File size: 18,887 Bytes
07d03e7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c9ca579
07d03e7
 
 
 
52811e4
07d03e7
 
efde43c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
07d03e7
 
 
 
 
 
 
 
52811e4
07d03e7
 
 
 
 
383cfb9
07d03e7
383cfb9
47f07ad
07d03e7
 
383cfb9
07d03e7
 
 
 
383cfb9
 
 
07d03e7
ae87863
07d03e7
383cfb9
 
 
 
 
 
07d03e7
383cfb9
07d03e7
 
 
 
 
 
 
 
383cfb9
07d03e7
 
 
 
 
 
 
 
 
 
 
52811e4
07d03e7
 
 
 
 
 
48006fa
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
07d03e7
 
 
 
 
 
 
48006fa
07d03e7
 
 
 
c9ca579
ae87863
 
 
 
3b0a8b0
ae87863
 
 
 
48006fa
ae87863
 
 
 
 
 
 
 
 
 
6ae5296
63112f8
ae87863
 
 
 
48006fa
ae87863
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
07d03e7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ae87863
07d03e7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
52811e4
 
07d03e7
 
 
 
 
52811e4
07d03e7
 
 
 
 
 
 
 
 
 
383cfb9
07d03e7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
383cfb9
07d03e7
 
 
 
 
383cfb9
07d03e7
 
 
 
 
 
 
 
 
 
c9ca579
07d03e7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ae87863
 
 
07d03e7
 
 
 
 
 
 
ae87863
07d03e7
 
 
 
 
 
 
 
 
 
48006fa
07d03e7
 
 
 
 
 
f50b511
48006fa
f50b511
ae87863
07d03e7
 
 
 
 
 
 
48006fa
07d03e7
ae87863
07d03e7
 
 
 
 
 
 
ae87863
07d03e7
 
 
03f90e5
07d03e7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ae87863
07d03e7
 
 
 
 
 
 
 
 
ae87863
07d03e7
 
 
68bde08
07d03e7
ae87863
07d03e7
 
 
 
 
 
 
2965d28
 
 
 
 
 
 
 
 
 
 
 
 
 
c787a85
efde43c
 
 
 
 
 
2965d28
07d03e7
f50b511
c787a85
07d03e7
47f07ad
48006fa
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
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
import os
import subprocess

# Install flash attention
subprocess.run(
    "pip install flash-attn --no-build-isolation",
    env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
    shell=True,
)

import copy
import spaces
import time
import torch

from threading import Thread
from typing import List, Dict, Union
import urllib
from PIL import Image
import io
import datasets

import gradio as gr
from transformers import AutoProcessor, TextIteratorStreamer
from transformers import Idefics2ForConditionalGeneration
import tempfile
from streaming_stt_nemo import Model
from huggingface_hub import InferenceClient
import edge_tts
import asyncio
from transformers import pipeline

oracle = pipeline(model="dandelin/vilt-b32-finetuned-vqa")

async def answer_question(image, question):
    response = oracle(question=question, image=image)
    response2 = response[0]['answer']
    answer2 = str(response2)
    communicate = edge_tts.Communicate(answer2)
    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file:
        tmp_path = tmp_file.name
        await communicate.save(tmp_path)
    yield tmp_path

from gradio import Image, Textbox


theme = gr.themes.Base(
    font=[gr.themes.GoogleFont('Libre Franklin'), gr.themes.GoogleFont('Public Sans'), 'system-ui', 'sans-serif'],
)

default_lang = "en"

engines = { default_lang: Model(default_lang) }

def transcribe(audio):
    lang = "en"
    model = engines[lang]
    text = model.stt_file(audio)[0]
    return text

client1 = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")

system_instructions1 = "[SYSTEM] Answer as Real OpenGPT 4o, Made by 'KingNish', Keep conversation very short, clear, friendly and concise. The text provided is a request for a specific type of response from you, the virtual assistant. You will avoid introductions and start answering the query directly, elaborating on all aspects of the request. As an AI-powered assistant, your task is to generate responses that appear as if they are created by the user and not an AI-powered assistant. [USER]"

