Spaces:
Running
on
Zero
Running
on
Zero
Delete app_old.py
Browse files- app_old.py +0 -374
app_old.py
DELETED
@@ -1,374 +0,0 @@
|
|
1 |
-
import gradio as gr
|
2 |
-
import torch
|
3 |
-
import spaces
|
4 |
-
from PIL import Image, ImageDraw, ImageFont
|
5 |
-
# from src.condition import Condition
|
6 |
-
from diffusers.pipelines import FluxPipeline
|
7 |
-
import numpy as np
|
8 |
-
import requests
|
9 |
-
from huggingface_hub import hf_hub_download
|
10 |
-
from safetensors.torch import load_file
|
11 |
-
import torch.multiprocessing as mp
|
12 |
-
###
|
13 |
-
import argparse
|
14 |
-
import logging
|
15 |
-
import math
|
16 |
-
import os
|
17 |
-
import re
|
18 |
-
import random
|
19 |
-
import shutil
|
20 |
-
from contextlib import nullcontext
|
21 |
-
from pathlib import Path
|
22 |
-
from PIL import Image
|
23 |
-
import accelerate
|
24 |
-
import datasets
|
25 |
-
import numpy as np
|
26 |
-
import torch
|
27 |
-
import torch.nn.functional as F
|
28 |
-
from torch import Tensor, nn
|
29 |
-
import torch.utils.checkpoint
|
30 |
-
import transformers
|
31 |
-
from accelerate import Accelerator
|
32 |
-
from accelerate.logging import get_logger
|
33 |
-
from accelerate.state import AcceleratorState
|
34 |
-
from accelerate.utils import ProjectConfiguration, set_seed
|
35 |
-
from huggingface_hub import create_repo, upload_folder
|
36 |
-
from packaging import version
|
37 |
-
from tqdm.auto import tqdm
|
38 |
-
from transformers import CLIPTextModel, CLIPTokenizer
|
39 |
-
from transformers import CLIPVisionModelWithProjection, CLIPImageProcessor
|
40 |
-
from transformers.utils import ContextManagers
|
41 |
-
from omegaconf import OmegaConf
|
42 |
-
from copy import deepcopy
|
43 |
-
import diffusers
|
44 |
-
from diffusers import AutoencoderKL, DDPMScheduler, StableDiffusionPipeline
|
45 |
-
from diffusers.optimization import get_scheduler
|
46 |
-
from diffusers.training_utils import EMAModel, compute_dream_and_update_latents, compute_snr
|
47 |
-
from diffusers.utils import check_min_version, deprecate, make_image_grid
|
48 |
-
from diffusers.utils.hub_utils import load_or_create_model_card, populate_model_card
|
49 |
-
from diffusers.utils.import_utils import is_xformers_available
|
50 |
-
from diffusers.utils.torch_utils import is_compiled_module
|
51 |
-
from einops import rearrange
|
52 |
-
from src.flux.sampling import denoise, get_noise, get_schedule, prepare, unpack
|
53 |
-
from src.flux.util import (configs, load_ae, load_clip,
|
54 |
-
load_flow_model2, load_t5, save_image, tensor_to_pil_image, load_checkpoint)
|
55 |
-
from src.flux.modules.layers import DoubleStreamBlockLoraProcessor, SingleStreamBlockLoraProcessor, IPDoubleStreamBlockProcessor, IPSingleStreamBlockProcessor, ImageProjModel
|
56 |
-
from src.flux.xflux_pipeline import XFluxSampler
|
57 |
-
|
58 |
-
from image_datasets.dataset import loader, eval_image_pair_loader, image_resize
|
59 |
-
|
60 |
-
from safetensors.torch import load_file
|
61 |
-
import json
|
62 |
-
|
63 |
-
|
64 |
-
# logger = get_logger(__name__, log_level="INFO")
|
65 |
-
|
66 |
-
|
67 |
-
def get_models(name: str, device, offload: bool, is_schnell: bool):
|
68 |
-
t5 = load_t5(device, max_length=256 if is_schnell else 512)
|
69 |
-
clip = load_clip(device)
|
70 |
-
clip.requires_grad_(False)
