Spaces:
Sleeping
Sleeping
Ajout de l'agent résumé YouTube
Browse files- Ajout des outils get_transcript et summarize
- Configuration de l’agent Hugging Face smolagents
- Utilise Qwen2.5-Coder-32B-Instruct pour générer le résumé
app.py
CHANGED
|
@@ -1,22 +1,49 @@
|
|
| 1 |
-
from smolagents import CodeAgent,DuckDuckGoSearchTool, HfApiModel,load_tool,
|
|
|
|
| 2 |
import datetime
|
| 3 |
import requests
|
| 4 |
import pytz
|
| 5 |
-
import yaml
|
| 6 |
from tools.final_answer import FinalAnswerTool
|
| 7 |
|
| 8 |
from Gradio_UI import GradioUI
|
| 9 |
|
| 10 |
-
#
|
| 11 |
@tool
|
| 12 |
-
def
|
| 13 |
-
|
| 14 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
Args:
|
| 16 |
-
|
| 17 |
-
|
| 18 |
"""
|
| 19 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 20 |
|
| 21 |
@tool
|
| 22 |
def get_current_time_in_timezone(timezone: str) -> str:
|
|
@@ -33,6 +60,7 @@ def get_current_time_in_timezone(timezone: str) -> str:
|
|
| 33 |
except Exception as e:
|
| 34 |
return f"Error fetching time for timezone '{timezone}': {str(e)}"
|
| 35 |
|
|
|
|
| 36 |
|
| 37 |
final_answer = FinalAnswerTool()
|
| 38 |
|
|
@@ -55,7 +83,7 @@ with open("prompts.yaml", 'r') as stream:
|
|
| 55 |
|
| 56 |
agent = CodeAgent(
|
| 57 |
model=model,
|
| 58 |
-
tools=[final_answer], ## add your tools here (don't remove final answer)
|
| 59 |
max_steps=6,
|
| 60 |
verbosity_level=1,
|
| 61 |
grammar=None,
|
|
|
|
| 1 |
+
from smolagents import CodeAgent,DuckDuckGoSearchTool, HfApiModel,load_tool, toolInferenceClientModel
|
| 2 |
+
from youtube_transcript_api import YouTubeTranscriptApi # pip install youtube-transcript-api
|
| 3 |
import datetime
|
| 4 |
import requests
|
| 5 |
import pytz
|
| 6 |
+
import yaml, textwrap, re
|
| 7 |
from tools.final_answer import FinalAnswerTool
|
| 8 |
|
| 9 |
from Gradio_UI import GradioUI
|
| 10 |
|
| 11 |
+
# ---------- TOOLS ----------
|
| 12 |
@tool
|
| 13 |
+
def get_transcript(url: str) -> str:
|
| 14 |
+
"""
|
| 15 |
+
Télécharge la transcription (FR ou EN) d’une vidéo YouTube.
|
| 16 |
+
Args:
|
| 17 |
+
url: lien complet YouTube
|
| 18 |
+
"""
|
| 19 |
+
# extraction robuste de l’ID
|
| 20 |
+
video_id = re.search(r"(?:v=|youtu\.be/)([^&\n?#]+)", url)
|
| 21 |
+
if not video_id:
|
| 22 |
+
return "Impossible de détecter l’ID vidéo."
|
| 23 |
+
video_id = video_id.group(1)
|
| 24 |
+
|
| 25 |
+
try:
|
| 26 |
+
transcript = YouTubeTranscriptApi.get_transcript(
|
| 27 |
+
video_id, languages=['fr', 'en'])
|
| 28 |
+
return " ".join(seg["text"] for seg in transcript)
|
| 29 |
+
except Exception as e:
|
| 30 |
+
return f"Erreur transcript : {e}"
|
| 31 |
+
|
| 32 |
+
@tool
|
| 33 |
+
def summarize(text: str, max_chars: int = 1500) -> str:
|
| 34 |
+
"""
|
| 35 |
+
Résume le texte en ~10 bullet points (Markdown).
|
| 36 |
Args:
|
| 37 |
+
text: transcription brute
|
| 38 |
+
max_chars: longueur max passée au LLM
|
| 39 |
"""
|
| 40 |
+
# coupe si trop long pour le contexte
|
| 41 |
+
snippet = text[:max_chars]
|
| 42 |
+
prompt = (
|
| 43 |
+
"Résume le texte suivant en 8‑10 bullet points clairs :\n\n"
|
| 44 |
+
f"{snippet}\n\n# Résumé :"
|
| 45 |
+
)
|
| 46 |
+
return prompt # on renvoie le prompt ; l’agent fera l’appel LLM
|
| 47 |
|
| 48 |
@tool
|
| 49 |
def get_current_time_in_timezone(timezone: str) -> str:
|
|
|
|
| 60 |
except Exception as e:
|
| 61 |
return f"Error fetching time for timezone '{timezone}': {str(e)}"
|
| 62 |
|
| 63 |
+
# ---------- AGENT ----------
|
| 64 |
|
| 65 |
final_answer = FinalAnswerTool()
|
| 66 |
|
|
|
|
| 83 |
|
| 84 |
agent = CodeAgent(
|
| 85 |
model=model,
|
| 86 |
+
tools=tools=[final_answer, get_transcript, summarize], ## add your tools here (don't remove final answer)
|
| 87 |
max_steps=6,
|
| 88 |
verbosity_level=1,
|
| 89 |
grammar=None,
|