Spaces:
No application file
No application file
Commit ·
ac1b258
1
Parent(s): 0c57adf
Update App.py
Browse files
App.py
CHANGED
|
@@ -1,7 +1,127 @@
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import gradio as gr
|
|
|
|
|
|
|
| 2 |
|
| 3 |
-
def
|
| 4 |
-
return "Hello " + name + "!!"
|
| 5 |
|
| 6 |
-
|
| 7 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# app.py
|
| 2 |
+
import random
|
| 3 |
+
import torch
|
| 4 |
import gradio as gr
|
| 5 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 6 |
+
from util_funcs import getLengthParam, calcAnswerLengthByProbability, cropContext
|
| 7 |
|
| 8 |
+
def chat_function(Message, History): # model, tokenizer
|
|
|
|
| 9 |
|
| 10 |
+
input_user = Message
|
| 11 |
+
|
| 12 |
+
history = History or []
|
| 13 |
+
|
| 14 |
+
chat_history_ids = torch.zeros((1, 0), dtype=torch.int) if history == [] else torch.tensor(history[-1][2], dtype=torch.long)
|
| 15 |
+
|
| 16 |
+
# encode the new user input, add parameters and return a tensor in Pytorch
|
| 17 |
+
lengthId = getLengthParam(input_user, tokenizer)
|
| 18 |
+
new_user_input_ids = tokenizer.encode(f"|0|{lengthId}|" \
|
| 19 |
+
+ input_user + tokenizer.eos_token, return_tensors="pt")
|
| 20 |
+
# append the new user input tokens to the chat history
|
| 21 |
+
chat_history_ids = torch.cat([chat_history_ids, new_user_input_ids], dim=-1)
|
| 22 |
+
|
| 23 |
+
# Длину ожидаемой фразы мы рассчитаем на основании последнего инпута
|
| 24 |
+
# Например, я не люблю когда на мой длинный ответ отвечают короткой фразой
|
| 25 |
+
# Но пойдем через вероятности:
|
| 26 |
+
# при длинном инпуте 60% что будет длинный ответ (3), 30% что средний (2), 10% что короткий (1)
|
| 27 |
+
# при среднем инпуте 50% что ответ будет средний (2), и по 25% на оба остальных случая
|
| 28 |
+
# при коротком инпуте 50% что ответ будет короткий (1), 30% что средний (2) и 20% что длинный (3)
|
| 29 |
+
# см. функцию calcAnswerLengthByProbability()
|
| 30 |
+
|
| 31 |
+
next_len = calcAnswerLengthByProbability(lengthId)
|
| 32 |
+
|
| 33 |
+
# encode the new user input, add parameters and return a tensor in Pytorch
|
| 34 |
+
new_user_input_ids = tokenizer.encode(f"|1|{next_len}|", return_tensors="pt")
|
| 35 |
+
|
| 36 |
+
# append the new user input tokens to the chat history
|
| 37 |
+
chat_history_ids = torch.cat([chat_history_ids, new_user_input_ids], dim=-1)
|
| 38 |
+
|
| 39 |
+
chat_history_ids = cropContext(chat_history_ids, 10)
|
| 40 |
+
|
| 41 |
+
print(tokenizer.decode(chat_history_ids[-1]))# uncomment for debug
|
| 42 |
+
|
| 43 |
+
# save previous len
|
| 44 |
+
input_len = chat_history_ids.shape[-1]
|
| 45 |
+
# generated a response; PS you can read about the parameters at hf.co/blog/how-to-generate
|
| 46 |
+
|
| 47 |
+
temperature = 0.6
|
| 48 |
+
|
| 49 |
+
# Обрезаем контекст до нужной длины с конца
|
| 50 |
+
|
| 51 |
+
# Создадим копию изначальных данных на случай если придется перегенерировать ответ
|
| 52 |
+
chat_history_ids_initial = chat_history_ids
|
| 53 |
+
|
| 54 |
+
while True:
|
| 55 |
+
chat_history_ids = model.generate(
|
| 56 |
+
chat_history_ids,
|
| 57 |
+
num_return_sequences=1,
|
