Spaces:
Sleeping
Sleeping
Update agent.py
Browse files
agent.py
CHANGED
|
@@ -0,0 +1,523 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import json
|
| 2 |
+
import time
|
| 3 |
+
from datetime import datetime
|
| 4 |
+
from langchain_groq import ChatGroq
|
| 5 |
+
from langchain.tools import tool
|
| 6 |
+
from langgraph.graph import StateGraph, END
|
| 7 |
+
from langchain_core.runnables import RunnableLambda
|
| 8 |
+
from typing import TypedDict
|
| 9 |
+
from arxiv import Search, Client, SortCriterion
|
| 10 |
+
from duckduckgo_search import DDGS
|
| 11 |
+
import requests
|
| 12 |
+
from bs4 import BeautifulSoup
|
| 13 |
+
import re
|
| 14 |
+
import os
|
| 15 |
+
from langchain.tools import WikipediaQueryRun
|
| 16 |
+
from langchain.utilities import WikipediaAPIWrapper
|
| 17 |
+
|
| 18 |
+
@tool
|
| 19 |
+
def wikipedia_search(query: str, language: str = "en") -> str:
|
| 20 |
+
"""
|
| 21 |
+
Recherche des informations sur Wikipedia.
|
| 22 |
+
Args:
|
| 23 |
+
query: La requête de recherche
|
| 24 |
+
language: Code de langue (par défaut 'en')
|
| 25 |
+
"""
|
| 26 |
+
try:
|
| 27 |
+
wikipedia = WikipediaAPIWrapper(language=language, top_k_results=3)
|
| 28 |
+
tool = WikipediaQueryRun(api_wrapper=wikipedia)
|
| 29 |
+
return tool.run(query)
|
| 30 |
+
except Exception as e:
|
| 31 |
+
return f"Erreur Wikipedia: {str(e)}"
|
| 32 |
+
import pandas as pd
|
| 33 |
+
from dotenv import load_dotenv
|
| 34 |
+
|
| 35 |
+
load_dotenv()
|
| 36 |
+
|
| 37 |
+
# === VOTRE CODE AGENT (copié tel quel) ===
|
| 38 |
+
@tool
|
| 39 |
+
def math_tool(expression: str) -> str:
|
| 40 |
+
"""
|
| 41 |
+
Évalue une expression mathématique simple donnée sous forme de chaîne de caractères.
|
| 42 |
+
Exemple : "2 + 5 * 3"
|
| 43 |
+
"""
|
| 44 |
+
try:
|
| 45 |
+
allowed_chars = set('0123456789+-*/.() ')
|
| 46 |
+
if not all(c in allowed_chars for c in expression):
|
| 47 |
+
return "Erreur : Expression contient des caractères non autorisés"
|
| 48 |
+
result = eval(expression)
|
| 49 |
+
return f"Résultat : {result}"
|
| 50 |
+
except Exception as e:
|
| 51 |
+
return f"Erreur : {str(e)}"
|
| 52 |
+
|
| 53 |
+
@tool
|
| 54 |
+
def search_arxiv(query: str, max_results: int = 3) -> str:
|
| 55 |
+
"""
|
| 56 |
+
Recherche d'articles scientifiques sur arXiv.
|
| 57 |
+
Retourne les titres, auteurs, résumés et liens PDF.
|
| 58 |
+
"""
|
| 59 |
+
try:
|
| 60 |
+
if not query.strip():
|
| 61 |
+
return "Erreur : la requête de recherche arXiv est vide."
|
| 62 |
+
|
| 63 |
+
client = Client(num_retries=3)
|
| 64 |
+
search = Search(
|
| 65 |
+
query=query,
|
| 66 |
+
max_results=max_results,
|
| 67 |
+
sort_by=SortCriterion.Relevance
|
| 68 |
+
)
|
| 69 |
+
|
| 70 |
+
response = ""
|
| 71 |
+
for result in client.results(search):
|
| 72 |
+
response += f" Titre : {result.title.strip()}\n"
|
| 73 |
+
response += f" Auteurs : {', '.join([a.name for a in result.authors])}\n"
|
| 74 |
+
response += f" Résumé : {result.summary.strip()[:300]}...\n"
|
| 75 |
+
response += f" Lien : {result.pdf_url}\n\n"
|
| 76 |
+
|
| 77 |
+
return response or "Aucun résultat trouvé sur arXiv."
