File size: 13,128 Bytes
e368cec
 
 
 
52026b9
619dcd0
e368cec
cfd968b
b34109c
e368cec
45c2aa5
 
 
e368cec
 
 
 
944dd2b
868a596
9bf990d
e368cec
bd0bce1
e368cec
 
bd0bce1
e368cec
 
 
 
45c2aa5
 
 
 
 
 
52026b9
 
45c2aa5
 
 
 
 
 
619dcd0
2af380b
e368cec
fcd973b
 
 
 
 
e368cec
5777088
94bd22c
fcd973b
 
 
 
 
94bd22c
 
f7a6c5f
 
 
 
 
 
 
 
 
e368cec
868a596
 
e368cec
868a596
e368cec
 
 
 
94bd22c
 
a071819
e368cec
5777088
f7a6c5f
868a596
 
f7a6c5f
868a596
f7a6c5f
 
 
 
 
 
 
 
 
 
 
 
e368cec
 
 
94bd22c
 
a071819
e368cec
bd0bce1
f7a6c5f
 
 
 
 
 
 
 
 
 
3bb2685
e368cec
fcd973b
 
 
 
 
e368cec
5777088
f7a6c5f
 
 
 
 
 
 
 
 
e368cec
 
 
a071819
 
 
 
e368cec
5777088
f7a6c5f
 
 
 
 
 
 
 
 
 
 
 
e368cec
868a596
 
e368cec
868a596
e368cec
 
 
a071819
 
944dd2b
 
f7a6c5f
868a596
 
f7a6c5f
868a596
f7a6c5f
 
 
 
 
 
 
 
 
 
 
 
83e2394
944dd2b
b3e524c
 
 
 
 
944dd2b
 
94bd22c
b3e524c
 
 
 
 
94bd22c
 
f7a6c5f
b34109c
 
 
 
 
 
f7a6c5f
944dd2b
868a596
 
944dd2b
868a596
944dd2b
 
 
 
94bd22c
 
a071819
944dd2b
 
f7a6c5f
868a596
 
98bef81
f7a6c5f
868a596
9bf990d
 
98bef81
f7a6c5f
 
 
b34109c
 
 
 
 
 
 
 
f7a6c5f
944dd2b
 
 
94bd22c
 
a071819
f7a6c5f
 
 
 
b34109c
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
import concurrent.futures 
import random
import gradio as gr
import requests
import io, base64, json, os
import spaces
from PIL import Image
from .models import IMAGE_GENERATION_MODELS, IMAGE_EDITION_MODELS, VIDEO_GENERATION_MODELS, MUSEUM_UNSUPPORTED_MODELS, DESIRED_APPEAR_MODEL, load_pipeline
from .fetch_museum_results import draw_from_imagen_museum, draw2_from_imagen_museum, draw_from_videogen_museum, draw2_from_videogen_museum

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

class ModelManager:
    def __init__(self):
        self.model_ig_list = IMAGE_GENERATION_MODELS
        self.model_ie_list = IMAGE_EDITION_MODELS
        self.model_vg_list = VIDEO_GENERATION_MODELS
        self.excluding_model_list = MUSEUM_UNSUPPORTED_MODELS
        self.desired_model_list = DESIRED_APPEAR_MODEL
        self.loaded_models = {}

    def load_model_pipe(self, model_name):
        if not model_name in self.loaded_models:
            pipe = load_pipeline(model_name)
            self.loaded_models[model_name] = pipe
        else:
            pipe = self.loaded_models[model_name]
        return pipe

    @spaces.GPU(duration=20)
    def NSFW_filter(self, prompt):
        model_id = "meta-llama/Meta-Llama-Guard-2-8B"
        device = "cuda"
        dtype = torch.bfloat16
        tokenizer = AutoTokenizer.from_pretrained(model_id, token=os.environ['HF_GUARD'])
        model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=dtype, device_map=device, token=os.environ['HF_GUARD'])
        chat = [{"role": "user", "content": prompt}]
        input_ids = tokenizer.apply_chat_template(chat, return_tensors="pt").to(device)
        output = model.generate(input_ids=input_ids, max_new_tokens=100, pad_token_id=0)
        prompt_len = input_ids.shape[-1]
        result = tokenizer.decode(output[0][prompt_len:], skip_special_tokens=True)
        return result
    