|
71 |
-
model = load_flow_model2(name, device="cpu")
|
72 |
-
vae = load_ae(name, device="cpu" if offload else device)
|
73 |
-
return model, vae, t5, clip
|
74 |
-
|
75 |
-
args = OmegaConf.load("inference_configs/inference.yaml") #OmegaConf.load(parse_args())
|
76 |
-
is_schnell = args.model_name == "flux-schnell"
|
77 |
-
set_seed(args.seed)
|
78 |
-
# logging_dir = os.path.join(args.output_dir, args.logging_dir)
|
79 |
-
device = "cuda"
|
80 |
-
dit, vae, t5, clip = get_models(name=args.model_name, device=device, offload=False, is_schnell=is_schnell)
|
81 |
-
|
82 |
-
# # load image encoder
|
83 |
-
# ip_image_encoder = CLIPVisionModelWithProjection.from_pretrained(os.getenv("CLIP_VIT")).to(
|
84 |
-
# # accelerator.device, dtype=torch.bfloat16
|
85 |
-
# device, dtype=torch.bfloat16
|
86 |
-
# )
|
87 |
-
# ip_clip_image_processor = CLIPImageProcessor()
|
88 |
-
|
89 |
-
if args.use_ip:
|
90 |
-
sampler = XFluxSampler(clip=clip, t5=t5, ae=vae, model=dit, device=device, ip_loaded=True, spatial_condition=False, clip_image_processor=ip_clip_image_processor, image_encoder=ip_image_encoder, improj=ip_improj)
|
91 |
-
elif args.use_spatial_condition:
|
92 |
-
sampler = XFluxSampler(clip=clip, t5=t5, ae=vae, model=dit, device=device, ip_loaded=False, spatial_condition=True, clip_image_processor=None, image_encoder=None, improj=None,share_position_embedding=args.share_position_embedding)
|
93 |
-
else:
|
94 |
-
sampler = XFluxSampler(clip=clip, t5=t5, ae=vae, model=dit, device=device, ip_loaded=False, spatial_condition=False, clip_image_processor=None, image_encoder=None, improj=None)
|
95 |
-
|
96 |
-
|
97 |
-
# @spaces.GPU
|
98 |
-
def generate(image, edit_prompt):
|
99 |
-
print("hello?????????!!!!!")
|
100 |
-
|
101 |
-
# accelerator_project_config = ProjectConfiguration(project_dir=args.output_dir, logging_dir=logging_dir)
|
102 |
-
|
103 |
-
# accelerator = Accelerator(
|
104 |
-
# gradient_accumulation_steps=1,
|
105 |
-
# mixed_precision=args.mixed_precision,
|
106 |
-
# log_with=args.report_to,
|
107 |
-
# project_config=accelerator_project_config,
|
108 |
-
# )
|
109 |
-
|
110 |
-
# Make one log on every process with the configuration for debugging.
|
111 |
-
# logging.basicConfig(
|
112 |
-
# format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
|
113 |
-
# datefmt="%m/%d/%Y %H:%M:%S",
|
114 |
-
# level=logging.INFO,
|
115 |
-
# )
|
116 |
-
# logger.info(accelerator.state, main_process_only=False)
|
117 |
-
# if accelerator.is_local_main_process:
|
118 |
-
# datasets.utils.logging.set_verbosity_warning()
|
119 |
-
# transformers.utils.logging.set_verbosity_warning()
|
120 |
-
# diffusers.utils.logging.set_verbosity_info()
|
121 |
-
# else:
|
122 |
-
# datasets.utils.logging.set_verbosity_error()
|
123 |
-
# transformers.utils.logging.set_verbosity_error()
|
124 |
-
# diffusers.utils.logging.set_verbosity_error()
|
125 |
-
|
126 |
-
|
127 |
-
# if accelerator.is_main_process:
|
128 |
-
# if args.output_dir is not None:
|
129 |
-
# os.makedirs(args.output_dir, exist_ok=True)
|
130 |
-
# gpt_eval_path = os.path.join(args.output_dir,"Eval")
|
131 |
-
# os.makedirs(gpt_eval_path, exist_ok=True)
|
132 |
-
|
133 |
-
# dit, vae, t5, clip = get_models(name=args.model_name, device=accelerator.device, offload=False, is_schnell=is_schnell)
|
