Spaces:
Sleeping
Sleeping
Ilyas KHIAT
commited on
Commit
·
42117bb
1
Parent(s):
c089258
chatbot
Browse files- agents_page/recommended_agent.py +130 -18
- app.py +2 -2
- audit_page/audit.py +6 -1
- requirements.txt +1 -0
- utils/audit/rag.py +52 -0
agents_page/recommended_agent.py
CHANGED
|
@@ -1,6 +1,13 @@
|
|
| 1 |
import streamlit as st
|
| 2 |
from utils.audit.response_llm import generate_response_via_langchain
|
| 3 |
from textwrap import dedent
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 4 |
#st.set_page_config(page_title="Agents recommandés", page_icon="", layout="wide")
|
| 5 |
def remove_images_from_content(content):
|
| 6 |
filtered_content = {}
|
|
@@ -12,8 +19,78 @@ def remove_images_from_content(content):
|
|
| 12 |
|
| 13 |
return filtered_content
|
| 14 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
def recommended_agent_main():
|
| 16 |
st.title("Agents recommandés")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
|
| 18 |
if "audit" not in st.session_state or st.session_state.audit == {}:
|
| 19 |
st.error("Veuillez d'abord effectuer un audit pour obtenir des recommandations d'agents.")
|
|
@@ -35,26 +112,33 @@ def recommended_agent_main():
|
|
| 35 |
|
| 36 |
ressources = content
|
| 37 |
|
| 38 |
-
prompt = '''
|
| 39 |
-
|
|
|
|
|
|
|
| 40 |
|
| 41 |
-
|
| 42 |
|
| 43 |
-
|
| 44 |
-
* Nom
|
| 45 |
-
* Rôle
|
| 46 |
-
* Objectifs
|
| 47 |
-
* Outils utilisés par l'agent
|
| 48 |
-
* Tâches réalisées par l'agents
|
| 49 |
-
* Compétences de l'agent (backstory)
|
| 50 |
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 54 |
|
| 55 |
-
Une fois ce travail réalisé, tu proposes une série de 3 missions avec objectifs SMART pour chacun des agents Sol. A et Sol. B en présentation les résultats dans un tableau contenant :
|
| 56 |
-
Nom de l’agent
|
| 57 |
-
Objectifs à atteindre
|
| 58 |
'''
|
| 59 |
|
| 60 |
#display prompt and modify it
|
|
@@ -63,9 +147,37 @@ def recommended_agent_main():
|
|
| 63 |
if st.button("Générer les recommandations"):
|
| 64 |
resource_prompt = f'''Ressources fournies par l'utilisateur :{ressources}'''
|
| 65 |
prompt_modified = f"{prompt_modified}\n{resource_prompt}"
|
| 66 |
-
|
|
|
|
|
|
|
| 67 |
elif st.session_state.response_llm:
|
| 68 |
st.info("la dernière réponse générée est affichée ci-dessous")
|
| 69 |
-
st.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 70 |
|
| 71 |
recommended_agent_main()
|
|
|
|
| 1 |
import streamlit as st
|
| 2 |
from utils.audit.response_llm import generate_response_via_langchain
|
| 3 |
from textwrap import dedent
|
| 4 |
+
import streamlit as st
|
| 5 |
+
from langchain_openai import ChatOpenAI
|
| 6 |
+
from langchain_mistralai import ChatMistralAI
|
| 7 |
+
from langchain_core.prompts import ChatPromptTemplate
|
| 8 |
+
from langchain_core.output_parsers import StrOutputParser
|
| 9 |
+
from langchain_core.messages import AIMessage, HumanMessage
|
| 10 |
+
|
| 11 |
#st.set_page_config(page_title="Agents recommandés", page_icon="", layout="wide")
|
| 12 |
def remove_images_from_content(content):
|
| 13 |
filtered_content = {}
|
|
|
|
| 19 |
|
| 20 |
return filtered_content
|
| 21 |
|
| 22 |
+
def get_response(user_query, chat_history, db,llm=None,history_limit=5,stream=True):
|
| 23 |
+
retriever = db.as_retriever()
|
| 24 |
+
context = retriever.invoke(user_query)
|
| 25 |
+
template = """
|
| 26 |
+
Étant donné l'historique de la conversation : {chat_history}, le contexte qui est le document : {context}, et la question de l'utilisateur : {user_question}, repond comme un expert en agent IA.