| 58 |
+
min_length = 2,
|
| 59 |
+
max_length=512,
|
| 60 |
+
no_repeat_ngram_size=3,
|
| 61 |
+
do_sample=True,
|
| 62 |
+
top_k=50,
|
| 63 |
+
top_p=0.9,
|
| 64 |
+
temperature = temperature,
|
| 65 |
+
mask_token_id=tokenizer.mask_token_id,
|
| 66 |
+
eos_token_id=tokenizer.eos_token_id,
|
| 67 |
+
unk_token_id=tokenizer.unk_token_id,
|
| 68 |
+
pad_token_id=tokenizer.pad_token_id,
|
| 69 |
+
device='cpu'
|
| 70 |
+
)
|
| 71 |
+
|
| 72 |
+
answer = tokenizer.decode(chat_history_ids[:, input_len:][0], skip_special_tokens=True)
|
| 73 |
+
|
| 74 |
+
if (len(answer) > 0 and answer[-1] != ',' and answer[-1] != ':'):
|
| 75 |
+
break
|
| 76 |
+
else:
|
| 77 |
+
if (temperature <= 0.1):
|
| 78 |
+
temperature -= 0.1
|
| 79 |
+
|
| 80 |
+
# Случай когда надо перегенерировать ответ наступил, берем изначальный тензор
|
| 81 |
+
chat_history_ids = chat_history_ids_initial
|
| 82 |
+
|
| 83 |
+
history.append((input_user, answer, chat_history_ids.tolist()))
|
| 84 |
+
html = "<div class='chatbot'>"
|
| 85 |
+
for user_msg, resp_msg, _ in history:
|
| 86 |
+
if user_msg != '-':
|
| 87 |
+
html += f"<div class='user_msg'>{user_msg}</div>"
|
| 88 |
+
if resp_msg != '-':
|
| 89 |
+
html += f"<div class='resp_msg'>{resp_msg}</div>"
|
| 90 |
+
html += "</div>"
|
| 91 |
+
return html, history
|
| 92 |
+
|
| 93 |
+
# Download checkpoint:
|
| 94 |
+
|
| 95 |
+
checkpoint = "avorozhko/ruDialoGpt3-medium-finetuned-context"
|
| 96 |
+
tokenizer = AutoTokenizer.from_pretrained(checkpoint)
|
| 97 |
+
model = AutoModelForCausalLM.from_pretrained(checkpoint)
|
| 98 |
+
model = model.eval()
|
| 99 |
+
|
| 100 |
+
# Gradio
|
| 101 |
+
title = "Чат-бот для поднятия настроения"
|
| 102 |
+
description = """
|
| 103 |
+
Данный бот постарается поднять вам настроение, так как он знает 26700 анекдотов.
|
| 104 |
+
Но чувство юмора у него весьма специфичное.
|
| 105 |
+
Бот не знает матерных слов и откровенных пошлостей, но кто такой Вовочка и Поручик Ржевский знает )
|
| 106 |
+
"""
|
| 107 |
+
article = "<p style='text-align: center'><a href='https://huggingface.co/avorozhko/ruDialoGpt3-medium-finetuned-context'>Бот на основе дообученной GPT-3</a></p>"
|
| 108 |
+
|
| 109 |
+
iface = gr.Interface(fn=chat_function,
|
| 110 |
+
inputs=[gr.inputs.Textbox(lines=3, placeholder="Что вы хотите сказать боту..."), "state"],
|
| 111 |
+
outputs=["html", "state"],
|
| 112 |
+
title=title, description=description, article=article,
|
| 113 |
+
theme='dark-grass',
|
| 114 |
+
css= """
|
| 115 |
+
.chatbox {display:flex;flex-direction:column}
|
| 116 |
+
.user_msg, .resp_msg {padding:4px;margin-bottom:4px;border-radius:4px;width:80%}
|
| 117 |
+
.user_msg {background-color:#1e4282;color:white;align-self:start}
|
| 118 |
+
.resp_msg {background-color:#552a2a;align-self:self-end}
|
| 119 |
+
.panels.unaligned {flex-direction: column !important;align-items: initial!important;}
|
| 120 |
+
.panels.unaligned :last-child {order: -1 !important;}
|
| 121 |
+
""",
|
| 122 |
+
allow_screenshot=False,
|
| 123 |
+
allow_flagging='never'
|
| 124 |
+
)
|
| 125 |
+
|
| 126 |
+
if __name__ == "__main__":
|
| 127 |
+
iface.launch(debug=True, share=True)
|