bias", "out_blocks.4.skip_linear.weight", "out_blocks.4.skip_linear.bias", "out_blocks.5.norm1.weight", "out_blocks.5.norm1.bias", "out_blocks.5.norm2.weight", "out_blocks.5.norm2.bias", "out_blocks.5.attn.qkv.weight", "out_blocks.5.attn.proj.weight", "out_blocks.5.attn.proj.bias", "out_blocks.5.norm3.weight", "out_blocks.5.norm3.bias", "out_blocks.5.mlp.fc1.weight", "out_blocks.5.mlp.fc1.bias", "out_blocks.5.mlp.fc2.weight", "out_blocks.5.mlp.fc2.bias", "out_blocks.5.skip_linear.weight", "out_blocks.5.skip_linear.bias", "out_blocks.6.norm1.weight", "out_blocks.6.norm1.bias", "out_blocks.6.norm2.weight", "out_blocks.6.norm2.bias", "out_blocks.6.attn.qkv.weight", "out_blocks.6.attn.proj.weight", "out_blocks.6.attn.proj.bias", "out_blocks.6.norm3.weight", "out_blocks.6.norm3.bias", "out_blocks.6.mlp.fc1.weight", "out_blocks.6.mlp.fc1.bias", "out_blocks.6.mlp.fc2.weight", "out_blocks.6.mlp.fc2.bias", "out_blocks.6.skip_linear.weight", "out_blocks.6.skip_linear.bias", "out_blocks.7.norm1.weight", "out_blocks.7.norm1.bias", "out_blocks.7.norm2.weight", "out_blocks.7.norm2.bias", "out_blocks.7.attn.qkv.weight", "out_blocks.7.attn.proj.weight", "out_blocks.7.attn.proj.bias", "out_blocks.7.norm3.weight", "out_blocks.7.norm3.bias", "out_blocks.7.mlp.fc1.weight", "out_blocks.7.mlp.fc1.bias", "out_blocks.7.mlp.fc2.weight", "out_blocks.7.mlp.fc2.bias", "out_blocks.7.skip_linear.weight", "out_blocks.7.skip_linear.bias", "out_blocks.8.norm1.weight", "out_blocks.8.norm1.bias", "out_blocks.8.norm2.weight", "out_blocks.8.norm2.bias", "out_blocks.8.attn.qkv.weight", "out_blocks.8.attn.proj.weight", "out_blocks.8.attn.proj.bias", "out_blocks.8.norm3.weight", "out_blocks.8.norm3.bias", "out_blocks.8.mlp.fc1.weight", "out_blocks.8.mlp.fc1.bias", "out_blocks.8.mlp.fc2.weight", "out_blocks.8.mlp.fc2.bias", "out_blocks.8.skip_linear.weight", "out_blocks.8.skip_linear.bias", "out_blocks.9.norm1.weight", "out_blocks.9.norm1.bias", "out_blocks.9.norm2.weight", "out_blocks.9.norm2.bias", "out_blocks.9.attn.qkv.weight", "out_blocks.9.attn.proj.weight", "out_blocks.9.attn.proj.bias", "out_blocks.9.norm3.weight", "out_blocks.9.norm3.bias", "out_blocks.9.mlp.fc1.weight", "out_blocks.9.mlp.fc1.bias", "out_blocks.9.mlp.fc2.weight", "out_blocks.9.mlp.fc2.bias", "out_blocks.9.skip_linear.weight", "out_blocks.9.skip_linear.bias", "out_blocks.10.norm1.weight", "out_blocks.10.norm1.bias", "out_blocks.10.norm2.weight", "out_blocks.10.norm2.bias", "out_blocks.10.attn.qkv.weight", "out_blocks.10.attn.proj.weight", "out_blocks.10.attn.proj.bias", "out_blocks.10.norm3.weight", "out_blocks.10.norm3.bias", "out_blocks.10.mlp.fc1.weight", "out_blocks.10.mlp.fc1.bias", "out_blocks.10.mlp.fc2.weight", "out_blocks.10.mlp.fc2.bias", "out_blocks.10.skip_linear.weight", "out_blocks.10.skip_linear.bias", "out_blocks.11.norm1.weight", "out_blocks.11.norm1.bias", "out_blocks.11.norm2.weight", "out_blocks.11.norm2.bias", "out_blocks.11.attn.qkv.weight", "out_blocks.11.attn.proj.weight", "out_blocks.11.attn.proj.bias", "out_blocks.11.norm3.weight", "out_blocks.11.norm3.bias", "out_blocks.11.mlp.fc1.weight", "out_blocks.11.mlp.fc1.bias", "out_blocks.11.mlp.fc2.weight", "out_blocks.11.mlp.fc2.bias", "out_blocks.11.skip_linear.weight", "out_blocks.11.skip_linear.bias", "out_blocks.12.norm1.weight", "out_blocks.12.norm1.bias", "out_blocks.12.norm2.weight", "out_blocks.12.norm2.bias", "out_blocks.12.attn.qkv.weight", "out_blocks.12.attn.proj.weight", "out_blocks.12.attn.proj.bias", "out_blocks.12.norm3.weight", "out_blocks.12.norm3.bias", "out_blocks.12.mlp.fc1.weight", "out_blocks.12.mlp.fc1.bias", "out_blocks.12.mlp.fc2.weight", "out_blocks.12.mlp.fc2.bias", "out_blocks.12.skip_linear.weight", "out_blocks.12.skip_linear.bias", "out_blocks.13.norm1.weight", "out_blocks.13.norm1.bias", "out_blocks.13.norm2.weight", "out_blocks.13.norm2.bias", "out_blocks.13.attn.qkv.weight", "out_blocks.13.attn.proj.weight", "out_blocks.13.attn.proj.bias", "out_blocks.13.norm3.weight", "out_blocks.13.norm3.bias", "out_blocks.13.mlp.fc1.weight", "out_blocks.13.mlp.fc1.bias", "out_blocks.13.mlp.fc2.weight", "out_blocks.13.mlp.fc2.bias", "out_blocks.13.skip_linear.weight", "out_blocks.13.skip_linear.bias", "out_blocks.14.norm1.weight", "out_blocks.14.norm1.bias", "out_blocks.14.norm2.weight", "out_blocks.14.norm2.bias", "out_blocks.14.attn.qkv.weight", "out_blocks.14.attn.proj.weight", "out_blocks.14.attn.proj.bias", "out_blocks.14.norm3.weight", "out_blocks.14.norm3.bias", "out_blocks.14.mlp.fc1.weight", "out_blocks.14.mlp.fc1.bias", "out_blocks.14.mlp.fc2.weight", "out_blocks.14.mlp.fc2.bias", "out_blocks.14.skip_linear.weight", "out_blocks.14.skip_linear.bias", "norm.weight", "norm.bias", "decoder_pred.weight", "decoder_pred.bias", "token_embedding.weight"}
    nnet_mapping_dict = {}
    autoencoder_mapping_dict = {}
    clip_text_mapping_dict = {}
    