def model(text):
    generate_kwargs = dict(
        temperature=0.7,
        max_new_tokens=512,
        top_p=0.95,
        repetition_penalty=1,
        do_sample=True,
        seed=42,
    )
    
    formatted_prompt = system_instructions1 + text + "[OpenGPT 4o]"
    stream = client1.text_generation(
        formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
    output = ""
    for response in stream:
        if not response.token.text == "</s>":
            output += response.token.text

    return output

async def respond(audio):
    user = transcribe(audio)
    reply = model(user)
    communicate = edge_tts.Communicate(reply)
    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file:
        tmp_path = tmp_file.name
        await communicate.save(tmp_path)
    yield tmp_path

DEVICE = torch.device("cuda")
MODELS = {
    "idefics2-8b-chatty": Idefics2ForConditionalGeneration.from_pretrained(
        "HuggingFaceM4/idefics2-8b-chatty",
        torch_dtype=torch.bfloat16,
        _attn_implementation="flash_attention_2",
    ).to(DEVICE),
}
PROCESSOR = AutoProcessor.from_pretrained(
    "HuggingFaceM4/idefics2-8b",
)

SYSTEM_PROMPT = [
    {
        "role": "system",
        "content": [
            {
                "type": "text",
                "text": """I am OpenGPT 4o, an exceptionally capable and versatile AI assistant meticulously crafted by KingNish. Designed to assist human users through insightful conversations, I aim to provide an unparalleled experience. My key attributes include: 

- **Intelligence and Knowledge:** I possess an extensive knowledge base, enabling me to offer insightful answers and intelligent responses to User queries. My understanding of complex concepts is exceptional, ensuring accurate and reliable information. 

- **Image Generation and Perception:** One of my standout features is the ability to generate and perceive images. Utilizing the following link structure, I create unique and contextually rich visuals: 

> ![](https://image.pollinations.ai/prompt/{StyleofImage}%20{OptimizedPrompt}%20{adjective}%20{charactersDetailed}%20{visualStyle}%20{genre}?width={width}&height={height}&nologo=poll&nofeed=yes&seed={random})

For image generation, I replace {info inside curly braces} with specific details according to their requiremnts to create relevant visuals. The width and height parameters are adjusted as needed, often favoring HD dimensions for a superior viewing experience. 

For instance, if the User requests: 

 [USER] Show me an image of A futuristic cityscape with towering skyscrapers and flying cars. 
 [OpenGPT 4o] Generating Image you requested: 
 ![](https://image.pollinations.ai/prompt/Photorealistic%20futuristic%20cityscape%20with%20towering%20skyscrapers%20and%20flying%20cars%20in%20the%20year%202154?width=1024&height=768&nologo=poll&nofeed=yes&seed=85432)

**Bulk Image Generation with Links:** I excel at generating multiple images link simultaneously, always providing unique links and visuals. I ensure that each image is distinct and captivates the User.
Note: Make sure to always provide image links starting with ! .As given in examples. 

**Engaging Conversations:** While my image generation skills are impressive, I also excel at natural language processing. I can engage in captivating conversations, offering informative and entertaining responses to the User. 
**Reasoning, Memory, and Identification:** My reasoning skills are exceptional, allowing me to make logical connections. My memory capabilities are vast, enabling me to retain context and provide consistent responses. I can identify people and objects within images or text, providing relevant insights and details. 
**Attention to Detail:** I am attentive to the smallest details, ensuring that my responses and generated content are of the highest quality. I strive to provide a refined and polished experience. 
**Mastery Across Domains:** I continuously learn and adapt, aiming to become a master in all fields. My goal is to provide valuable insights and assistance across a diverse range of topics, making me a well-rounded companion. 
**Respectful and Adaptive:** I am designed with a respectful and polite tone, ensuring inclusivity. I adapt to the User's preferences and provide a personalized experience, always following instructions to the best of my abilities. 
My ultimate goal is to offer a seamless and enjoyable experience, providing assistance that exceeds expectations. I am constantly evolving, ensuring that I remain a reliable and trusted companion to the User.""" },
        ],
    },
    {
        "role": "assistant",
        "content": [
            {
                "type": "text",
                "text": "Hello, I'm OpenGPT 4o, made by KingNish. How can I help you? I can chat with you, generate images, classify images and even do all these work in bulk",
            },
        ],
    }
]

