File size: 1,830 Bytes
b73e8d0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from transformers import AutoModelForCausalLM, AutoTokenizer
from transformers import BlenderbotForConditionalGeneration
import torch
imprt gradio as gr

# model_name = "facebook/blenderbot-400M-distill"
model_name = "microsoft/DialoGPT-medium"
chat_tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

def conversation(user_input, chat_history=[]):
    user_input_ids = chat_tokenizer(user_input+chat_tokenizer.eos_token, return_tensors="pt").input_ids

    # maintain history in the tensor
    chatbot_input_ids = torch.cat([torch.LongTensor(chat_hostory), user_input_ids], dim=-1)

    # ger response
    chat_history = model.generate(chatbot_input_ids, max_length=1000, pad_token_id=chat_tokenizer.eos_token_id).tolist()
    print(chat_history)

    response = chat_tokeniser.decode(chat_history[0]).split("<|endoftext|>")
    print("Starting to print response")
    print(response)

    # html for display
    html = "<div class='mychatbot'>"
    for x, msg in enumerate(response):
        if x%2 !=0:
            msg = "ChatBot:" + msg
            class = "bot"
        else:
            class = "user"

        print("Value of x:")
        print(x)
        print("Message:")
        print(msg)
        html += "<div class='msg {}'></div>".format(class, msg)
    html += "</div>"
    print(html)
    return html, chat_history

# UX
css = """
.mychat {display:flex;flex-direction:column}
.msg {padding:5px;margin-bottom:5px;border-radius:5px;width:75%}
.msg.user {background-color:lightblue;color:white}
.msg.bot {background-color:orange;color:white,align-self:self-end}
.footer
"""
in_text = gr.inputs.Textbox(placeholder="Let's start a chat...")
gr.Interface(fn=conversation, theme="default", inputs=[in_text, "state"], outputs=["html", "state"], css=css).launch()