File size: 9,963 Bytes
9aa6aea
 
 
 
 
 
 
 
562648a
9aa6aea
 
 
 
 
 
 
 
 
 
 
 
 
 
ce50c4e
9aa6aea
 
6a25365
9aa6aea
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
862c154
9aa6aea
 
862c154
9aa6aea
 
 
 
862c154
9aa6aea
 
 
 
 
505f63b
9aa6aea
862c154
9aa6aea
 
 
6a25365
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9aa6aea
 
 
 
 
 
 
 
 
 
 
cd9f1f6
9aa6aea
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6a25365
 
 
 
 
 
 
 
 
 
 
 
562648a
6a25365
 
 
 
 
 
 
9aa6aea
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6a25365
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9aa6aea
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
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
# -*- coding: utf-8 -*-

# ===================================================
#
#    Author        : Fan Zhang
#    Email         : zhangfan@baai.ac.cn
#    Institute     : Beijing Academy of Artificial Intelligence (BAAI)
#    Create On     : 2023-12-11 15:35
#    Last Modified : 2023-12-20 16:39
#    File Name     : generation_frontend.py
#    Description   :
#
# ===================================================

import base64
import json
import io
import time
from PIL import Image
import requests

import gradio as gr

from .constants import EVA_IMAGE_SIZE
from .meta import ConvMeta, Role, DataMeta
from .utils import frontend_logger as logging
from .constants import LICENSE, TERM_OF_USE

CONTROLLER_URL = ""

def submit(
    meta,
    enable_grd,
    left,
    top,
    right,
    bottom,
    image,
    text,
):
    if meta is None:
        meta = ConvMeta()

    meta.pop_error()
    if meta.has_gen:
        meta.clear()

    if enable_grd:
        if text == "" and image is None:
            logging.info(f"{meta.log_id}: invalid input: no valid data for grounding input")
            gr.Error("text or image must be given if enable grounding generation")
            return meta.format_chatbot(), meta, False, 0, 0, EVA_IMAGE_SIZE, EVA_IMAGE_SIZE, None, ""

        meta.append(Role.USER, DataMeta.build(text=text, image=image, coordinate=[left, top, right, bottom]))
    elif image is not None and text != "":
        logging.info(f"{meta.log_id}: invalid input: give text and image simultaneously for single modality input")
        gr.Error("Do not submit text and image data at the same time!!!")
        return meta.format_chatbot(), meta, False, 0, 0, EVA_IMAGE_SIZE, EVA_IMAGE_SIZE, None, ""
    elif image is not None:
        meta.append(Role.USER, DataMeta.build(image=image))
    elif text != "":
        meta.append(Role.USER, DataMeta.build(text=text))
    return meta.format_chatbot(), meta, False, 0, 0, EVA_IMAGE_SIZE, EVA_IMAGE_SIZE, None, ""


def clear_history(meta):
    if meta is None:
        meta = ConvMeta()
    meta.clear()
    return meta.format_chatbot(), meta


def generate(meta, classifier_free_guidance, steps):
    if meta is None:
        meta = ConvMeta()

    meta.pop_error()
    meta.pop()
    prompt = meta.format_prompt()

    prompt_list, image_list = [], {}
    for idx, p in enumerate(prompt):
        if isinstance(p, Image.Image):
            key = f"[<IMAGE{idx}>]"
            prompt_list.append(["IMAGE", key])

            buf = io.BytesIO()
            p.save(buf, format="PNG")
            image_list[key] = (key, io.BytesIO(buf.getvalue()), "image/png")
        else:
            prompt_list.append(["TEXT", p])


    if len(image_list) == 0:
        image_list = None

    logging.info(f"{meta.log_id}: construct generation reqeust with prompt {prompt_list}")

    t0 = time.time()
    try:
        rsp = requests.post(
            CONTROLLER_URL + "/v1/mmg",
            files=image_list,
            data={
                "log_id": meta.log_id,
                "prompt": json.dumps(prompt_list),
                "classifier_free_guidance": classifier_free_guidance,
                "steps": steps,
            },
        )
    except Exception as ex:
        rsp = requests.Response()
        rsp.status_code = 1099
        rsp._content = str(ex).encode()
    t1 = time.time()

    logging.info(f"{meta.log_id}: get response with status code: {rsp.status_code}, time: {(t1-t0)*1000:.3f}ms")

    if rsp.status_code == requests.codes.ok:
        content = json.loads(rsp.text)
        if content["code"] == 0:
            image = Image.open(io.BytesIO(base64.b64decode(content["data"])))
            meta.append(Role.ASSISTANT, DataMeta.build(image=image, resize=False))
        else:
            meta.append(Role.ASSISTANT, DataMeta.build(text=f"GENERATE FAILED: {content['data']}", is_error=True))
    else:
        meta.append(Role.ASSISTANT, DataMeta.build(text=f"GENERATE FAILED: http failed with code {rsp.status_code}, msg: {rsp.text}", is_error=True))

