Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -1,146 +1,321 @@
|
|
1 |
import os
|
2 |
-
from
|
3 |
import gradio as gr
|
|
|
|
|
|
|
|
|
|
|
4 |
import requests
|
5 |
-
|
6 |
-
from typing import List, Dict, Union, Optional
|
7 |
import pandas as pd
|
8 |
-
import
|
9 |
-
import
|
10 |
-
|
11 |
-
import
|
12 |
-
import
|
13 |
-
from
|
14 |
-
from
|
15 |
-
|
|
|
|
|
|
|
|
|
16 |
|
17 |
load_dotenv()
|
18 |
|
19 |
# (Keep Constants as is)
|
20 |
# --- Constants ---
|
21 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
22 |
-
|
23 |
|
24 |
# --- Basic Agent Definition ---
|
|
|
|
|
25 |
|
26 |
-
|
27 |
-
def __init__(self, model_name: str = "gemini-pro"):
|
28 |
-
"""
|
29 |
-
Multi-modal agent powered by Google Gemini with:
|
30 |
-
- Web search
|
31 |
-
- Wikipedia access
|
32 |
-
- Document processing
|
33 |
-
"""
|
34 |
-
self.model = genai.GenerativeModel(model_name)
|
35 |
-
self.wiki = wikipediaapi.Wikipedia('en')
|
36 |
-
self.searx_url = "https://searx.space/search" # Public Searx instance
|
37 |
-
|
38 |
-
print("BasicAgent initialized.")
|
39 |
-
|
40 |
-
def __call__(self, question: str) -> str:
|
41 |
-
print(f"Agent received question (first 50 chars): {question[:50]}...")
|
42 |
-
fixed_answer = self.process_request(question)
|
43 |
-
print(f"Agent returning answer: {fixed_answer}")
|
44 |
-
return fixed_answer
|
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 |
-
"""Process PDF using Gemini's vision capability"""
|
98 |
-
try:
|
99 |
-
# For Gemini 1.5 or later which supports file uploads
|
100 |
-
with open(file_path, "rb") as f:
|
101 |
-
file = genai.upload_file(f)
|
102 |
-
response = self.model.generate_content(
|
103 |
-
["Extract and summarize the key points from this document:", file]
|
104 |
-
)
|
105 |
-
return response.text
|
106 |
-
except:
|
107 |
-
# Fallback for older Gemini versions
|
108 |
-
try:
|
109 |
-
import PyPDF2
|
110 |
-
with open(file_path, 'rb') as f:
|
111 |
-
reader = PyPDF2.PdfReader(f)
|
112 |
-
return "\n".join([page.extract_text() for page in reader.pages])
|
113 |
-
except ImportError:
|
114 |
-
return "PDF processing requires PyPDF2 (pip install PyPDF2)"
|
115 |
-
|
116 |
-
def _process_word(self, file_path: str) -> str:
|
117 |
-
"""Process Word documents"""
|
118 |
-
try:
|
119 |
-
from docx import Document
|
120 |
-
doc = Document(file_path)
|
121 |
-
return "\n".join([para.text for para in doc.paragraphs])
|
122 |
-
except ImportError:
|
123 |
-
return "Word processing requires python-docx (pip install python-docx)"
|
124 |
|
125 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
126 |
"""
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
131 |
"""
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
|
140 |
-
|
141 |
-
|
142 |
-
|
143 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
144 |
|
145 |
|
146 |
|
|
|
1 |
import os
|
2 |
+
from typing import Annotated, Optional, TypedDict
|
3 |
import gradio as gr
|
4 |
+
from langchain_core.messages import AnyMessage, HumanMessage, SystemMessage
|
5 |
+
from langchain_openai import ChatOpenAI
|
6 |
+
from langgraph.graph.message import add_messages
|
7 |
+
from langgraph.graph import StateGraph, START
|
8 |
+
from langgraph.prebuilt import tools_condition, ToolNode