134 |
-
# dit, vae, t5, clip = get_models(name=args.model_name, device=device, offload=False, is_schnell=is_schnell)
|
135 |
-
|
136 |
-
if args.use_lora:
|
137 |
-
lora_attn_procs = {}
|
138 |
-
if args.use_ip:
|
139 |
-
ip_attn_procs = {}
|
140 |
-
if args.double_blocks is None:
|
141 |
-
double_blocks_idx = list(range(19))
|
142 |
-
else:
|
143 |
-
double_blocks_idx = [int(idx) for idx in args.double_blocks.split(",")]
|
144 |
-
|
145 |
-
if args.single_blocks is None:
|
146 |
-
single_blocks_idx = list(range(38))
|
147 |
-
elif args.single_blocks is not None:
|
148 |
-
single_blocks_idx = [int(idx) for idx in args.single_blocks.split(",")]
|
149 |
-
|
150 |
-
if args.use_lora:
|
151 |
-
for name, attn_processor in dit.attn_processors.items():
|
152 |
-
match = re.search(r'\.(\d+)\.', name)
|
153 |
-
if match:
|
154 |
-
layer_index = int(match.group(1))
|
155 |
-
|
156 |
-
if name.startswith("double_blocks") and layer_index in double_blocks_idx:
|
157 |
-
# if accelerator.is_main_process:
|
158 |
-
# print("setting LoRA Processor for", name)
|
159 |
-
lora_attn_procs[name] = DoubleStreamBlockLoraProcessor(
|
160 |
-
dim=3072, rank=args.rank
|
161 |
-
)
|
162 |
-
elif name.startswith("single_blocks") and layer_index in single_blocks_idx:
|
163 |
-
# if accelerator.is_main_process:
|
164 |
-
# print("setting LoRA Processor for", name)
|
165 |
-
lora_attn_procs[name] = SingleStreamBlockLoraProcessor(
|
166 |
-
dim=3072, rank=args.rank
|
167 |
-
)
|
168 |
-
else:
|
169 |
-
lora_attn_procs[name] = attn_processor
|
170 |
-
|
171 |
-
dit.set_attn_processor(lora_attn_procs)
|
172 |
-
|
173 |
-
# if args.use_ip:
|
174 |
-
# # unpack checkpoint
|
175 |
-
# checkpoint = load_checkpoint(args.ip_local_path, args.ip_repo_id, args.ip_name)
|
176 |
-
# prefix = "double_blocks."
|
177 |
-
# # blocks = {}
|
178 |
-
# proj = {}
|
179 |
-
|
180 |
-
# for key, value in checkpoint.items():
|
181 |
-
# # if key.startswith(prefix):
|
182 |
-
# # blocks[key[len(prefix):].replace('.processor.', '.')] = value
|
183 |
-
# if key.startswith("ip_adapter_proj_model"):
|
184 |
-
# proj[key[len("ip_adapter_proj_model."):]] = value
|
185 |
-
|
186 |
-
# # # load image encoder
|
187 |
-
# # ip_image_encoder = CLIPVisionModelWithProjection.from_pretrained(os.getenv("CLIP_VIT")).to(
|
188 |
-
# # # accelerator.device, dtype=torch.bfloat16
|
189 |
-
# # device, dtype=torch.bfloat16
|
190 |
-
# # )
|
191 |
-
# # ip_clip_image_processor = CLIPImageProcessor()
|
192 |
-
|
193 |
-
# # setup image embedding projection model
|
194 |
-
# ip_improj = ImageProjModel(4096, 768, 4)
|
195 |
-
# ip_improj.load_state_dict(proj)
|
196 |
-
# # ip_improj = ip_improj.to(accelerator.device, dtype=torch.bfloat16)
|
197 |
-
# ip_improj = ip_improj.to(device, dtype=torch.bfloat16)
|
198 |
-
|
199 |
-
# ip_attn_procs = {}
|
200 |
-
|
201 |
-
# for name, _ in dit.attn_processors.items():
|
202 |
-
# ip_state_dict = {}
|
203 |
-
# for k in checkpoint.keys():
|
204 |
-
# if name in k:
|
205 |
-
# ip_state_dict[k.replace(f'{name}.', '')] = checkpoint[k]
|
206 |
-
# if ip_state_dict:
|
207 |
-
# ip_attn_procs[name] = IPDoubleStreamBlockProcessor(4096, 3072)
|
208 |
-
# ip_attn_procs[name].load_state_dict(ip_state_dict)
|
209 |
-
# ip_attn_procs[name].to(accelerator.device, dtype=torch.bfloat16)
|