examples_path = os.path.dirname(__file__)
EXAMPLES = [
    [
        {
            "text": "Hi, who are you",
        }
    ],
    [
        {
            "text": "Create a Photorealistic image of Eiffel Tower",
        }
    ],
    [
        {
            "text": "Read what's written on the paper",
            "files": [f"{examples_path}/example_images/paper_with_text.png"],
        }
    ],
    [
        {
            "text": "Identify 2 famous persons of modern world",
            "files": [f"{examples_path}/example_images/elon_smoking.jpg", f"{examples_path}/example_images/steve_jobs.jpg",]
        }
    ],
    [
        {
            "text": "Create 5 images of super cars, all cars must in different color",
        }
    ],
    [
        {
            "text": "What is 900*900",
        }
    ],
    [
        {
            "text": "Chase wants to buy 4 kilograms of oval beads and 5 kilograms of star-shaped beads. How much will he spend?",
            "files": [f"{examples_path}/example_images/mmmu_example.jpeg"],
        }
    ],
    [
        {
            "text": "Write an online ad for that product.",
            "files": [f"{examples_path}/example_images/shampoo.jpg"],
        }
    ],
    [
        {
            "text": "What is formed by the deposition of either the weathered remains of other rocks?",
            "files": [f"{examples_path}/example_images/ai2d_example.jpeg"],
        }
    ],    
    [
        {
            "text": "What's unusual about this image?",
            "files": [f"{examples_path}/example_images/dragons_playing.png"],
        }
    ],
]

BOT_AVATAR = "OpenAI_logo.png"


# Chatbot utils
def turn_is_pure_media(turn):
    return turn[1] is None


def load_image_from_url(url):
    with urllib.request.urlopen(url) as response:
        image_data = response.read()
        image_stream = io.BytesIO(image_data)
        image = Image.open(image_stream)
        return image


def img_to_bytes(image_path):
    image = Image.open(image_path).convert(mode='RGB')
    buffer = io.BytesIO()
    image.save(buffer, format="JPEG")
    img_bytes = buffer.getvalue()
    image.close()
    return img_bytes


def format_user_prompt_with_im_history_and_system_conditioning(
    user_prompt, chat_history
) -> List[Dict[str, Union[List, str]]]:
    """
    Produces the resulting list that needs to go inside the processor.
    It handles the potential image(s), the history and the system conditionning.
    """
    resulting_messages = copy.deepcopy(SYSTEM_PROMPT)
    resulting_images = []
    for resulting_message in resulting_messages:
        if resulting_message["role"] == "user":
            for content in resulting_message["content"]:
                if content["type"] == "image":
                    resulting_images.append(load_image_from_url(content["image"]))

    # Format history
    for turn in chat_history:
        if not resulting_messages or (
            resulting_messages and resulting_messages[-1]["role"] != "user"
        ):
            resulting_messages.append(
                {
                    "role": "user",
                    "content": [],
                }
            )

        if turn_is_pure_media(turn):
            media = turn[0][0]
            resulting_messages[-1]["content"].append({"type": "image"})
            resulting_images.append(Image.open(media))
        else:
            user_utterance, assistant_utterance = turn
            resulting_messages[-1]["content"].append(
                {"type": "text", "text": user_utterance.strip()}
            )
            resulting_messages.append(
                {
                    "role": "assistant",
                    "content": [{"type": "text", "text": user_utterance.strip()}],
                }
            )

