File size: 15,803 Bytes
702d7ab 3b3520e a8f4003 ca0effb a8f4003 be387b1 a8f4003 7d74414 c36a40b 1281b5f 702d7ab c36a40b c396b0b aee5f88 1281b5f 702d7ab a8f4003 1281b5f a8f4003 1281b5f 7e486cf 1281b5f a8f4003 e1abe10 3b3520e 1281b5f 3051da3 1281b5f 3051da3 1281b5f |
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 |
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig,HfArgumentParser,TrainingArguments,pipeline, logging
import os
from datasets import load_dataset
import pandas as pd
import pyarrow as pa
import pyarrow.dataset as ds
from datasets import Dataset
import re
from langchain_community.embeddings import SentenceTransformerEmbeddings
from langchain_community.vectorstores import FAISS
from sklearn.metrics.pairwise import cosine_similarity
import json
import pickle
import numpy as np
model_name = "sentence-transformers/all-MiniLM-L6-v2"
embedding_llm = SentenceTransformerEmbeddings(model_name=model_name)
loaded_vector_db = FAISS.load_local('index.faiss', embedding_llm, allow_dangerous_deserialization=True)
def load_data(text_filename='docs_text.json', embeddings_filename='docs_embeddings.json'):
with open(text_filename, 'r', encoding='utf-8') as f:
docs_text = json.load(f)
with open(embeddings_filename, 'r') as f:
docs_embeddings = json.load(f)
return docs_text, docs_embeddings
#docs_text, docs_embeddings = load_data()
def mot_cle(path):
with open(path, 'r') as fichier:
contenu = fichier.read()
# Séparer les mots en utilisant la virgule comme séparateur
mots = contenu.split(',')
# Afficher les mots pour vérifier
for mot in mots:
print(mot.strip())
# stocker les mots dans un tableau (une liste)
tableau_de_mots = [mot.strip() for mot in mots]
return tableau_de_mots
def pip(question,docs_text, docs_embeddings,mots_a_verifier):
query_text = question
query_embedding = embedding_llm.embed_query(query_text)
query_embedding_array = np.array(query_embedding)
docs_embeddings=np.array(docs_embeddings)
# Question à analyser
question = query_text
# Convertir la question en une liste de mots
mots_question = question.lower().split()
bi_grammes = [' '.join([mots_question[i], mots_question[i+1]]) for i in range(len(mots_question)-1)]
#mots_a_verifier_lower=[mot.lower() for mot in mots_a_verifier]
mots_a_verifier_lower = {mot.lower(): mot for mot in mots_a_verifier}
mots_question_lower=[mot.lower() for mot in mots_question]
bi_grammes_lower=[mot.lower() for mot in bi_grammes]
# Trouver les mots de la question qui sont dans le tableau
mots_trouves1 = [mots_a_verifier_lower[mot] for mot in mots_a_verifier_lower if mot in bi_grammes_lower]
if not mots_trouves1:
mots_trouves1 = [mots_a_verifier_lower[mot] for mot in mots_a_verifier_lower if mot in mots_question_lower ]
# Afficher les mots trouvés
mots_trouves=mots_trouves1
if not mots_trouves:
similarities = [cosine_similarity(doc.reshape(1,-1), query_embedding_array.reshape(1,-1)) for doc in docs_embeddings]
sorted_docs = sorted(zip(docs_text, docs_embeddings, similarities), key=lambda x: x[2], reverse=True)
similar_docs1 = [(doc,sim) for doc, _, sim in sorted_docs if sim > 0.72]
if not similar_docs1:
similar_docs2 = [(doc,sim) for doc, _, sim in sorted_docs if sim > 0.65]
if not similar_docs2:
similar_docs = [(doc,sim) for doc, _, sim in sorted_docs if sim > 0.4]
if not similar_docs:
similar_docsA = [(doc,sim) for doc, _, sim in sorted_docs if (sim >= 0.3 and sim<0.4)]
if not similar_docsA:
print("As a chatbot for Djezzy, I can provide information exclusively about our affiliated companies. Unfortunately, I'm unable to respond to inquiries outside of that scope.")
generate2="As a chatbot for Djezzy, I can provide information exclusively about our affiliated companies. Unfortunately, I'm unable to respond to inquiries outside of that scope."
generates.append(generate2)
else:
print("I apologize, I don't fully understand your question. You can contact our customer service for answers to your needs, or if you can provide more details, I would be happy to help.")
generate1="I apologize, I don't fully understand your question. You can contact our customer service for answers to your needs, or if you can provide more details, I would be happy to help."