    @spaces.GPU(duration=120)
    def generate_image_ig(self, prompt, model_name):
        #if self.NSFW_filter(prompt) == 'safe':
        pipe = self.load_model_pipe(model_name)
        result = pipe(prompt=prompt)
        # else:
        #    result = ''
        return result

    def generate_image_ig_api(self, prompt, model_name):
        # if self.NSFW_filter(prompt) == 'safe':
        pipe = self.load_model_pipe(model_name)
        result = pipe(prompt=prompt)
        # else:
        #    result = ''
        return result

    def generate_image_ig_museum(self, model_name):
        model_name = model_name.split('_')[1]
        result_list = draw_from_imagen_museum("t2i", model_name)
        image_link = result_list[0]
        prompt = result_list[1]

        return image_link, prompt


    def generate_image_ig_parallel_anony(self, prompt, model_A, model_B):
        # Using list comprehension to get the difference between two lists
        picking_list = [item for item in self.model_ig_list if item not in self.excluding_model_list]
        if model_A == "" and model_B == "":
            model_names = random.sample([model for model in picking_list], 2)
        else:
            model_names = [model_A, model_B]

        with concurrent.futures.ThreadPoolExecutor() as executor:
            futures = [executor.submit(self.generate_image_ig, prompt, model) if model.startswith("imagenhub")
                       else executor.submit(self.generate_image_ig_api, prompt, model) for model in model_names]
            results = [future.result() for future in futures]
        return results[0], results[1], model_names[0], model_names[1]

    def generate_image_ig_museum_parallel_anony(self, model_A, model_B):
        # Using list comprehension to get the difference between two lists
        picking_list = [item for item in self.model_ig_list if item not in self.excluding_model_list]
        if model_A == "" and model_B == "":
            model_names = random.sample([model for model in picking_list], 2)
        else:
            model_names = [model_A, model_B]

        with concurrent.futures.ThreadPoolExecutor() as executor:
            model_1 = model_names[0].split('_')[1]
            model_2 = model_names[1].split('_')[1]
            result_list = draw2_from_imagen_museum("t2i", model_1, model_2)
            image_links = result_list[0]
            prompt_list = result_list[1]

        return image_links[0], image_links[1], model_names[0], model_names[1], prompt_list[0]
    
    def generate_image_ig_parallel(self, prompt, model_A, model_B):
        model_names = [model_A, model_B]
        with concurrent.futures.ThreadPoolExecutor() as executor:
            futures = [executor.submit(self.generate_image_ig, prompt, model) if model.startswith("imagenhub")
                       else executor.submit(self.generate_image_ig_api, prompt, model) for model in model_names]
            results = [future.result() for future in futures]
        return results[0], results[1]

    def generate_image_ig_museum_parallel(self, model_A, model_B):
        with concurrent.futures.ThreadPoolExecutor() as executor:
            model_1 = model_A.split('_')[1]
            model_2 = model_B.split('_')[1]
            result_list = draw2_from_imagen_museum("t2i", model_1, model_2)
            image_links = result_list[0]
            prompt_list = result_list[1]
        return image_links[0], image_links[1], prompt_list[0]


    @spaces.GPU(duration=200)
    def generate_image_ie(self, textbox_source, textbox_target, textbox_instruct, source_image, model_name):
        # if self.NSFW_filter(" ".join([textbox_source, textbox_target, textbox_instruct])) == 'safe':
        pipe = self.load_model_pipe(model_name)
        result = pipe(src_image = source_image, src_prompt = textbox_source, target_prompt = textbox_target, instruct_prompt = textbox_instruct)
        # else:
        #    result = ''
        return result

    def generate_image_ie_museum(self, model_name):
        model_name = model_name.split('_')[1]
        result_list = draw_from_imagen_museum("tie", model_name)
        image_links = result_list[0]
        prompt_list = result_list[1]
        # image_links = [src, model]
        # prompt_list = [source_caption, target_caption, instruction]
        return image_links[0], image_links[1], prompt_list[0], prompt_list[1], prompt_list[2]

    def generate_image_ie_parallel(self, textbox_source, textbox_target, textbox_instruct, source_image, model_A, model_B):
        model_names = [model_A, model_B]
        with concurrent.futures.ThreadPoolExecutor() as executor:
            futures = [
                executor.submit(self.generate_image_ie, textbox_source, textbox_target, textbox_instruct, source_image,
                                model) for model in model_names]
            results = [future.result() for future in futures]
        return results[0], results[1]

    def generate_image_ie_museum_parallel(self, model_A, model_B):
        model_names = [model_A, model_B]
        with concurrent.futures.ThreadPoolExecutor() as executor:
            model_1 = model_names[0].split('_')[1]
            model_2 = model_names[1].split('_')[1]
            result_list = draw2_from_imagen_museum("tie", model_1, model_2)
            image_links = result_list[0]
            prompt_list = result_list[1]
            # image_links = [src, model_A, model_B]
            # prompt_list = [source_caption, target_caption, instruction]
        return image_links[0], image_links[1], image_links[2], prompt_list[0], prompt_list[1], prompt_list[2]