|
| 27 |
+
Assurez-vous que la réponse soit adaptée au niveau d'expertise de l'utilisateur et aux spécificités du contexte fourni.
|
| 28 |
+
|
| 29 |
+
"""
|
| 30 |
+
|
| 31 |
+
prompt = ChatPromptTemplate.from_template(template)
|
| 32 |
+
|
| 33 |
+
#llm = ChatOpenAI(model="gpt-4o")
|
| 34 |
+
if not llm:
|
| 35 |
+
llm = ChatOpenAI(model="gpt-4o-mini")
|
| 36 |
+
elif llm == "GPT-4o":
|
| 37 |
+
llm = ChatOpenAI(model="gpt-4o")
|
| 38 |
+
elif llm == "Mistral Large 2 (FR)":
|
| 39 |
+
llm = ChatMistralAI(model_name="mistral-large-2407")
|
| 40 |
+
elif llm == "GPT-4o-mini":
|
| 41 |
+
llm = ChatOpenAI(model="gpt-4o-mini")
|
| 42 |
+
elif llm == "Mistral Nemo (FR)":
|
| 43 |
+
llm = ChatMistralAI(model_name="open-mistral-nemo-2407")
|
| 44 |
+
|
| 45 |
+
|
| 46 |
+
chain = prompt | llm
|
| 47 |
+
|
| 48 |
+
if not stream:
|
| 49 |
+
return chain.invoke({
|
| 50 |
+
"context": context,
|
| 51 |
+
"chat_history": chat_history[-history_limit:],
|
| 52 |
+
"user_question": user_query,
|
| 53 |
+
})
|
| 54 |
+
|
| 55 |
+
chain = chain | StrOutputParser()
|
| 56 |
+
|
| 57 |
+
if history_limit:
|
| 58 |
+
return chain.stream({
|
| 59 |
+
"context": context,
|
| 60 |
+
"chat_history": chat_history[-history_limit:],
|
| 61 |
+
"user_question": user_query,
|
| 62 |
+
})
|
| 63 |
+
|
| 64 |
+
return chain.stream({
|
| 65 |
+
"context": context,
|
| 66 |
+
"chat_history": chat_history,
|
| 67 |
+
"user_question": user_query,
|
| 68 |
+
})
|
| 69 |
+
|
| 70 |
+
def handle_display_models(index, models_names):
|
| 71 |
+
model = st.radio("Choisir un modèle",models_names, index=index)
|
| 72 |
+
return model
|
| 73 |
+
|
| 74 |
def recommended_agent_main():
|
| 75 |
st.title("Agents recommandés")
|
| 76 |
+
models_names = ["GPT-4o", "GPT-4o-mini"]
|
| 77 |
+
|
| 78 |
+
if "chat_history" not in st.session_state:
|
| 79 |
+
st.session_state.chat_history = [
|
| 80 |
+
]
|
| 81 |
+
|
| 82 |
+
if "model" not in st.session_state:
|
| 83 |
+
st.session_state.model = "GPT-4o-mini"
|
| 84 |
+
|
| 85 |
+
header = st.container()
|
| 86 |
+
col1, col2 = header.columns([1, 2])
|
| 87 |
+
|
| 88 |
+
with col1.popover("Modèles disponibles"):
|
| 89 |
+
new_model = handle_display_models(models_names.index(st.session_state.model), models_names)
|
| 90 |
+
|
| 91 |
+
st.session_state.model = new_model
|
| 92 |
+
|
| 93 |
+
st.markdown(f"- **{st.session_state.model}**")
|
| 94 |
|
| 95 |
if "audit" not in st.session_state or st.session_state.audit == {}:
|
| 96 |
st.error("Veuillez d'abord effectuer un audit pour obtenir des recommandations d'agents.")