    print("####### evaluating changed paramters")
    total_diff_parameters = 0
    print(">>> evaluating nnet changed paramters")
    nnet_standard = UViT(**config.nnet)
    nnet_standard.load_state_dict(torch.load("models/uvit_v1.pth", map_location='cpu'), False)
    nnet_standard = get_peft_model(nnet_standard,lora_config)
    compare_and_print_models(nnet_standard, nnet, nnet_mapping_dict)
    del nnet_standard


    # print(">>> evaluating autoencoder changed paramters")
    # autoencoder_standard = libs.autoencoder.get_model(**config.autoencoder)
    # # total_diff_parameters += compare_model(autoencoder_standard, autoencoder, autoencoder_mapping_dict)
    # del autoencoder_standard
    
    print(">>> evaluating clip text changed paramters")
    clip_text_strandard = FrozenCLIPEmbedder(version=config.clip_text_model, device=device).to("cpu")
    compare_and_print_models(clip_text_strandard, clip_text_model, clip_text_mapping_dict)
    del clip_text_strandard
    

    clip_text_model.to(device)
    autoencoder.to(device)
    nnet.to(device)
    
    # 基于给定的prompt进行生成
    prompts = json.load(open(args.prompt_path, "r"))
    for prompt_index, prompt in enumerate(prompts):
        # 根据训练策略
        if "boy" in prompt:
            prompt = prompt.replace("boy", "<new1> boy")
        else:
            prompt = prompt.replace("girl", "<new1> girl")

        config.prompt = prompt
        print("sampling with prompt:", prompt)
        sample(prompt_index, prompt, config, nnet, clip_text_model, autoencoder, device)
        
    # print("total changed paramters:", total_diff_parameters)

if __name__ == "__main__":
    main()