|
| 78 |
+
|
| 79 |
+
except Exception as e:
|
| 80 |
+
return f"Erreur lors de la recherche sur arXiv : {str(e)}"
|
| 81 |
+
|
| 82 |
+
@tool
|
| 83 |
+
def search_web(query: str, max_results: int = 3) -> str:
|
| 84 |
+
""" Recherche sur le web en utilisant DuckDuckGo.
|
| 85 |
+
Retourne les titres et liens des résultats.
|
| 86 |
+
Si la requête est vide, retourne une erreur.
|
| 87 |
+
"""
|
| 88 |
+
if not query.strip():
|
| 89 |
+
return "Erreur : La requête de recherche est vide."
|
| 90 |
+
with DDGS() as ddgs:
|
| 91 |
+
results = ddgs.text(query, max_results=max_results)
|
| 92 |
+
response = ""
|
| 93 |
+
for res in results:
|
| 94 |
+
response += f"- {res['title']}: {res['href']}\n"
|
| 95 |
+
return response or "Aucun résultat trouvé."
|
| 96 |
+
|
| 97 |
+
@tool
|
| 98 |
+
def html_scraper_tool(prompt: str) -> str:
|
| 99 |
+
"""
|
| 100 |
+
Extrait une URL depuis un prompt texte et scrappe la page correspondante.
|
| 101 |
+
Ex : "Scrappe-moi le site : www.google.com"
|
| 102 |
+
"""
|
| 103 |
+
match = re.search(r'(https?://)?(www\.[^\s]+)', prompt)
|
| 104 |
+
if not match:
|
| 105 |
+
return " Aucune URL valide trouvée dans le prompt."
|
| 106 |
+
|
| 107 |
+
url = match.group(0)
|
| 108 |
+
if not url.startswith("http"):
|
| 109 |
+
url = "https://" + url
|
| 110 |
+
|
| 111 |
+
try:
|
| 112 |
+
response = requests.get(url, timeout=5)
|
| 113 |
+
response.raise_for_status()
|
| 114 |
+
soup = BeautifulSoup(response.text, 'html.parser')
|
| 115 |
+
title = soup.title.string.strip() if soup.title else "Aucun titre trouvé"
|
| 116 |
+
return f" Page title : {title}"
|
| 117 |
+
|
| 118 |
+
except Exception as e:
|
| 119 |
+
return f" Erreur lors du scraping de '{url}' : {str(e)}"
|
| 120 |
+
|
| 121 |
+
@tool
|
| 122 |
+
def wikipedia_search(query: str, language: str = "en") -> str:
|
| 123 |
+
"""
|
| 124 |
+
Recherche des informations sur Wikipedia.
|
| 125 |
+
Args:
|
| 126 |
+
query: La requête de recherche
|
| 127 |
+
language: Code de langue (par défaut 'en')
|
| 128 |
+
"""
|
| 129 |
+
try:
|
| 130 |
+
print(f"Langue utilisée : {language}")
|
| 131 |
+
print(f"Recherche Wikipedia pour : {query}")
|
| 132 |
+
|
| 133 |
+
wikipedia = WikipediaAPIWrapper(language=language, top_k_results=3)
|
| 134 |
+
print("WikipediaAPIWrapper instancié.")
|
| 135 |
+
|
| 136 |
+
tool = WikipediaQueryRun(api_wrapper=wikipedia)
|
| 137 |
+
print("WikipediaQueryRun instancié.")
|
| 138 |
+
|
| 139 |
+
result = tool.run(query)
|
| 140 |
+
print("Résultat obtenu.")
|
| 141 |
+
return result
|
| 142 |
+
except Exception as e:
|
| 143 |
+
print(f"Erreur Wikipedia : {str(e)}")
|
| 144 |
+
return f"Erreur Wikipedia: {str(e)}"
|
| 145 |
+
|
| 146 |
+
@tool
|
| 147 |
+
def reverse_text(text: str) -> str:
|
| 148 |
+
"""
|
| 149 |
+
Inverse une chaîne de caractères.