210 |
-
# else:
|
211 |
-
# ip_attn_procs[name] = dit.attn_processors[name]
|
212 |
-
# dit.set_attn_processor(ip_attn_procs)
|
213 |
-
|
214 |
-
|
215 |
-
vae.requires_grad_(False)
|
216 |
-
t5.requires_grad_(False)
|
217 |
-
clip.requires_grad_(False)
|
218 |
-
|
219 |
-
|
220 |
-
|
221 |
-
# weight_dtype = torch.float32
|
222 |
-
# if accelerator.mixed_precision == "fp16":
|
223 |
-
# weight_dtype = torch.float16
|
224 |
-
# args.mixed_precision = accelerator.mixed_precision
|
225 |
-
# elif accelerator.mixed_precision == "bf16":
|
226 |
-
# weight_dtype = torch.bfloat16
|
227 |
-
# args.mixed_precision = accelerator.mixed_precision
|
228 |
-
|
229 |
-
|
230 |
-
# print(f"Resuming from checkpoint {args.ckpt_dir}")
|
231 |
-
# dit_stat_dict = load_file(args.ckpt_dir)
|
232 |
-
# Get path from Hub
|
233 |
-
model_path = hf_hub_download(
|
234 |
-
repo_id="Boese0601/ByteMorpher",
|
235 |
-
filename="dit.safetensors"
|
236 |
-
)
|
237 |
-
state_dict = load_file(model_path)
|
238 |
-
dit.load_state_dict(state_dict)
|
239 |
-
dit = dit.to(weight_dtype)
|
240 |
-
dit.eval()
|
241 |
-
|
242 |
-
# test_dataloader = loader(**args.data_config)
|
243 |
-
test_dataloader = eval_image_pair_loader(**args.data_config)
|
244 |
-
|
245 |
-
|
246 |
-
|
247 |
-
# from deepspeed import initialize
|
248 |
-
dit = accelerator.prepare(dit)
|
249 |
-
|
250 |
-
# if accelerator.is_main_process:
|
251 |
-
# accelerator.init_trackers(args.tracker_project_name, {"test": None})
|
252 |
-
|
253 |
-
# logger.info("***** Running Evaluation *****")
|
254 |
-
# logger.info(f" Instantaneous batch size = {args.eval_batch_size}")
|
255 |
-
|
256 |
-
|
257 |
-
|
258 |
-
# progress_bar = tqdm(
|
259 |
-
# range(0, len(test_dataloader)),
|
260 |
-
# initial=0,
|
261 |
-
# desc="Steps",
|
262 |
-
# disable=not accelerator.is_local_main_process,
|
263 |
-
# )
|
264 |
-
|
265 |
-
# for step, batch in enumerate(test_dataloader):
|
266 |
-
# with accelerator.accumulate(dit):
|
267 |
-
# img, tgt_image, prompt, edit_prompt, img_name, edit_name = batch
|
268 |
-
img = image_resize(image, 512)
|
269 |
-
w, h = img.size
|
270 |
-
new_w = (w // 32) * 32
|
271 |
-
new_h = (h // 32) * 32
|
272 |
-
img = img.resize((new_w, new_h))
|
273 |
-
img = torch.from_numpy((np.array(img) / 127.5) - 1)
|
274 |
-
img = img.permute(2, 0, 1).unsqueeze(0)
|
275 |
-
|
276 |
-
edit_prompt = edit_prompt
|
277 |
-
|
278 |
-
# if args.use_ip:
|
279 |
-
# sampler = XFluxSampler(clip=clip, t5=t5, ae=vae, model=dit, device=accelerator.device, ip_loaded=True, spatial_condition=False, clip_image_processor=ip_clip_image_processor, image_encoder=ip_image_encoder, improj=ip_improj)
|
280 |
-
# elif args.use_spatial_condition:
|
281 |
-
# sampler = XFluxSampler(clip=clip, t5=t5, ae=vae, model=dit, device=accelerator.device, ip_loaded=False, spatial_condition=True, clip_image_processor=None, image_encoder=None, improj=None,share_position_embedding=args.share_position_embedding)
|
282 |
-
# else:
|
283 |
-
# sampler = XFluxSampler(clip=clip, t5=t5, ae=vae, model=dit, device=accelerator.device, ip_loaded=False, spatial_condition=False, clip_image_processor=None, image_encoder=None, improj=None)
|
284 |
-
with torch.no_grad():
|
285 |
-
result = sampler(prompt=edit_prompt,
|
286 |
-
width=args.sample_width,
|
287 |