generates.append(generate1)
else:
context="\n---------------------\n".join([doc for doc,_ in similar_docs[:4]]if len(similar_docs) >=3 else [doc for doc, _ in similar_docs[:1]])
system_message=" "
prompt = f"[INST] <<SYS>>\n As Djezzy's chatbot\nread each paraphrase in the context and Answer the question .\ndo not take into consideration the paragraphs which have no relation to the question\n if there is not a paragraph that is related to the question, respond that for this question it's best to reach out to our customer service team . They'll be able to assist you with your needs\n just give me the answer I don't want any other details \n ###context:{context}<</SYS>>\n\n ###question: {query_text} [/INST]"
#prompt = f" <bos><start_of_turn>user \n read each paraphrase in the context and Answer the question .\ndo not take into consideration the paragraphs which have no relation to the question\n if there is not a paragraph that is related to the question, respond that for this question it's best to reach out to our customer service team . They'll be able to assist you with your needs\n just give me the answer I don't want any other details \n###context:\n{contexts[0]}\n ###question:\nWhat are the benefits of opting for the Djezzy Legend 100 DA package? \n###answer:\n{reponses[0]}<eos>\nuser \n read each paraphrase in the context and Answer the question .\ndo not take into consideration the paragraphs which have no relation to the question\n if there is not a paragraph that is related to the question, respond that for this question it's best to reach out to our customer service team . They'll be able to assist you with your needs\n just give me the answer I don't want any other details \n###context:\n{contexts[1]}\n ###question:\nWhat are the benefits of opting for the Djezzy Legend 100 DA package? \n###answer:\n{reponses[1]}<eos>\nuser read each paraphrase in the context and Answer the question .\ndo not take into consideration the paragraphs which have no relation to the question\n if there is not a paragraph that is related to the question, respond that for this question it's best to reach out to our customer service team . They'll be able to assist you with your needs\n just give me the answer I don't want any other details \n###context:\n{context}\n###question:\n{query_text}\n###answer:\n<end_of_turn>\n <start_of_turn>model" # replace the command here with something relevant to your task
#pipe = pipeline(task="text-generation", model=model, tokenizer=tokenizer,temperature=0.1,top_p=0.9, max_length=4000)
#result = pipe(prompt)
#repons=result[0]['generated_text'].split('[/INST]')[1].strip()
#generate=repons.replace("<start_of_turn>model", "")
#generates.append(generate)
#print(generate)
#print(result[0]['generated_text'])
else:
context = "\n---------------------\n".join([doc for doc, _ in similar_docs2[:2]] if len(similar_docs2) >= 2 else [doc for doc, _ in similar_docs2[:1]])
system_message=" "
prompt = f"[INST] <<SYS>>\n As Djezzy's chatbot\nread each paraphrase in the context and Answer the question .\ndo not take into consideration the paragraphs which have no relation to the question\n if there is not a paragraph that is related to the question, respond that for this question it's best to reach out to our customer service team . They'll be able to assist you with your needs\n just give me the answer I don't want any other details \n ###context:{context}<</SYS>>\n\n ###question: {query_text} [/INST]"
#prompt = f" <bos><start_of_turn>user \n read each paraphrase in the context and Answer the question .\ndo not take into consideration the paragraphs which have no relation to the question\n if there is not a paragraph that is related to the question, respond that for this question it's best to reach out to our customer service team . They'll be able to assist you with your needs\n just give me the answer I don't want any other details \n###context:\n{contexts[0]}\n ###question:\nWhat are the benefits of opting for the Djezzy Legend 100 DA package? \n###answer:\n{reponses[0]}<eos>\nuser \n read each paraphrase in the context and Answer the question .\ndo not take into consideration the paragraphs which have no relation to the question\n if there is not a paragraph that is related to the question, respond that for this question it's best to reach out to our customer service team . They'll be able to assist you with your needs\n just give me the answer I don't want any other details \n###context:\n{contexts[1]}\n ###question:\nWhat are the benefits of opting for the Djezzy Legend 100 DA package? \n###answer:\n{reponses[1]}<eos>\nuser read each paraphrase in the context and Answer the question .\ndo not take into consideration the paragraphs which have no relation to the question\n if there is not a paragraph that is related to the question, respond that for this question it's best to reach out to our customer service team . They'll be able to assist you with your needs\n just give me the answer I don't want any other details \n###context:\n{context}\n###question:\n{query_text}\n###answer:\n<end_of_turn>\n <start_of_turn>model" # replace the command here with something relevant to your task
#pipe = pipeline(task="text-generation", model=model, tokenizer=tokenizer,temperature=0.1,top_p=0.9, max_length=4000)
#result = pipe(prompt)
#repons=result[0]['generated_text'].split('[/INST]')[1].strip()
#generate=repons.replace("<start_of_turn>model", "")
#generates.append(generate)
#print(generate)
#print(result[0]['generated_text'])
else:
context="\n---------------------\n".join([doc for doc,_ in similar_docs1[:1]])
system_message=" "
