import gradio as gr
from transformers import BlenderbotTokenizer, BlenderbotForConditionalGeneration
model_name = "facebook/blenderbot-400M-distill"
tokenizer = BlenderbotTokenizer.from_pretrained(model_name)
model = BlenderbotForConditionalGeneration.from_pretrained(model_name)
# declare translate model
def translate(text,mode):
if mode== "ztoe":
from transformers import AutoModelWithLMHead,AutoTokenizer,pipeline
mode_name = 'liam168/trans-opus-mt-zh-en'
model = AutoModelWithLMHead.from_pretrained(mode_name)
tokenizer = AutoTokenizer.from_pretrained(mode_name)
translation = pipeline("translation_zh_to_en", model=model, tokenizer=tokenizer)
translate_result = translation(text, max_length=400)
if mode == "etoz":
from transformers import AutoModelWithLMHead,AutoTokenizer,pipeline
mode_name = 'liam168/trans-opus-mt-en-zh'
model = AutoModelWithLMHead.from_pretrained(mode_name)
tokenizer = AutoTokenizer.from_pretrained(mode_name)
translation = pipeline("translation_en_to_zh", model=model, tokenizer=tokenizer)
translate_result = translation(text, max_length=400)
return translate_result
#initial list and variable
chat_history=[]
output=[]
bname=''
#declare emoji function
def add_emoji(response):
keyword_emoji_dict = {
"happy": "๐",
"sad": "๐ข",
"sorry":"๐",
"love": "โค๏ธ",
"like": "๐",
"dislike": "๐",
"Why": "๐ฅบ",
"cat":"๐ฑ",
"dog":"๐ถ",
"ๅจ" : "๐"
}
for keyword, emoji in keyword_emoji_dict.items():
response = response.replace(keyword, f"{keyword} {emoji}")
return response
# Define the keywords and their corresponding keywords
def tran_shortform(response):
keyword_shortform_dict = {
"yolo": "You only live once",
"lol":"funny" ,
"nvm": "nevermind",
"lmk": "tell me",
"btw": "by the way",
"dk":"don't know",
"idk":"I don't know" ,
"๐": "happy",
"๐ข": "sad",
"๐": "sorry",
"โค๏ธ": "love",
"๐": "like",
"๐": "dislike",
"๐ฅบ": "Why",
"๐ฑ": "cat",
"๐ถ": "dog",
}
response= response.lower()
for keyword, st in keyword_shortform_dict.items():
response = response.replace(keyword, f"{st}")
return response
# Define the keywords and their corresponding keywords
def add_shortform(response):
keyword_shortform_dict = {
"You only live once": "YOLO",
"funny": "LOL",
"laugh":"LOL",
"nevermind": "nvm",
"sorry": "sorryyyyy",
"tell me": "LMK",
"By the way": "BTW",
"don't know":"DK",
"i don't know":"IDK"
}
response= response.lower()
for keyword, st in keyword_shortform_dict.items():
response = response.replace(keyword, f"{st}")
return response
#------------------------------------------------------------------------------------------------------
#Chatbot start
def chatbot(text,name):
global chat_history #for chatbot to response
global outputsss # for user to read
global bname #the previous bot name
global raw # the text input by user. It will not be changed in the program
raw=text #distinguish text for the user to read it(raw) and text for the chatbot (text)
# if the user do not name the chatbot.
if name=='':
name="Mr. Chatbot"
#since the program will keep running in the website. When the name of chatbot change, the history and chatbot will be refreshed
if bname != name:
chat_history= []
outputsss=[]
bname=name # same the name
#change the shortform and emoji to english
text = tran_shortform(text)
# Try to detect the language of the input text
is_chinese = any(0x4e00 <= ord(char) <= 0x9fff for char in text.lower())
# If the input language is Chinese, convert the text to lowercase and check if it contains any Chinese characters
if is_chinese:
text = translate(text,"ztoe")
text=f"{text}"
text=text[23:(len(text)-3)]
# Look for keywords in the previous chat history
keyword_responses = {
"how are you": "I'm doing well๐, thank you for asking!",
"bye": "Goodbye!๐๐ป",
"bye.": "Goodbye!๐๐ป",
"thank you": "You're welcome!๐",
"thank you.": "You're welcome!๐",
"hello": f'I am {bname}. Nice to meet you!๐',
"hello.": f'I am {bname}. Nice to meet you!๐',
"hi": f'I am {bname}. Nice to meet you!๐',
}
# Generate a response based on the previous messages
if len(chat_history) > 0:
# Get the last message from the chat history
last_message = chat_history[-1][1]
# Generate a response based on the last message
encoded_input = tokenizer.encode(last_message + tokenizer.eos_token + text, return_tensors='pt')
generated = model.generate(encoded_input, max_length=1024, do_sample=True)
response = tokenizer.decode(generated[0], skip_special_tokens=True)
response=f"{response}"
else:
# If there is no previous message, generate a response using the default method
encoded_input = tokenizer(text, return_tensors='pt')
generated = model.generate(**encoded_input)
response = tokenizer.batch_decode(generated, skip_special_tokens=True)[0]
response=f"{response}"
# Add emojis to the response
response = add_emoji(response)
# Add short form to the response
response = add_shortform(response)
# output the standard response
if text.lower() in keyword_responses:
response = keyword_responses[text.lower()]
# update history
chat_history.append((text,response))
# If the input language was Chinese, translate the response back to Chinese
if is_chinese:
response = translate(response,"etoz")
# remove the header of the output
# translate simplified Chinese to traditional Chinese
from hanziconv import HanziConv
response = HanziConv.toTraditional(f"{response}")
response = f"{response} "
response=response[23:(len(response)-4)]
#if the response is amended. Show the amended text in blanket)
if raw.lower() == text.lower():
outputsss.append((raw,response))
else:
outputsss.append((f"{raw} ({text})", response))
return (outputsss)
#------------------------------------------------------------------------------------------------------
iface =gr.Interface(fn=chatbot,
inputs=[gr.inputs.Textbox(label="Chat๐", placeholder="Say somehting๐ฌ"),
gr.inputs.Textbox(label="Name the Bot๐บ", placeholder="give me a name๐")],
outputs=[gr.Chatbot(label="Chat Hereโบ๏ธ")],
title="๐ธEmphatic Chatbotโค๏ธ",
allow_flagging=False,
layout="vertical",
theme='gstaff/xkcd' , # credit: https://huggingface.co/spaces/gstaff/xkcd/tree/main
examples=[["ไฝ ๅฅฝ"], ["Hello"],["I am sad"],["ๆๅพๅทๅฟ"]],
article="Disclaimer
The chatbot does not have emotions or the ability to empathize in the same way that humans doThe chatbot can simulate empathy by recognizing and responding to certain emotional cues in language, but responses are generated by algorithms and not based on personal experiences or feelingsThe information and advice provided by the chatbot should not be used as a substitute for professional medical, psychological, or legal adviceUsers should always consult with qualified professionals in these fields for personalized recommendations that take into account their individual circumstancesThe chatbot is not infallible and may encounter technical difficulties or make errors that can impact its performanceThe information presented by the chatbot should always be cross-checked and authenticated before any action is taken based on itInteractions with the chatbot may be subject to recording or monitoring for the purposes of quality assurance and trainingAny information collected by the chatbot must be handled in compliance with relevant data protection laws and regulations.")
#.launch(share=True)
iface.launch()