File size: 11,645 Bytes
4a8ac8b
 
 
 
 
 
 
 
7e5261e
 
 
 
 
 
 
 
 
 
 
4e7357a
7e5261e
 
 
6cf979c
983867e
 
6cf979c
 
 
 
 
 
983867e
6cf979c
 
 
983867e
6cf979c
06bea5a
7e5261e
4a8ac8b
 
db07900
4a8ac8b
 
 
 
7ea7941
4a8ac8b
 
8276bc3
7ea7941
 
4a8ac8b
8c8d4ad
7ea7941
 
 
4a8ac8b
 
 
 
 
 
 
 
8c8d4ad
4a8ac8b
 
 
 
 
8276bc3
4a8ac8b
 
 
7ea7941
8c8d4ad
7ea7941
 
 
 
 
666a2c5
7ea7941
4a8ac8b
f105330
7ea7941
4e7357a
06bea5a
 
 
 
 
 
 
 
 
 
 
983867e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
06bea5a
983867e
 
 
 
06bea5a
983867e
 
06bea5a
 
 
 
 
 
 
 
 
 
 
983867e
 
 
 
 
 
 
 
 
 
 
 
 
4a8ac8b
 
 
 
 
7ea7941
4a8ac8b
 
 
 
 
 
7ea7941
 
 
 
 
 
 
 
 
 
 
 
 
983867e
 
7ea7941
 
 
4a8ac8b
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
import gradio as gr
from huggingface_hub import InferenceClient
import json
import uuid
from PIL import Image
from bs4 import BeautifulSoup
import requests
import random
from transformers import LlavaProcessor, LlavaForConditionalGeneration, TextIteratorStreamer
from threading import Thread
import re
import time 
import torch
import cv2

model_id = "llava-hf/llava-interleave-qwen-0.5b-hf"

processor = LlavaProcessor.from_pretrained(model_id)

model = LlavaForConditionalGeneration.from_pretrained(model_id, low_cpu_mem_usage=True)
model.to("cpu")


def llava(message, history):
    if message["files"]:
        image = user_prompt["files"][0]    
    else:
        for hist in history:
            if type(hist[0])==tuple:
                image = hist[0][0]
        
    txt = user_prompt["text"]
        
    gr.Info("Analyzing image")
    image = Image.open(image).convert("RGB")
    prompt = f"<|im_start|>user <image>\n{user_prompt}<|im_end|><|im_start|>assistant"
    
    inputs = processor(prompt, image, return_tensors="pt")
    return inputs

def extract_text_from_webpage(html_content):
    soup = BeautifulSoup(html_content, 'html.parser')
    for tag in soup(["script", "style", "header", "footer"]):
        tag.extract()
    return soup.get_text(strip=True)

def search(query):
    term = query
    start = 0
    all_results = []
    max_chars_per_page = 8000
    with requests.Session() as session:
        resp = session.get(
            url="https://www.google.com/search",
            headers={"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:109.0) Gecko/20100101 Firefox/111.0"},
            params={"q": term, "num": 3, "udm": 14},
            timeout=5,
            verify=None,
        )
        resp.raise_for_status()
        soup = BeautifulSoup(resp.text, "html.parser")
        result_block = soup.find_all("div", attrs={"class": "g"})
        for result in result_block:
            link = result.find("a", href=True)
            link = link["href"]
            try:
                webpage = session.get(link, headers={"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:109.0) Gecko/20100101 Firefox/111.0"}, timeout=5, verify=False)
                webpage.raise_for_status()
                visible_text = extract_text_from_webpage(webpage.text)
                if len(visible_text) > max_chars_per_page:
                    visible_text = visible_text[:max_chars_per_page]
                all_results.append({"link": link, "text": visible_text})
            except requests.exceptions.RequestException:
                all_results.append({"link": link, "text": None})
    return all_results

# Initialize inference clients for different models
client_gemma = InferenceClient("google/gemma-1.1-7b-it")
client_mixtral = InferenceClient("NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO")
client_llama = InferenceClient("meta-llama/Meta-Llama-3-8B-Instruct")

# Define the main chat function
def respond(message, history):
    func_caller = []
    vqa = ""

    user_prompt = message
    # Handle image processing
    if message["files"]:
        inputs = llava(message, history)
        streamer = TextIteratorStreamer(processor, skip_prompt=True, **{"skip_special_tokens": True})
        generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=1024)

        thread = Thread(target=model.generate, kwargs=generation_kwargs)
        thread.start()
    
        buffer = ""
        for new_text in streamer:
            buffer += new_text
            yield buffer
    else:
        functions_metadata = [
            {"type": "function", "function": {"name": "web_search", "description": "Search query on google", "parameters": {"type": "object", "properties": {"query": {"type": "string", "description": "web search query"}}, "required": ["query"]}}},
            {"type": "function", "function": {"name": "general_query", "description": "Reply general query of USER", "parameters": {"type": "object", "properties": {"prompt": {"type": "string", "description": "A detailed prompt"}}, "required": ["prompt"]}}},
            {"type": "function", "function": {"name": "image_generation", "description": "Generate image for user", "parameters": {"type": "object", "properties": {"query": {"type": "string", "description": "image generation prompt"}, "number_of_image": {"type": "integer", "description": "number of images to generate"}}, "required": ["query"]}}},
            {"type": "function", "function": {"name": "image_qna", "description": "Answer question asked by user related to image", "parameters": {"type": "object", "properties": {"query": {"type": "string", "description": "Question by user"}}, "required": ["query"]}}},
        ]
    
        message_text = message["text"]
        func_caller.append({"role": "user", "content": f'[SYSTEM]You are a helpful assistant. You have access to the following functions: \n {str(functions_metadata)}\n\nTo use these functions respond with:\n<functioncall> {{ "name": "function_name", "arguments": {{ "arg_1": "value_1", "arg_1": "value_1", ... }} }}  </functioncall>  [USER] {message_text} {vqa}'})
    