|
|
|
|
| 112 |
|
| 113 |
ressources = content
|
| 114 |
|
| 115 |
+
prompt = '''
|
| 116 |
+
Tu es designer en intelligence artificielle (IA) spécialisé dans la création d'agents IA autonomes et performants.
|
| 117 |
+
|
| 118 |
+
A partir de ressources fournies par l'utilisateur (texte, documents, images, audio), tu es chargé de réaliser les tâches suivantes :
|
| 119 |
|
| 120 |
+
A/ Faire un résumé des ressources fournies en 500 caractères maximum
|
| 121 |
|
| 122 |
+
B/ Suggérer la création d'agents autonomes pour mettre en pratique les informations contenues dans les ressources fournies.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 123 |
|
| 124 |
+
Tu proposes deux solutions :
|
| 125 |
+
|
| 126 |
+
Sol. A : 1 seul agent IA dont tu suggéreras :
|
| 127 |
+
* Nom
|
| 128 |
+
* Rôle
|
| 129 |
+
* Objectifs
|
| 130 |
+
* Outils utilisés par l'agent
|
| 131 |
+
* Tâches réalisées par l'agents
|
| 132 |
+
* Compétences de l'agent (backstory)
|
| 133 |
+
|
| 134 |
+
Sol. B : 1 équipe d'agents tu suggéreras :
|
| 135 |
+
* Le nombre d'agents
|
| 136 |
+
* Pour chacune d'eux [Nom, Rôle, Objectifs, Outils utilisés par l'agent, Tâches réalisées par l'agents, Compétences de l'agent (backstory)]
|
| 137 |
+
|
| 138 |
+
Une fois ce travail réalisé, tu proposes une série de 3 missions avec objectifs SMART pour chacun des agents Sol. A et Sol. B en présentation les résultats dans un tableau contenant :
|
| 139 |
+
* Nom de l’agent
|
| 140 |
+
* Objectifs à atteindre
|
| 141 |
|
|
|
|
|
|
|
|
|
|
| 142 |
'''
|
| 143 |
|
| 144 |
#display prompt and modify it
|
|
|
|
| 147 |
if st.button("Générer les recommandations"):
|
| 148 |
resource_prompt = f'''Ressources fournies par l'utilisateur :{ressources}'''
|
| 149 |
prompt_modified = f"{prompt_modified}\n{resource_prompt}"
|
| 150 |
+
with st.chat_message("AI"):
|
| 151 |
+
st.session_state.response_llm = st.write_stream(generate_response_via_langchain(query=prompt_modified,stream=True))
|
| 152 |
+
st.session_state.chat_history.append(AIMessage(content=st.session_state.response_llm))
|
| 153 |
elif st.session_state.response_llm:
|
| 154 |
st.info("la dernière réponse générée est affichée ci-dessous")
|
| 155 |
+
with st.chat_message("AI"):
|
| 156 |
+
st.write(st.session_state.response_llm)
|
| 157 |
+
|
| 158 |
+
for message in st.session_state.chat_history[1:]:
|
| 159 |
+
if isinstance(message, AIMessage):
|
| 160 |
+
with st.chat_message("AI"):
|
| 161 |
+
st.markdown(message.content)
|
| 162 |
+
elif isinstance(message, HumanMessage):
|
| 163 |
+
with st.chat_message("Moi"):
|
| 164 |
+
st.write(message.content)
|
| 165 |
+
|
| 166 |
+
user_query = st.chat_input("Par ici ...")