|
| 150 |
+
"""
|
| 151 |
+
return text[::-1]
|
| 152 |
+
|
| 153 |
+
@tool
|
| 154 |
+
def process_excel(file_path: str) -> str:
|
| 155 |
+
"""
|
| 156 |
+
Traite un fichier Excel et extrait des informations.
|
| 157 |
+
"""
|
| 158 |
+
try:
|
| 159 |
+
df = pd.read_excel(file_path)
|
| 160 |
+
return df.to_string()
|
| 161 |
+
except Exception as e:
|
| 162 |
+
return f"Erreur Excel: {str(e)}"
|
| 163 |
+
from youtube_transcript_api import YouTubeTranscriptApi
|
| 164 |
+
|
| 165 |
+
@tool
|
| 166 |
+
def get_youtube_transcript(video_url: str) -> str:
|
| 167 |
+
"""
|
| 168 |
+
Récupère la transcription d'une vidéo YouTube.
|
| 169 |
+
"""
|
| 170 |
+
try:
|
| 171 |
+
video_id = video_url.split("watch?v=")[1]
|
| 172 |
+
transcript = YouTubeTranscriptApi.get_transcript(video_id)
|
| 173 |
+
return " ".join([entry['text'] for entry in transcript])
|
| 174 |
+
except Exception as e:
|
| 175 |
+
return f"Erreur YouTube: {str(e)}"
|
| 176 |
+
|
| 177 |
+
|
| 178 |
+
|
| 179 |
+
tools = {
|
| 180 |
+
"math": math_tool,
|
| 181 |
+
"search": search_web,
|
| 182 |
+
"arxiv": search_arxiv,
|
| 183 |
+
"html_scraper": html_scraper_tool,
|
| 184 |
+
"wikipedia": wikipedia_search,
|
| 185 |
+
"reverse_text": reverse_text,
|
| 186 |
+
"process_excel": process_excel,
|
| 187 |
+
"get_youtube_transcript": get_youtube_transcript
|
| 188 |
+
}
|
| 189 |
+
|
| 190 |
+
api_key = os.getenv("API_KEY_GROQ")
|
| 191 |
+
if not api_key:
|
| 192 |
+
raise ValueError("La variable d'environnement 'API_KEY_GROQ' n'est pas définie.")
|
| 193 |
+
|
| 194 |
+
llm = ChatGroq(
|
| 195 |
+
model="llama-3.1-8b-instant",
|
| 196 |
+
temperature=0.7,
|
| 197 |
+
max_tokens=1024,
|
| 198 |
+
api_key=api_key
|
| 199 |
+
)
|
| 200 |
+
|
| 201 |
+
class AgentState(TypedDict):
|
| 202 |
+
input: str
|
| 203 |
+
tool: str
|
| 204 |
+
processed_input: str
|
| 205 |
+
tool_output: str
|
| 206 |
+
final_answer: str
|
| 207 |
+
|
| 208 |
+
def call_llm(state: AgentState) -> AgentState:
|
| 209 |
+
"""
|
| 210 |
+
Appelle le LLM pour déterminer l'outil approprié.
|
| 211 |
+
"""
|
| 212 |
+
system_prompt = f"""
|
| 213 |
+
Analyze this user request: "{state['input']}"
|
| 214 |
+
|
| 215 |
+
Available tools:
|
| 216 |
+
- 'math': for calculations and mathematical expressions
|
| 217 |
+
- 'search': to perform web searches
|
| 218 |
+
- 'arxiv': to search for scientific papers on arXiv
|
| 219 |
+
- 'html_scraper': to scrape HTML content
|
| 220 |
+
- 'wikipedia': to search for information on Wikipedia
|
| 221 |
+
- 'reverse_text': to reverse a given text
|
| 222 |
+
- 'process_excel': to process Excel files
|
| 223 |
+
- 'get_youtube_transcript': to retrieve YouTube video transcripts
|
| 224 |
+
|
| 225 |
+
Respond **only** with the appropriate tool name.
|
| 226 |
+
NO SENTENCES, just the tool name.