|
9 |
import requests
|
|
|
|
|
10 |
import pandas as pd
|
11 |
+
from langchain.tools import Tool
|
12 |
+
from dotenv import load_dotenv
|
13 |
+
|
14 |
+
from arxiv_searcher import ArxivSearcher
|
15 |
+
from chess_algebraic_notation_retriever import ChessAlgebraicNotationMoveRetriever
|
16 |
+
from excel_file_reader import ExcelFileReader
|
17 |
+
from image_question_answer_tool import ImageQuestionAnswerTool
|
18 |
+
from python_code_question_answer_tool import PythonCodeQuestionAnswerTool
|
19 |
+
from tavily_searcher import TavilySearcher
|
20 |
+
from transcriber import Transcriber
|
21 |
+
from wikipedia_searcher import WikipediaSearcher
|
22 |
+
from youtube_video_question_answer_tool import YoutubeVideoQuestionAnswerTool
|
23 |
|
24 |
load_dotenv()
|
25 |
|
26 |
# (Keep Constants as is)
|
27 |
# --- Constants ---
|
28 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
29 |
+
ASSOCIATED_FILE_ENDPOINT = f"{DEFAULT_API_URL}/files/"
|
30 |
|
31 |
# --- Basic Agent Definition ---
|
32 |
+
# ----- THIS IS WERE YOU CAN BUILD WHAT YOU WANT ------
|
33 |
+
#search_tool = DuckDuckGoSearchRun()
|
34 |
|
35 |
+
#search_tool = DuckDuckGoSearcherTool()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
36 |
|
37 |
+
def retrieve_task_file(task_id: str) -> Optional[bytes]:
|
38 |
+
"""
|
39 |
+
Retrieve the task file for a given task ID.
|
40 |
+
"""
|
41 |
+
try:
|
42 |
+
response = requests.get(ASSOCIATED_FILE_ENDPOINT + task_id, timeout=15)
|
43 |
+
response.raise_for_status()
|
44 |
+
if response.status_code != 200:
|
45 |
+
print(f"Error fetching file: {response.status_code}")
|
46 |
+
return None
|
47 |
+
#print(f"Fetched file: {response.content}")
|
48 |
+
return response.content
|
49 |
+
except requests.exceptions.RequestException as e:
|
50 |
+
print(f"Error fetching file: {e}")
|
51 |
+
return None
|
52 |
+
except Exception as e:
|
53 |
+
print(f"An unexpected error occurred fetching file: {e}")
|
54 |
+
return None
|
55 |
|
56 |
+
def retrieve_next_chess_move_in_algebraic_notation(task_file_path: str, is_black_turn: bool) -> str:
|
57 |
+
"""
|
58 |
+
Retrieve the next chess move in algebraic notation from an image path.
|
59 |
+
"""
|
60 |
+
if task_file_path is None:
|
61 |
+
return "Error: Task file not found."
|
62 |
+
# Retrieve the next chess move in algebraic notation
|
63 |
+
next_chess_move = ChessAlgebraicNotationMoveRetriever().retrieve(task_file_path, is_black_turn)
|
64 |
+
return next_chess_move
|
65 |
+
|
66 |
+
# Initialize the tool
|
67 |
+
retrieve_next_chess_move_in_algebraic_notation_tool = Tool(
|
68 |
+
name="retrieve_next_chess_move_in_algebraic_notation",
|
69 |
+
func=retrieve_next_chess_move_in_algebraic_notation,
|
70 |
+
description="Retrieve the next chess move in algebraic notation from an image path."
|
71 |
+
)
|
72 |
+
|
73 |
+
def transcribe_audio(file_path: str) -> str:
|
74 |
+
if file_path is None:
|
75 |
+
return "Error: Audio path not found."
|
76 |
+
# Transcribe the audio
|
77 |
+
return Transcriber().transcribe(file_path)
|
78 |
+
|
79 |
+
# Initialize the tool
|
80 |
+
transcribe_audio_tool = Tool(
|
81 |
+
name="transcribe_audio",
|
82 |
+
func=transcribe_audio,
|
83 |
+
description="Transcribe the audio from an audio path."