        response = client_gemma.chat_completion(func_caller, max_tokens=150)
        response = str(response)
        try:
            response = response[int(response.find("{")):int(response.index("</"))]
        except:
            print("A error occured")
        response = response.replace("\\n", "")
        response = response.replace("\\'", "'")
        response = response.replace('\\"', '"')
        print(f"\n{response}")
    
        func_caller.append({"role": "assistant", "content": f"<functioncall>{response}</functioncall>"})
    
        try:
            json_data = json.loads(str(response))
            if json_data["name"] == "web_search":
                query = json_data["arguments"]["query"]
                gr.Info("Searching Web")
                web_results = search(query)
                gr.Info("Extracting relevant Info")
                web2 = ' '.join([f"Link: {res['link']}\nText: {res['text']}\n\n" for res in web_results])
                messages = f"<|im_start|>system\nYou are OpenGPT 4o mini a helpful assistant made by KingNish. You are provided with WEB results from which you can find informations to answer users query in Structured and More better way. You do not say Unnecesarry things Only say thing which is important and relevant. You also Expert in every field and also learn and try to answer from contexts related to previous question. Try your best to give best response possible to user. You also try to show emotions using Emojis and reply like human, use short forms, friendly tone and emotions.<|im_end|>"
                for msg in history:
                    messages += f"\n<|im_start|>user\n{str(msg[0])}<|im_end|>"
                    messages += f"\n<|im_start|>assistant\n{str(msg[1])}<|im_end|>"
                messages+=f"\n<|im_start|>user\n{message_text} {vqa}<|im_end|>\n<|im_start|>web_result\n{web2}<|im_end|>\n<|im_start|>assistant\n"
                stream = client_mixtral.text_generation(messages, max_new_tokens=2000, do_sample=True, stream=True, details=True, return_full_text=False)
                output = ""
                for response in stream:
                    if not response.token.text == "<|im_end|>":
                        output += response.token.text
                        yield output
            elif json_data["name"] == "image_generation":
                query = json_data["arguments"]["query"]
                gr.Info("Generating Image, Please wait 10 sec...")
                seed = random.randint(1, 99999)
                query = query.replace(" ", "%20")
                image = f"![](https://image.pollinations.ai/prompt/{query}?seed={seed})"
                yield image
                time.sleep(8)
                gr.Info("We are going to Update Our Image Generation Engine to more powerful ones in Next Update. ThankYou")
            elif json_data["name"] == "image_qna":
                inputs = llava(message, history)
                streamer = TextIteratorStreamer(processor, skip_prompt=True, **{"skip_special_tokens": True})
                generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=1024)

                thread = Thread(target=model.generate, kwargs=generation_kwargs)
                thread.start()
    
                buffer = ""
                for new_text in streamer:
                    buffer += new_text
                    yield buffer
            else:
                messages = f"<|start_header_id|>system\nYou are OpenGPT 4o mini a helpful assistant made by KingNish. You answers users query like human friend. You are also Expert in every field and also learn and try to answer from contexts related to previous question. Try your best to give best response possible to user. You also try to show emotions using Emojis and reply like human, use short forms, friendly tone and emotions.<|end_header_id|>"
                for msg in history:
                    messages += f"\n<|start_header_id|>user\n{str(msg[0])}<|end_header_id|>"
                    messages += f"\n<|start_header_id|>assistant\n{str(msg[1])}<|end_header_id|>"
                messages+=f"\n<|start_header_id|>user\n{message_text} {vqa}<|end_header_id|>\n<|start_header_id|>assistant\n"
                stream = client_llama.text_generation(messages, max_new_tokens=2000, do_sample=True, stream=True, details=True, return_full_text=False)
                output = ""
                for response in stream:
                    if not response.token.text == "<|eot_id|>":
                        output += response.token.text
                        yield output
        except:
            messages = f"<|start_header_id|>system\nYou are OpenGPT 4o mini a helpful assistant made by KingNish. You answers users query like human friend. You are also Expert in every field and also learn and try to answer from contexts related to previous question. Try your best to give best response possible to user. You also try to show emotions using Emojis and reply like human, use short forms, friendly tone and emotions.<|end_header_id|>"
            for msg in history:
                messages += f"\n<|start_header_id|>user\n{str(msg[0])}<|end_header_id|>"
                messages += f"\n<|start_header_id|>assistant\n{str(msg[1])}<|end_header_id|>"
            messages+=f"\n<|start_header_id|>user\n{message_text} {vqa}<|end_header_id|>\n<|start_header_id|>assistant\n"
            stream = client_llama.text_generation(messages, max_new_tokens=2000, do_sample=True, stream=True, details=True, return_full_text=False)
            output = ""
            for response in stream:
                if not response.token.text == "<|eot_id|>":
                    output += response.token.text
                    yield output

# Create the Gradio interface
demo = gr.ChatInterface(
    fn=respond,
    chatbot=gr.Chatbot(show_copy_button=True, likeable=True, layout="panel"),
    title="OpenGPT 4o mini",
    textbox=gr.MultimodalTextbox(),
    multimodal=True,
    concurrency_limit=20,
    examples=[
        {"text": "Hy, who are you?",},
        {"text": "What's the current price of Bitcoin",},
        {"text": "Create A Beautiful image of Effiel Tower at Night",},
        {"text": "Write me a Python function to calculate the first 10 digits of the fibonacci sequence.",},
        {"text": "What's the colour of car in given image", "files": ["./car1.png"]},
        {"text": "Read what's written on paper", "files": ["./paper_with_text.png"]},
    ],
    cache_examples=False,
)
demo.launch()