|
| 227 |
+
"""
|
| 228 |
+
|
| 229 |
+
try:
|
| 230 |
+
response = llm.invoke(system_prompt)
|
| 231 |
+
tool_name = response.content.strip().lower()
|
| 232 |
+
tool_name = tool_name.strip("'\"")
|
| 233 |
+
|
| 234 |
+
if tool_name in ['math', 'search', 'arxiv', 'html_scraper', 'wikipedia', 'reverse_text', 'process_excel', 'get_youtube_transcript']:
|
| 235 |
+
state['tool'] = tool_name
|
| 236 |
+
return state
|
| 237 |
+
except Exception as e:
|
| 238 |
+
print(f"Erreur lors de l'appel LLM : {e}")
|
| 239 |
+
return state
|
| 240 |
+
|
| 241 |
+
def extract_math_expression(state: AgentState) -> AgentState:
|
| 242 |
+
"""
|
| 243 |
+
Extrait l'expression mathématique de l'input pour les demandes de type math.
|
| 244 |
+
"""
|
| 245 |
+
if state['tool'] == 'math':
|
| 246 |
+
system_prompt = f"""
|
| 247 |
+
Extrayez uniquement l'expression mathématique de cette question : "{state['input']}"
|
| 248 |
+
|
| 249 |
+
Exemples :
|
| 250 |
+
- "Quel est le résultat de 100 / 4 ?" → "100 / 4"
|
| 251 |
+
- "Calcule 15 + 27 * 3" → "15 + 27 * 3"
|
| 252 |
+
- "Quelle est la racine carrée de 144?" → "144 ** 0.5"
|
| 253 |
+
- "2 plus 3 fois 5" → "2 + 3 * 5"
|
| 254 |
+
- "Combien font 12 * 8 ?" → "12 * 8"
|
| 255 |
+
- "Quel est le résultat de 23 * (5 + 7) ?" → "23 * (5 + 7)"
|
| 256 |
+
|
| 257 |
+
Répondez uniquement par l'expression mathématique, sans explication.
|
| 258 |
+
"""
|
| 259 |
+
try:
|
| 260 |
+
response = llm.invoke(system_prompt)
|
| 261 |
+
math_expression = response.content.strip()
|
| 262 |
+
state['processed_input'] = math_expression
|
| 263 |
+
except Exception as e:
|
| 264 |
+
print(f"Erreur extraction math : {e}")
|
| 265 |
+
else:
|
| 266 |
+
state['processed_input'] = state['input']
|
| 267 |
+
return state
|
| 268 |
+
|
| 269 |
+
def generate_response(state: AgentState) -> AgentState:
|
| 270 |
+
"""
|
| 271 |
+
Génère la réponse finale pour l'utilisateur.
|
| 272 |
+
"""
|
| 273 |
+
system_prompt = f"""
|
| 274 |
+
The tool '{state['tool']}' returned: "{state['tool_output']}"
|
| 275 |
+
|
| 276 |
+
Formulate a clear and natural response for the user .
|
| 277 |
+
Integrate the result smoothly into your reply.
|
| 278 |
+
"""
|
| 279 |
+
try:
|
| 280 |
+
response = llm.invoke(system_prompt)
|
| 281 |
+
state['final_answer'] = response.content
|
| 282 |
+
return state
|
| 283 |
+
except Exception as e:
|
| 284 |
+
state['final_answer'] = f"Réponse générée avec succès : {state['tool_output']}"
|
| 285 |
+
return state
|
| 286 |
+
|
| 287 |
+
def create_agent_graph():
|
| 288 |
+
workflow = StateGraph(AgentState)
|
| 289 |
+
|
| 290 |
+
workflow.add_node("llm_decision", call_llm)
|
| 291 |
+
workflow.add_node("process", extract_math_expression)
|
| 292 |
+
workflow.add_node("response_generation", generate_response)
|
| 293 |
+
|
| 294 |
+
workflow.add_node("math_tool", RunnableLambda(lambda state: {
|
| 295 |
+
**state,
|
| 296 |
+
"tool_output": tools["math"].invoke(state["processed_input"])
|
| 297 |
+
}))
|
| 298 |
+
|
| 299 |
+
workflow.add_node("search_tool", RunnableLambda(lambda state: {
|
| 300 |
+
**state,
|
| 301 |
+
"tool_output": tools["search"](state["processed_input"])
|
| 302 |
+
}))
|
| 303 |
+
|
| 304 |
+
workflow.add_node("arxiv_tool", RunnableLambda(lambda state: {
|
| 305 |
+