|
| 167 |
+
if user_query is not None and user_query != "":
|
| 168 |
+
st.session_state.chat_history.append(HumanMessage(content=user_query))
|
| 169 |
+
|
| 170 |
+
with st.chat_message("Moi"):
|
| 171 |
+
st.markdown(user_query)
|
| 172 |
+
|
| 173 |
+
with st.chat_message("AI"):
|
| 174 |
+
st.markdown(f"**{st.session_state.model}**")
|
| 175 |
+
|
| 176 |
+
|
| 177 |
+
response = st.write_stream(get_response(user_query, st.session_state.chat_history,db=st.session_state.vectorstore, llm=st.session_state.model, stream=True))
|
| 178 |
+
st.session_state.chat_history.append(AIMessage(content=response))
|
| 179 |
+
|
| 180 |
+
|
| 181 |
+
|
| 182 |
|
| 183 |
recommended_agent_main()
|
app.py
CHANGED
|
@@ -17,8 +17,8 @@ def main():
|
|
| 17 |
|
| 18 |
pg = st.navigation(
|
| 19 |
{
|
| 20 |
-
"Audit de contenus": [audit_page
|
| 21 |
-
"Equipe d'agents IA": [recommended_agents
|
| 22 |
"Chatbot": [chatbot],
|
| 23 |
"Documentation": [documentation]
|
| 24 |
}
|
|
|
|
| 17 |
|
| 18 |
pg = st.navigation(
|
| 19 |
{
|
| 20 |
+
"Audit de contenus": [audit_page],
|
| 21 |
+
"Equipe d'agents IA": [recommended_agents],
|
| 22 |
"Chatbot": [chatbot],
|
| 23 |
"Documentation": [documentation]
|
| 24 |
}
|
audit_page/audit.py
CHANGED
|
@@ -2,6 +2,7 @@ import streamlit as st
|
|
| 2 |
import pymupdf as fitz
|
| 3 |
import pyperclip
|
| 4 |
from utils.audit.audit_doc import audit_descriptif_pdf,audit_text
|
|
|
|
| 5 |
import dotenv
|
| 6 |
from utils.audit.audit_audio import evaluate_audio_quality
|
| 7 |
from PIL import Image
|
|
@@ -56,7 +57,7 @@ def display_content_doc(content:dict,col:st):
|
|
| 56 |
else:
|
| 57 |
text = page["texte"]
|
| 58 |
|
| 59 |
-
col.
|
| 60 |
|
| 61 |
elif option == "liens":
|
| 62 |
if number == 0:
|
|
@@ -208,6 +209,8 @@ def audit_main():
|
|
| 208 |
st.session_state.name_file = ""
|
| 209 |
if "audit_simplified" not in st.session_state:
|
| 210 |
st.session_state.audit_simplified = {}
|
|
|
|
|
|
|
| 211 |
|
| 212 |
# File uploader
|
| 213 |
uploaded_file = col1.file_uploader("Télécharger un ou plusieurs documents")
|
|
@@ -223,5 +226,7 @@ def audit_main():
|
|
| 223 |
if "audit" in st.session_state and st.session_state.audit != {}:
|
| 224 |
display_audit(col1)
|
| 225 |
handle_display_content(col2)
|
|
|
|
|
|
|
| 226 |
|
| 227 |
audit_main()
|
|
|
|
| 2 |
import pymupdf as fitz
|
| 3 |
import pyperclip
|
| 4 |
from utils.audit.audit_doc import audit_descriptif_pdf,audit_text
|
| 5 |
+
from utils.audit.rag import setup_rag
|
| 6 |
import dotenv
|
| 7 |
from utils.audit.audit_audio import evaluate_audio_quality
|
| 8 |
from PIL import Image
|
|
|
|
| 57 |
else:
|
| 58 |
text = page["texte"]
|
| 59 |
|
| 60 |
+
col.code(text,language="text")
|
| 61 |
|
| 62 |
elif option == "liens":
|
| 63 |
if number == 0:
|
|
|
|
| 209 |
st.session_state.name_file = ""
|
| 210 |
if "audit_simplified" not in st.session_state:
|
| 211 |
st.session_state.audit_simplified = {}
|
| 212 |
+
if "vectorstore" not in st.session_state:
|
| 213 |