|
84 |
+
)
|
85 |
+
|
86 |
+
# Initialize the tool
|
87 |
+
answer_python_code_tool = PythonCodeQuestionAnswerTool()
|
88 |
+
|
89 |
+
# Initialize the tool
|
90 |
+
answer_image_question_tool = ImageQuestionAnswerTool()
|
91 |
+
|
92 |
+
# Initialize the tool
|
93 |
+
answer_youtube_video_question_tool = YoutubeVideoQuestionAnswerTool()
|
94 |
+
|
95 |
+
'''def answer_youtube_video_question(youtube_video_url: str, question: str) -> str:
|
96 |
+
"""
|
97 |
+
Answer the question based on the youtube video.
|
98 |
+
"""
|
99 |
+
if youtube_video_url is None:
|
100 |
+
return "Error: Video not found."
|
101 |
+
# Download the video
|
102 |
+
video_path = YoutubeVideoDownloader().download_video(youtube_video_url)
|
103 |
+
# Answer the question
|
104 |
+
return VideoQuestionAnswer().answer(video_path, question)
|
105 |
+
# Initialize the tool
|
106 |
+
answer_youtube_video_question_tool = Tool(
|
107 |
+
name="answer_youtube_video_question",
|
108 |
+
func=answer_youtube_video_question,
|
109 |
+
description="Answer the question based on the youtube video."
|
110 |
+
)'''
|
111 |
+
|
112 |
+
def read_excel_file(file_path: str) -> str:
|
113 |
+
if file_path is None:
|
114 |
+
return "Error: File not found."
|
115 |
+
return ExcelFileReader().read_file(file_path)
|
116 |
+
|
117 |
+
# Initialize the tool
|
118 |
+
read_excel_file_tool = Tool(
|
119 |
+
name="read_excel_file",
|
120 |
+
func=read_excel_file,
|
121 |
+
description="Read the excel file."
|
122 |
+
)
|
123 |
+
|
124 |
+
# Initialize the tool
|
125 |
+
wikipedia_search_tool = Tool(
|
126 |
+
name="wikipedia_search",
|
127 |
+
func=WikipediaSearcher().search,
|
128 |
+
description="Search Wikipedia for a given query."
|
129 |
+
)
|
130 |
+
|
131 |
+
# Initialize the tool
|
132 |
+
arxiv_search_tool = Tool(
|
133 |
+
name="arxiv_search",
|
134 |
+
func=ArxivSearcher().search,
|
135 |
+
description="Search Arxiv for a given query."
|
136 |
+
)
|
137 |
+
|
138 |
+
tavily_search_tool = Tool(
|
139 |
+
name="tavily_search",
|
140 |
+
func=TavilySearcher().search,
|
141 |
+
description="Search the web for a given query."
|
142 |
+
)
|
143 |
+
|
144 |
+
def format_gaia_answer(answer: str) -> str:
|
145 |
+
llm = ChatOpenAI(model="o3-mini", openai_api_key=os.getenv("OPENAI_API_KEY"))
|
146 |
+
prompt = f"""
|
147 |
+
You are formatting answers for the GAIA benchmark, which requires responses to be concise and unambiguous.
|
148 |
+
Given the answer: {answer}
|
149 |
+
Return the answer in the correct GAIA format:
|
150 |
+
- If the answer is a single word or number, return it without any additional text or formatting.
|
151 |
+
- If the answer is a list, return a comma-separated list without any additional text or formatting.
|
152 |
+
- If the answer is a string, return it without any additional text or formatting.
|
153 |
+
Do not include any prefixes, dots, enumerations, explanations, or quotation marks.
|
154 |
+
Do not include any additional text or formatting.