**state,
|
| 306 |
+
"tool_output": tools["arxiv"](state["processed_input"])
|
| 307 |
+
}))
|
| 308 |
+
|
| 309 |
+
workflow.add_node("html_scraper_tool", RunnableLambda(lambda state: {
|
| 310 |
+
**state,
|
| 311 |
+
"tool_output": tools["html_scraper"](state["processed_input"])
|
| 312 |
+
}))
|
| 313 |
+
workflow.add_node("wikipedia_search", RunnableLambda(lambda state: {
|
| 314 |
+
**state,
|
| 315 |
+
"tool_output": tools["wikipedia"](state["processed_input"])
|
| 316 |
+
}))
|
| 317 |
+
workflow.add_node("reverse_text", RunnableLambda(lambda state: {
|
| 318 |
+
**state,
|
| 319 |
+
"tool_output": tools["reverse_text"](state["processed_input"])
|
| 320 |
+
}))
|
| 321 |
+
|
| 322 |
+
workflow.add_node("process_excel", RunnableLambda(lambda state: {
|
| 323 |
+
**state,
|
| 324 |
+
"tool_output": tools["process_excel"](state["processed_input"])
|
| 325 |
+
}))
|
| 326 |
+
|
| 327 |
+
workflow.add_node("get_youtube_transcript", RunnableLambda(lambda state: {
|
| 328 |
+
**state,
|
| 329 |
+
"tool_output": tools["get_youtube_transcript"](state["processed_input"])
|
| 330 |
+
}))
|
| 331 |
+
|
| 332 |
+
|
| 333 |
+
|
| 334 |
+
workflow.set_entry_point("llm_decision")
|
| 335 |
+
workflow.add_edge("llm_decision", "process")
|
| 336 |
+
|
| 337 |
+
def router(state: AgentState) -> str:
|
| 338 |
+
if state["tool"] == "math":
|
| 339 |
+
return "math_tool"
|
| 340 |
+
elif state["tool"] == "search":
|
| 341 |
+
return "search_tool"
|
| 342 |
+
elif state["tool"] == "arxiv":
|
| 343 |
+
return "arxiv_tool"
|
| 344 |
+
elif state["tool"] == "html_scraper":
|
| 345 |
+
return "html_scraper_tool"
|
| 346 |
+
elif state["tool"] == "wikipedia":
|
| 347 |
+
return "wikipedia_search"
|
| 348 |
+
elif state["tool"] == "reverse_text":
|
| 349 |
+
return "reverse_text"
|
| 350 |
+
elif state["tool"] == "process_excel":
|
| 351 |
+
return "process_excel"
|
| 352 |
+
elif state["tool"] == "get_youtube_transcript":
|
| 353 |
+
return "get_youtube_transcript"
|
| 354 |
+
|
| 355 |
+
workflow.add_conditional_edges("process", router, {
|
| 356 |
+
"math_tool": "math_tool",
|
| 357 |
+
"search_tool": "search_tool",
|
| 358 |
+
"arxiv_tool": "arxiv_tool",
|
| 359 |
+
"html_scraper_tool": "html_scraper_tool",
|
| 360 |
+
"wikipedia_search": "wikipedia_search",
|
| 361 |
+
"reverse_text": "reverse_text",
|
| 362 |
+
"process_excel": "process_excel",
|
| 363 |
+
"get_youtube_transcript": "get_youtube_transcript"
|
| 364 |
+
|
| 365 |
+
})
|
| 366 |
+
|
| 367 |
+
workflow.add_edge("math_tool", "response_generation")
|
| 368 |
+
workflow.add_edge("search_tool", "response_generation")
|
| 369 |
+
workflow.add_edge("arxiv_tool", "response_generation")
|
| 370 |
+
workflow.add_edge("html_scraper_tool", "response_generation")
|
| 371 |
+
workflow.add_edge("wikipedia_search", "response_generation")
|
| 372 |
+
workflow.add_edge("reverse_text", "response_generation")
|
| 373 |
+
workflow.add_edge("process_excel", "response_generation")
|
| 374 |
+
workflow.add_edge("get_youtube_transcript", "response_generation")
|
| 375 |
+
workflow.add_edge("response_generation", END)
|
| 376 |
+
|
| 377 |
+
return workflow.compile()
|
| 378 |
+
|
| 379 |
+
def run_agent(user_input: str) -> str:
|
| 380 |
+
"""
|
| 381 |
+
Exécute l'agent avec une entrée utilisateur.