    def generate_image_ie_parallel_anony(self, textbox_source, textbox_target, textbox_instruct, source_image, model_A, model_B):
        # Using list comprehension to get the difference between two lists
        picking_list = [item for item in self.model_ie_list if item not in self.excluding_model_list]
        if model_A == "" and model_B == "":
            model_names = random.sample([model for model in picking_list], 2)
        else:
            model_names = [model_A, model_B]
        with concurrent.futures.ThreadPoolExecutor() as executor:
            futures = [executor.submit(self.generate_image_ie, textbox_source, textbox_target, textbox_instruct, source_image, model) for model in model_names]
            results = [future.result() for future in futures]
        return results[0], results[1], model_names[0], model_names[1]

    def generate_image_ie_museum_parallel_anony(self, model_A, model_B):
        # Using list comprehension to get the difference between two lists
        picking_list = [item for item in self.model_ie_list if item not in self.excluding_model_list]
        if model_A == "" and model_B == "":
            model_names = random.sample([model for model in picking_list], 2)
        else:
            model_names = [model_A, model_B]
        with concurrent.futures.ThreadPoolExecutor() as executor:
            model_1 = model_names[0].split('_')[1]
            model_2 = model_names[1].split('_')[1]
            result_list = draw2_from_imagen_museum("tie", model_1, model_2)
            image_links = result_list[0]
            prompt_list = result_list[1]
            # image_links = [src, model_A, model_B]
            # prompt_list = [source_caption, target_caption, instruction]
        return image_links[0], image_links[1], image_links[2], prompt_list[0], prompt_list[1], prompt_list[2], model_names[0], model_names[1]

    @spaces.GPU(duration=150)
    def generate_video_vg(self, prompt, model_name):
        # if self.NSFW_filter(prompt) == 'safe':
        pipe = self.load_model_pipe(model_name)
        result = pipe(prompt=prompt)
        # else:
        #    result = ''
        return result

    def generate_video_vg_api(self, prompt, model_name):
        # if self.NSFW_filter(prompt) == 'safe':
        pipe = self.load_model_pipe(model_name)
        result = pipe(prompt=prompt)
        # else:
        #    result = ''
        return result

    def generate_video_vg_museum(self, model_name):
        model_name = model_name.split('_')[1]
        result_list = draw_from_videogen_museum("t2v", model_name)
        video_link = result_list[0]
        prompt = result_list[1]

        return video_link, prompt

    def generate_video_vg_parallel_anony(self, prompt, model_A, model_B):
        # Using list comprehension to get the difference between two lists
        picking_list = [item for item in self.model_vg_list if item not in self.excluding_model_list]
        if model_A == "" and model_B == "":
            model_names = random.sample([model for model in picking_list], 2)
        else:
            model_names = [model_A, model_B]

        with concurrent.futures.ThreadPoolExecutor() as executor:
            futures = [executor.submit(self.generate_video_vg, prompt, model) if model.startswith("videogenhub")
                       else executor.submit(self.generate_video_vg_api, prompt, model) for model in model_names]
            results = [future.result() for future in futures]
        return results[0], results[1], model_names[0], model_names[1]

    def generate_video_vg_museum_parallel_anony(self, model_A, model_B):
        # Using list comprehension to get the difference between two lists
        picking_list = [item for item in self.model_vg_list if item not in self.excluding_model_list]
        #picking_list = [item for item in picking_list if item not in self.desired_model_list]
        if model_A == "" and model_B == "":
            model_names = random.sample([model for model in picking_list], 2)

            #override the random selection
            #model_names[random.choice([0, 1])] = random.choice(self.desired_model_list)
        else:
            model_names = [model_A, model_B]

        with concurrent.futures.ThreadPoolExecutor() as executor:
            model_1 = model_names[0].split('_')[1]
            model_2 = model_names[1].split('_')[1]
            result_list = draw2_from_videogen_museum("t2v", model_1, model_2)
            video_links = result_list[0]
            prompt_list = result_list[1]

        return video_links[0], video_links[1], model_names[0], model_names[1], prompt_list[0]
    
    def generate_video_vg_parallel(self, prompt, model_A, model_B):
        model_names = [model_A, model_B]
        with concurrent.futures.ThreadPoolExecutor() as executor:
            futures = [executor.submit(self.generate_video_vg, prompt, model) if model.startswith("videogenhub")
                       else executor.submit(self.generate_video_vg_api, prompt, model) for model in model_names]
            results = [future.result() for future in futures]
        return results[0], results[1]

    def generate_video_vg_museum_parallel(self, model_A, model_B):
        model_names = [model_A, model_B]
        with concurrent.futures.ThreadPoolExecutor() as executor:
            model_1 = model_A.split('_')[1]
            model_2 = model_B.split('_')[1]
            result_list = draw2_from_videogen_museum("t2v", model_1, model_2)
            video_links = result_list[0]
            prompt_list = result_list[1]
        return video_links[0], video_links[1], prompt_list[0]