    return meta.format_chatbot(), meta

def push_examples(examples, meta):
    if meta is None:
        meta = ConvMeta()

    meta.clear()

    if len(examples) == 1:
        prompt, = examples
        meta.append(Role.USER, DataMeta.build(text=prompt))
    elif len(examples) == 2:
        image, prompt = examples
        meta.append(Role.USER, DataMeta.build(image=Image.open(image)))
        meta.append(Role.USER, DataMeta.build(text=prompt))

    return meta.format_chatbot(), meta


def build_generation(args):
    global CONTROLLER_URL
    CONTROLLER_URL = args.controller_url

    with gr.Blocks(title="Emu", theme=gr.themes.Default(primary_hue="blue", secondary_hue="blue")) as demo:
        state = gr.State()

        with gr.Row():
            with gr.Column(scale=2):
                with gr.Row():
                    imagebox = gr.Image(type="pil")

                with gr.Row():
                    with gr.Accordion("Grounding Parameters", open=True, visible=True) as grounding_row:
                        enable_grd = gr.Checkbox(label="Enable")
                        left = gr.Slider(minimum=0, maximum=EVA_IMAGE_SIZE, value=0, step=1, interactive=True, label="left")
                        top = gr.Slider(minimum=0, maximum=EVA_IMAGE_SIZE, value=0, step=1, interactive=True, label="top")
                        right = gr.Slider(minimum=0, maximum=EVA_IMAGE_SIZE, value=EVA_IMAGE_SIZE, step=1, interactive=True, label="right")
                        bottom = gr.Slider(minimum=0, maximum=EVA_IMAGE_SIZE, value=EVA_IMAGE_SIZE, step=1, interactive=True, label="bottom")

                with gr.Row():
                    with gr.Accordion("Diffusion Parameters", open=True, visible=True) as parameters_row:
                        cfg = gr.Slider(minimum=1, maximum=30, value=3, step=0.5, interactive=True, label="classifier free guidance")
                        steps = gr.Slider(minimum=1, maximum=100, value=50, step=1, interactive=True, label="steps")

            with gr.Column(scale=6):
                chatbot = gr.Chatbot(
                    elem_id="chatbot",
                    label="Emu Chatbot",
                    visible=True,
                    height=720,
                )

                with gr.Row():
                    with gr.Column(scale=8):
                        textbox = gr.Textbox(
                            show_label=False,
                            placeholder="Enter text and add to prompt",
                            visible=True,
                            container=False,
                        )

                    with gr.Column(scale=1, min_width=60):
                        add_btn = gr.Button(value="Add")

                with gr.Row(visible=True) as button_row:
                    # upvote_btn = gr.Button(value="πŸ‘ Upvote", interactive=False)
                    # downvote_btn = gr.Button(value="πŸ‘Ž Downvote", interactive=False)
                    # regenerate_btn = gr.Button(value="πŸ”„ Regenerate", interactive=False)
                    clear_btn = gr.Button(value="πŸ—‘οΈ Clear History")
                    generate_btn = gr.Button(value="Generate")

        with gr.Row():
            examples_t2i = gr.Dataset(components=[gr.Textbox(visible=False)],
                label="Examples",
                samples=[
                    ["impressionist painting of an astronaut in a jungle"],
                    ["A poised woman with short, curly hair and a warm smile, dressed in elegant attire, standing in front of a historic stone bridge in a serene park at sunset."],
                ],
            )
        with gr.Row():
            examples_it2i = gr.Dataset(components=[gr.Image(type="pil", visible=False), gr.Textbox(visible=False)],
                samples=[
                    ["./examples/dog2.jpg", "wearing a red hat on the beach."],
                    ["./examples/balloon.webp", "floats above theforest"],
                ],
            )


        gr.Markdown(TERM_OF_USE)
        gr.Markdown(LICENSE)

        clear_btn.click(clear_history, inputs=state, outputs=[chatbot, state])

        textbox.submit(
            submit,
            inputs=[
                state,
                enable_grd,
                left,
                top,
                right,
                bottom,
                imagebox,
                textbox,
            ],
            outputs=[
                chatbot,
                state,
                enable_grd,
                left,
                top,
                right,
                bottom,
                imagebox,
                textbox,
            ],
        )

        add_btn.click(
            submit,
            inputs=[
                state,
                enable_grd,
                left,
                top,
                right,
                bottom,
                imagebox,
                textbox,
            ],
            outputs=[
                chatbot,
                state,
                enable_grd,
                left,
                top,
                right,
                bottom,
                imagebox,
                textbox,
            ],
        )

        generate_btn.click(
            generate,
            inputs=[
                state,
                cfg,
                steps,
            ],
            outputs=[
                chatbot,
                state,
            ]
        )

        examples_t2i.click(
            push_examples,
            inputs=[
                examples_t2i,
                state,
            ],
            outputs=[
                chatbot,
                state,
            ]
        )
        examples_it2i.click(
            push_examples,
            inputs=[
                examples_it2i,
                state,
            ],
            outputs=[
                chatbot,
                state,
            ]
        )


    return demo