|
| 382 |
+
"""
|
| 383 |
+
agent = create_agent_graph()
|
| 384 |
+
initial_state = AgentState(
|
| 385 |
+
input=user_input,
|
| 386 |
+
tool="",
|
| 387 |
+
processed_input="",
|
| 388 |
+
tool_output="",
|
| 389 |
+
final_answer=""
|
| 390 |
+
)
|
| 391 |
+
try:
|
| 392 |
+
result = agent.invoke(initial_state)
|
| 393 |
+
return result['final_answer']
|
| 394 |
+
except Exception as e:
|
| 395 |
+
return f"Erreur lors de l'exécution : {str(e)}"
|
| 396 |
+
|
| 397 |
+
# === SCRIPT D'ÉVALUATION ===
|
| 398 |
+
def evaluate_agent_on_dataset(input_file_path, output_file_path):
|
| 399 |
+
"""
|
| 400 |
+
Évalue l'agent sur un dataset de questions et sauvegarde les réponses.
|
| 401 |
+
|
| 402 |
+
Args:
|
| 403 |
+
input_file_path (str): Chemin vers le fichier JSON contenant les questions
|
| 404 |
+
output_file_path (str): Chemin vers le fichier de sortie pour les réponses
|
| 405 |
+
"""
|
| 406 |
+
|
| 407 |
+
# Charger les questions depuis le fichier JSON
|
| 408 |
+
try:
|
| 409 |
+
with open(input_file_path, 'r', encoding='utf-8') as f:
|
| 410 |
+
questions_data = json.load(f)
|
| 411 |
+
# récupèrer les 5 premières questions
|
| 412 |
+
questions_data = questions_data[:-1] # Limiter à 5 questions pour l'évaluation
|
| 413 |
+
print(f" Fichier chargé avec succès: {len(questions_data)} questions trouvées")
|
| 414 |
+
except FileNotFoundError:
|
| 415 |
+
print(f" Erreur: Le fichier {input_file_path} n'a pas été trouvé")
|
| 416 |
+
return
|
| 417 |
+
except json.JSONDecodeError as e:
|
| 418 |
+
print(f" Erreur lors du parsing JSON: {e}")
|
| 419 |
+
return
|
| 420 |
+
|
| 421 |
+
# Préparer la structure des résultats
|
| 422 |
+
results = []
|
| 423 |
+
start_time = datetime.now()
|
| 424 |
+
|
| 425 |
+
print(f"\n Début de l'évaluation - {start_time.strftime('%Y-%m-%d %H:%M:%S')}")
|
| 426 |
+
print("=" * 60)
|
| 427 |
+
|
| 428 |
+
# Traiter chaque question
|
| 429 |
+
for i, item in enumerate(questions_data, 1):
|
| 430 |
+
task_id = item.get('task_id', 'unknown')
|
| 431 |
+
question = item.get('question', '')
|
| 432 |
+
level = item.get('Level', 'unknown')
|
| 433 |
+
file_name = item.get('file_name', '')
|
| 434 |
+
|
| 435 |
+
print(f"\n Question {i}/{len(questions_data)}")
|
| 436 |
+
print(f"Task ID: {task_id}")
|
| 437 |
+
print(f"Level: {level}")
|
| 438 |
+
print(f"Question: {question[:100]}{'...' if len(question) > 100 else ''}")
|
| 439 |
+
|
| 440 |
+
if file_name:
|
| 441 |
+
print(f" Note: Cette question fait référence au fichier: {file_name}")
|
| 442 |
+
print(" L'agent ne peut pas traiter les fichiers joints actuellement.")
|
| 443 |
+
|
| 444 |
+
# Exécuter l'agent
|
| 445 |
+
try:
|
| 446 |
+
print("🤖 Traitement en cours...")