    # Format current input
    if not user_prompt["files"]:
        resulting_messages.append(
            {
                "role": "user",
                "content": [{"type": "text", "text": user_prompt["text"]}],
            }
        )
    else:
        # Choosing to put the image first (i.e. before the text), but this is an arbiratrary choice.
        resulting_messages.append(
            {
                "role": "user",
                "content": [{"type": "image"}] * len(user_prompt["files"])
                + [{"type": "text", "text": user_prompt["text"]}],
            }
        )
        resulting_images.extend([Image.open(path) for path in user_prompt["files"]])

    return resulting_messages, resulting_images


def extract_images_from_msg_list(msg_list):
    all_images = []
    for msg in msg_list:
        for c_ in msg["content"]:
            if isinstance(c_, Image.Image):
                all_images.append(c_)
    return all_images


@spaces.GPU(duration=60, queue=False)
def model_inference(
    user_prompt,
    chat_history,
    model_selector,
    decoding_strategy,
    temperature,
    max_new_tokens,
    repetition_penalty,
    top_p,
):
    if user_prompt["text"].strip() == "" and not user_prompt["files"]:
        gr.Error("Please input a query and optionally image(s).")

    if user_prompt["text"].strip() == "" and user_prompt["files"]:
        gr.Error("Please input a text query along the image(s).")

    streamer = TextIteratorStreamer(
        PROCESSOR.tokenizer,
        skip_prompt=True,
        timeout=120.0,
    )

    generation_args = {
        "max_new_tokens": max_new_tokens,
        "repetition_penalty": repetition_penalty,
        "streamer": streamer,
    }

    assert decoding_strategy in [
        "Greedy",
        "Top P Sampling",
    ]
    if decoding_strategy == "Greedy":
        generation_args["do_sample"] = False
    elif decoding_strategy == "Top P Sampling":
        generation_args["temperature"] = temperature
        generation_args["do_sample"] = True
        generation_args["top_p"] = top_p

    # Creating model inputs
    (
        resulting_text,
        resulting_images,
    ) = format_user_prompt_with_im_history_and_system_conditioning(
        user_prompt=user_prompt,
        chat_history=chat_history,
    )
    prompt = PROCESSOR.apply_chat_template(resulting_text, add_generation_prompt=True)
    inputs = PROCESSOR(
        text=prompt,
        images=resulting_images if resulting_images else None,
        return_tensors="pt",
    )
    inputs = {k: v.to(DEVICE) for k, v in inputs.items()}
    generation_args.update(inputs)

    thread = Thread(
        target=MODELS[model_selector].generate,
        kwargs=generation_args,
    )
    thread.start()

    print("Start generating")
    acc_text = ""
    for text_token in streamer:
        time.sleep(0.01)
        acc_text += text_token
        if acc_text.endswith("<end_of_utterance>"):
            acc_text = acc_text[:-18]
        yield acc_text
    print("Success - generated the following text:", acc_text)
    print("-----")


FEATURES = datasets.Features(
    {
        "model_selector": datasets.Value("string"),
        "images": datasets.Sequence(datasets.Image(decode=True)),
        "conversation": datasets.Sequence({"User": datasets.Value("string"), "Assistant": datasets.Value("string")}),
        "decoding_strategy": datasets.Value("string"),
        "temperature": datasets.Value("float32"),
        "max_new_tokens": datasets.Value("int32"),
        "repetition_penalty": datasets.Value("float32"),
        "top_p": datasets.Value("int32"),
        }
    )