-
height=args.sample_height,
|
288 |
-
num_steps=args.sample_steps,
|
289 |
-
image_prompt=None, # ip_adapter
|
290 |
-
true_gs=args.cfg_scale,
|
291 |
-
seed=args.seed,
|
292 |
-
ip_scale=args.ip_scale if args.use_ip else 1.0,
|
293 |
-
source_image=img if args.use_spatial_condition else None,
|
294 |
-
)
|
295 |
-
gen_img = result
|
296 |
-
|
297 |
-
|
298 |
-
|
299 |
-
# progress_bar.update(1)
|
300 |
-
|
301 |
-
# accelerator.wait_for_everyone()
|
302 |
-
# accelerator.end_training()
|
303 |
-
return gen_img
|
304 |
-
|
305 |
-
|
306 |
-
def get_samples():
|
307 |
-
sample_list = [
|
308 |
-
{
|
309 |
-
"image": "assets/0_camera_zoom/20486354.png",
|
310 |
-
"edit_prompt": "Zoom in on the coral and add a small blue fish in the background.",
|
311 |
-
},
|
312 |
-
]
|
313 |
-
return [
|
314 |
-
[
|
315 |
-
Image.open(sample["image"]).resize((512, 512)),
|
316 |
-
sample["edit_prompt"],
|
317 |
-
]
|
318 |
-
for sample in sample_list
|
319 |
-
]
|
320 |
-
|
321 |
-
|
322 |
-
header = """
|
323 |
-
# ByteMoprh
|
324 |
-
|
325 |
-
<div style="text-align: center; display: flex; justify-content: left; gap: 5px;">
|
326 |
-
<a href=""><img src="https://img.shields.io/badge/ariXv-Paper-A42C25.svg" alt="arXiv"></a>
|
327 |
-
<a href="https://huggingface.co/datasets/Boese0601/ByteMorph-Bench"><img src="https://img.shields.io/badge/🤗-Model-ffbd45.svg" alt="HuggingFace"></a>
|
328 |
-
<a href="https://github.com/Boese0601/ByteMorph"><img src="https://img.shields.io/badge/GitHub-Code-blue.svg?logo=github&" alt="GitHub"></a>
|
329 |
-
</div>
|
330 |
-
"""
|
331 |
-
|
332 |
-
|
333 |
-
def create_app():
|
334 |
-
with gr.Blocks() as app:
|
335 |
-
gr.Markdown(header, elem_id="header")
|
336 |
-
with gr.Row(equal_height=False):
|
337 |
-
with gr.Column(variant="panel", elem_classes="inputPanel"):
|
338 |
-
original_image = gr.Image(
|
339 |
-
type="pil", label="Condition Image", width=300, elem_id="input"
|
340 |
-
)
|
341 |
-
edit_prompt = gr.Textbox(lines=2, label="Edit Prompt", elem_id="edit_prompt")
|
342 |
-
submit_btn = gr.Button("Run", elem_id="submit_btn")
|
343 |
-
|
344 |
-
with gr.Column(variant="panel", elem_classes="outputPanel"):
|
345 |
-
output_image = gr.Image(type="pil", elem_id="output")
|
346 |
-
|
347 |
-
with gr.Row():
|
348 |
-
examples = gr.Examples(
|
349 |
-
examples=get_samples(),
|
350 |
-
inputs=[original_image, edit_prompt],
|
351 |
-
label="Examples",
|
352 |
-
)
|
353 |
-
|
354 |
-
submit_btn.click(
|
355 |
-
fn=generate,
|
356 |
-
inputs=[original_image, edit_prompt],
|
357 |
-
outputs=output_image,
|
358 |
-
)
|
359 |
-
gr.HTML(
|
360 |
-
"""
|
361 |
-
<div style="text-align: center;">
|
362 |
-
* This demo's template was modified from <a href="https://arxiv.org/abs/2411.15098" target="_blank">OminiControl</a>.
|
363 |
-
</div>
|
364 |
-
"""
|
365 |
-
)
|
366 |
-
return app
|
367 |
-
|
368 |
-
|
369 |
-
if __name__ == "__main__":
|
370 |
-
print("CUDA available:", torch.cuda.is_available())
|
371 |
-
print("CUDA version:", torch.version.cuda)
|
372 |
-
print("GPU device name:", torch.cuda.get_device_name(0) if torch.cuda.is_available() else "None")
|
373 |
-
# mp.set_start_method("spawn", force=True)
|
374 |
-
create_app().launch(debug=False, share=True, ssr_mode=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|