|
| 447 |
+
answer = run_agent(question)
|
| 448 |
+
error_message = None
|
| 449 |
+
|
| 450 |
+
except Exception as e:
|
| 451 |
+
print(f" Erreur lors du traitement: {str(e)}")
|
| 452 |
+
answer = f"Erreur: {str(e)}"
|
| 453 |
+
error_message = str(e)
|
| 454 |
+
|
| 455 |
+
# Sauvegarder le résultat
|
| 456 |
+
result = {
|
| 457 |
+
"username": "Bachir00",
|
| 458 |
+
"code_agent": "https://huggingface.co/spaces/Bachir00/Final_Assignment_Template/tree/main",
|
| 459 |
+
"task_id": task_id,
|
| 460 |
+
"submitted_answer": answer,
|
| 461 |
+
}
|
| 462 |
+
|
| 463 |
+
if error_message:
|
| 464 |
+
result["error_message"] = error_message
|
| 465 |
+
|
| 466 |
+
results.append(result)
|
| 467 |
+
|
| 468 |
+
print(f"Réponse: {answer[:150]}{'...' if len(answer) > 150 else ''}")
|
| 469 |
+
|
| 470 |
+
# Pause entre les requêtes pour éviter la surcharge
|
| 471 |
+
time.sleep(1)
|
| 472 |
+
|
| 473 |
+
# Sauvegarder tous les résultats
|
| 474 |
+
try:
|
| 475 |
+
with open(output_file_path, 'w', encoding='utf-8') as f:
|
| 476 |
+
json.dump(results, f, ensure_ascii=False, indent=2)
|
| 477 |
+
|
| 478 |
+
end_time = datetime.now()
|
| 479 |
+
duration = end_time - start_time
|
| 480 |
+
|
| 481 |
+
print("\n" + "=" * 60)
|
| 482 |
+
print("RÉSUMÉ DE L'ÉVALUATION")
|
| 483 |
+
print("=" * 60)
|
| 484 |
+
print(f" Total des questions traitées: {len(results)}")
|
| 485 |
+
print(f"⏱ Temps total: {duration}")
|
| 486 |
+
print(f" Résultats sauvegardés dans: {output_file_path}")
|
| 487 |
+
|
| 488 |
+
# Statistiques par status
|
| 489 |
+
success_count = sum(1 for r in results if r['status'] == 'success')
|
| 490 |
+
error_count = len(results) - success_count
|
| 491 |
+
|
| 492 |
+
print(f" Succès: {success_count}")
|
| 493 |
+
print(f" Erreurs: {error_count}")
|
| 494 |
+
|
| 495 |
+
if error_count > 0:
|
| 496 |
+
print(f" Taux de réussite: {success_count/len(results)*100:.1f}%")
|
| 497 |
+
|
| 498 |
+
# Statistiques par niveau
|
| 499 |
+
levels = {}
|
| 500 |
+
for result in results:
|
| 501 |
+
level = result['level']
|
| 502 |
+
if level not in levels:
|
| 503 |
+
levels[level] = 0
|
| 504 |
+
levels[level] += 1
|
| 505 |
+
|
| 506 |
+
print(f"\ Répartition par niveau:")
|
| 507 |
+
for level, count in sorted(levels.items()):
|
| 508 |
+
print(f" Niveau {level}: {count} questions")
|
| 509 |
+
|
| 510 |
+
print("\n Évaluation terminée avec succès!")
|
| 511 |
+
|
| 512 |
+
except Exception as e:
|
| 513 |
+
print(f" Erreur lors de la sauvegarde: {str(e)}")
|
| 514 |
+
|
| 515 |
+
# === UTILISATION ===
|
| 516 |
+
if __name__ == "__main__":
|
| 517 |
+
# Chemins des fichiers
|
| 518 |
+
input_file = "response_1748862846167.json" # Remplacez par le chemin de votre fichier JSON
|
| 519 |
+
output_file = f"agent_results_{datetime.now().strftime('%Y%m%d_%H%M%S')}.json"
|
| 520 |
+
|
| 521 |
+
# Lancer l'évaluation
|
| 522 |
+
evaluate_agent_on_dataset(input_file, output_file)
|
| 523 |
+
|