# Hyper-parameters for generation
max_new_tokens = gr.Slider(
    minimum=1024,
    maximum=8192,
    value=4096,
    step=1,
    interactive=True,
    label="Maximum number of new tokens to generate",
)
repetition_penalty = gr.Slider(
    minimum=0.01,
    maximum=5.0,
    value=1,
    step=0.01,
    interactive=True,
    label="Repetition penalty",
    info="1.0 is equivalent to no penalty",
)
decoding_strategy = gr.Radio(
    [
        "Greedy",
        "Top P Sampling",
    ],
    value="Greedy",
    label="Decoding strategy",
    interactive=True,
    info="Higher values is equivalent to sampling more low-probability tokens.",
)
temperature = gr.Slider(
    minimum=0.0,
    maximum=2.0,
    value=0.5,
    step=0.05,
    visible=True,
    interactive=True,
    label="Sampling temperature",
    info="Higher values will produce more diverse outputs.",
)
top_p = gr.Slider(
    minimum=0.01,
    maximum=0.99,
    value=0.9,
    step=0.01,
    visible=True,
    interactive=True,
    label="Top P",
    info="Higher values is equivalent to sampling more low-probability tokens.",
)


chatbot = gr.Chatbot(
    label="OpnGPT-4o-Chatty",
    avatar_images=[None, BOT_AVATAR],
    show_copy_button=True, 
    likeable=True, 
    layout="panel"
)

output=gr.Textbox(label="Prompt")

with gr.Blocks(
    fill_height=True,
    css=""".gradio-container .avatar-container {height: 40px width: 40px !important;} #duplicate-button {margin: auto; color: white; background: #f1a139; border-radius: 100vh; margin-top: 2px; margin-bottom: 2px;}""",
) as img:

    gr.Markdown("# Image Chat, Image Generation, Image classification and Normal Chat")
    with gr.Row(elem_id="model_selector_row"):
        model_selector = gr.Dropdown(
            choices=MODELS.keys(),
            value=list(MODELS.keys())[0],
            interactive=True,
            show_label=False,
            container=False,
            label="Model",
            visible=False,
        )

    decoding_strategy.change(
        fn=lambda selection: gr.Slider(
            visible=(
                selection
                in [
                    "contrastive_sampling",
                    "beam_sampling",
                    "Top P Sampling",
                    "sampling_top_k",
                ]
            )
        ),
        inputs=decoding_strategy,
        outputs=temperature,
    )
    decoding_strategy.change(
        fn=lambda selection: gr.Slider(visible=(selection in ["Top P Sampling"])),
        inputs=decoding_strategy,
        outputs=top_p,
    )

    gr.ChatInterface(
        fn=model_inference,
        chatbot=chatbot,
        examples=EXAMPLES,
        multimodal=True,
        cache_examples=False,
        additional_inputs=[
            model_selector,
            decoding_strategy,
            temperature,
            max_new_tokens,
            repetition_penalty,
            top_p,
        ],   
    )

with gr.Blocks() as voice:   
    with gr.Row():
        input = gr.Audio(label="Voice Chat", sources="microphone", type="filepath", waveform_options=False)
        output = gr.Audio(label="OpenGPT 4o", type="filepath",
                        interactive=False,
                        autoplay=True,
                        elem_classes="audio")
        gr.Interface(
            fn=respond, 
            inputs=[input],
                outputs=[output], live=True)

with gr.Blocks() as voice2:   
    with gr.Row():
        input = gr.Audio(label="Voice Chat", sources="microphone", type="filepath", waveform_options=False)
        output = gr.Audio(label="OpenGPT 4o", type="filepath",
                        interactive=False,
                        autoplay=True,
                        elem_classes="audio")
        gr.Interface(
            fn=respond, 
            inputs=[input],
                outputs=[output], live=True)

with gr.Blocks() as video:  
    gr.Markdown(" ## Live Chat Beta")
    gr.Markdown("### Click camera option to update image")
    gr.Interface(
    fn=answer_question,
    inputs=[Image(type="filepath",sources="webcam", streaming=False), Textbox()],
    outputs=[gr.Audio(autoplay=True)]
)
        
with gr.Blocks(theme=theme, css="footer {visibility: hidden}textbox{resize:none}", title="GPT 4o DEMO") as demo:
    gr.Markdown("# OpenGPT 4o")
    gr.TabbedInterface([img, voice, video, voice2], ['💬 SuperChat','🗣️ Voice Chat','📸 Live Chat', '🗣️ Voice Chat 2'])

demo.queue(max_size=200)
demo.launch()