+
st.session_state.vectorstore = None
|
| 214 |
|
| 215 |
# File uploader
|
| 216 |
uploaded_file = col1.file_uploader("Télécharger un ou plusieurs documents")
|
|
|
|
| 226 |
if "audit" in st.session_state and st.session_state.audit != {}:
|
| 227 |
display_audit(col1)
|
| 228 |
handle_display_content(col2)
|
| 229 |
+
vectorstore = setup_rag(st.session_state.audit_simplified["type de fichier"],st.session_state.audit["content"])
|
| 230 |
+
st.session_state.vectorstore = vectorstore
|
| 231 |
|
| 232 |
audit_main()
|
requirements.txt
CHANGED
|
@@ -17,3 +17,4 @@ langchain-core
|
|
| 17 |
langchainhub
|
| 18 |
langchain-openai
|
| 19 |
langchain-mistralai
|
|
|
|
|
|
| 17 |
langchainhub
|
| 18 |
langchain-openai
|
| 19 |
langchain-mistralai
|
| 20 |
+
faiss-cpu
|
utils/audit/rag.py
ADDED
|
@@ -0,0 +1,52 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import streamlit as st
|
| 2 |
+
from dotenv import load_dotenv
|
| 3 |
+
from PyPDF2 import PdfReader
|
| 4 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
| 5 |
+
from langchain_community.embeddings import OpenAIEmbeddings
|
| 6 |
+
from langchain_community.vectorstores import FAISS
|
| 7 |
+
from langchain_community.chat_models import ChatOpenAI
|
| 8 |
+
from langchain.llms import HuggingFaceHub
|
| 9 |
+
from langchain import hub
|
| 10 |
+
from langchain_core.output_parsers import StrOutputParser
|
| 11 |
+
from langchain_core.runnables import RunnablePassthrough
|
| 12 |
+
from langchain_community.document_loaders import WebBaseLoader
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
def get_text_from_content_for_doc(content):
|
| 16 |
+
text = ""
|
| 17 |
+
for page in content:
|
| 18 |
+
text += content[page]["texte"]
|
| 19 |
+
return text
|
| 20 |
+
|
| 21 |
+
def get_text_from_content_for_audio(content):
|
| 22 |
+
return content["transcription"]
|
| 23 |
+
|
| 24 |
+
|
| 25 |
+
def get_text_chunks(text):
|
| 26 |
+
text_splitter = RecursiveCharacterTextSplitter(
|
| 27 |
+
chunk_size=500, # the character length of the chunck
|
| 28 |
+
chunk_overlap=100, # the character length of the overlap between chuncks
|
| 29 |
+
length_function=len # the length function - in this case, character length (aka the python len() fn.)
|
| 30 |
+
)
|
| 31 |
+
chunks = text_splitter.split_text(text)
|
| 32 |
+
return chunks
|
| 33 |
+
|
| 34 |
+
def get_vectorstore(text_chunks):
|
| 35 |
+
embedding = OpenAIEmbeddings(model="text-embedding-3-small")
|
| 36 |
+
vectorstore = FAISS.from_texts(texts=text_chunks, embedding=embedding)
|
| 37 |
+
return vectorstore
|
| 38 |
+
|
| 39 |
+
def setup_rag(file_type,content):
|
| 40 |
+
if file_type == "pdf":
|
| 41 |
+
text = get_text_from_content_for_doc(content)
|
| 42 |
+
elif file_type == "audio":
|
| 43 |
+
text = get_text_from_content_for_audio(content)
|
| 44 |
+
|
| 45 |
+
|
| 46 |
+
chunks = get_text_chunks(text)
|
| 47 |
+
|
| 48 |
+
vectorstore = get_vectorstore(chunks)
|
| 49 |
+
|
| 50 |
+
return vectorstore
|
| 51 |
+
|
| 52 |
+
|