prompt = f"[INST] <<SYS>>\n As Djezzy's chatbot\nread 3 times each paraphrase in the context and Answer the question .\ndo not take into consideration the paragraphs which have no relation to the question\n if there is not a paragraph that is related to the question, respond that for this question it's best to reach out to our customer service team . They'll be able to assist you with your needs\n just give me the answer I don't want any other details \n differentiates between each price and gives the correct answer and does not distinguish between the offers of each price\n ###context:{context}<</SYS>>\n\n {query_text}[/INST]"
#prompt = f" <bos><start_of_turn>user \n read each paraphrase in the context and Answer the question .\ndo not take into consideration the paragraphs which have no relation to the question\n if there is not a paragraph that is related to the question, respond that for this question it's best to reach out to our customer service team . They'll be able to assist you with your needs\n just give me the answer I don't want any other details \n###context:\n{contexts[0]}\n ###question:\nWhat are the benefits of opting for the Djezzy Legend 100 DA package? \n###answer:\n{reponses[0]}<eos>\nuser \n read each paraphrase in the context and Answer the question .\ndo not take into consideration the paragraphs which have no relation to the question\n if there is not a paragraph that is related to the question, respond that for this question it's best to reach out to our customer service team . They'll be able to assist you with your needs\n just give me the answer I don't want any other details \n###context:\n{contexts[1]}\n ###question:\nWhat are the benefits of opting for the Djezzy Legend 100 DA package? \n###answer:\n{reponses[1]}<eos>\nuser read each paraphrase in the context and Answer the question .\ndo not take into consideration the paragraphs which have no relation to the question\n if there is not a paragraph that is related to the question, respond that for this question it's best to reach out to our customer service team . They'll be able to assist you with your needs\n just give me the answer I don't want any other details \n###context:\n{context}\n###question:\n{query_text}\n###answer:\n<end_of_turn>\n <start_of_turn>model" # replace the command here with something relevant to your task
#pipe = pipeline(task="text-generation", model=model, tokenizer=tokenizer,temperature=0.1,top_p=0.9, max_length=4000)
#result = pipe(prompt)
#repons=result[0]['generated_text'].split('[/INST]')[1].strip()
#generate=repons.replace("<start_of_turn>model", "")
#generates.append(generate)
#print(generate)
#print(result[0]['generated_text'])
else:
i=0
similar_docs=[]
for i in range(len(mots_trouves)):
k=mots_trouves[i]
result=vector_db.similarity_search(
query_text,
k=1,
filter={'document':mots_trouves[i] }
)
similar_docs.append(result[0])
context="\n---------------------\n".join([similar_docs[i].page_content for i in range(len(similar_docs))])
system_message=" "
prompt = f"[INST] <<SYS>>\n As Djezzy's chatbot\nread each paraphrase in the context and Answer the question .\ndo not take into consideration the paragraphs which have no relation to the question\n if there is not a paragraph that is related to the question, respond that for this question it's best to reach out to our customer service team . They'll be able to assist you with your needs\n just give me the answer I don't want any other details \n ###context:{context}<</SYS>>\n\n ###question: {query_text} [/INST]"
#prompt = f" <bos><start_of_turn>user \n read each paraphrase in the context and Answer the question .\ndo not take into consideration the paragraphs which have no relation to the question\n if there is not a paragraph that is related to the question, respond that for this question it's best to reach out to our customer service team . They'll be able to assist you with your needs\n just give me the answer I don't want any other details \n###context:\n{contexts[0]}\n ###question:\nWhat are the benefits of opting for the Djezzy Legend 100 DA package? \n###answer:\n{reponses[0]}<eos>\nuser \n read each paraphrase in the context and Answer the question .\ndo not take into consideration the paragraphs which have no relation to the question\n if there is not a paragraph that is related to the question, respond that for this question it's best to reach out to our customer service team . They'll be able to assist you with your needs\n just give me the answer I don't want any other details \n###context:\n{contexts[1]}\n ###question:\nWhat are the benefits of opting for the Djezzy Legend 100 DA package? \n###answer:\n{reponses[1]}<eos>\nuser read each paraphrase in the context and Answer the question .\ndo not take into consideration the paragraphs which have no relation to the question\n if there is not a paragraph that is related to the question, respond that for this question it's best to reach out to our customer service team . They'll be able to assist you with your needs\n just give me the answer I don't want any other details \n###context:\n{context}\n###question:\n{query_text}\n###answer:\n<end_of_turn>\n <start_of_turn>model" # replace the command here with something relevant to your task
#pipe = pipeline(task="text-generation", model=model, tokenizer=tokenizer,temperature=0.1,top_p=0.9, max_length=4000)
#result = pipe(prompt)
#repons=result[0]['generated_text'].split('[/INST]')[1].strip()
#generate=repons.replace("<start_of_turn>model", "")
#generates.append(generate)
#print(generate)
#print(result[0]['generated_text'])
return prompt
|