|
155 |
+
"""
|
156 |
+
response = llm.invoke(prompt)
|
157 |
+
# Delete double quotes
|
158 |
+
return response.content.strip().replace('"', '')
|
159 |
+
|
160 |
+
class AgentState(TypedDict):
|
161 |
+
# The document provided
|
162 |
+
messages: Annotated[list[AnyMessage], add_messages]
|
163 |
+
file_path: Optional[str]
|
164 |
+
|
165 |
+
class BasicAgent:
|
166 |
+
def __init__(self):
|
167 |
+
tools = [
|
168 |
+
tavily_search_tool,
|
169 |
+
arxiv_search_tool,
|
170 |
+
wikipedia_search_tool,
|
171 |
+
transcribe_audio_tool,
|
172 |
+
answer_python_code_tool,
|
173 |
+
answer_image_question_tool,
|
174 |
+
answer_youtube_video_question_tool,
|
175 |
+
read_excel_file_tool
|
176 |
+
]
|
177 |
+
'''llm = ChatGoogleGenerativeAI(
|
178 |
+
model="gemini-2.0-flash",
|
179 |
+
temperature=0.2,
|
180 |
+
api_key=os.getenv("GEMINI_API_KEY")
|
181 |
+
)'''
|
182 |
+
llm = ChatOpenAI(model="o3-mini", openai_api_key=os.getenv("OPENAI_API_KEY"))
|
183 |
+
self.llm_with_tools = llm.bind_tools(tools)
|
184 |
+
builder = StateGraph(AgentState)
|
185 |
+
|
186 |
+
# Define nodes: these do the work
|
187 |
+
builder.add_node("assistant", self.assistant)
|
188 |
+
builder.add_node("tools", ToolNode(tools))
|
189 |
+
|
190 |
+
# Define edges: these determine how the control flow moves
|
191 |
+
builder.add_edge(START, "assistant")
|
192 |
+
builder.add_conditional_edges(
|
193 |
+
"assistant",
|
194 |
+
# If the latest message requires a tool, route to tools
|
195 |
+
# Otherwise, provide a direct response
|
196 |
+
tools_condition,
|
197 |
+
)
|
198 |
+
builder.add_edge("tools", "assistant")
|
199 |
+
self.agent = builder.compile()
|
200 |
|
201 |
+
print("BasicAgent initialized.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
202 |
|
203 |
+
def assistant(self, state: AgentState):
|
204 |
+
# System message
|
205 |
+
textual_description_of_tools="""
|
206 |
+
tavily_search(query: str) -> str:
|
207 |
+
Search the web for a given query.
|
208 |
+
Args:
|
209 |
+
query: Query to search the web for (string).
|
210 |
+
Returns:
|
211 |
+
A single string containing the information found on the web.
|
212 |
+
arxiv_search(query: str) -> str:
|
213 |
+
Search Arxiv, that contains scientific papers, for a given query.
|
214 |
+
Args:
|
215 |
+
query: Query to search Arxiv for (string).
|
216 |
+
Returns:
|
217 |
+
A single string containing the answer to the question.
|
218 |
+
wikipedia_search(query: str) -> str:
|
219 |
+
Search Wikipedia for a given query.
|
220 |
+
Args:
|
221 |
+
query: Query to search Wikipedia for (string).
|
222 |
+
Returns:
|
223 |
+
A single string containing the answer to the question.
|
224 |
+
transcribe_audio(file_path: str) -> str:
|
225 |
+
Transcribe the audio from an audio path.
|
226 |
+
Args:
|
227 |
+
file_path: File path of the audio file (string).
|
228 |
+
Returns:
|
229 |
+
A single string containing the transcribed text from the audio.
|
230 |
+
|
231 |
+
answer_python_code(file_path: str, question: str) -> str:
|
232 |
+
Answer the question based on the python code.
|
233 |
+
Args:
|
234 |
+
file_path: File path of the python file (string).
|
235 |
+
question: Question to answer (string).
|
236 |
+
Returns:
|
237 |
+
A single string containing the answer to the question.
|
238 |
+
|
239 |
+
answer_image_question(file_path: str, question: str) -> str:
|
240 |
+
Answer the question based on the image.
|
241 |
+
Args:
|
242 |
+
file_path: File path of the image (string).
|
243 |
+
question: Question to answer (string).
|
244 |
+
Returns:
|
245 |
+
A single string containing the answer to the question.
|
246 |
+
|
247 |
+
download_youtube_video(youtube_video_url: str) -> str:
|
248 |
+
Download the Youtube video into a local file based on the URL
|
249 |
+
Args:
|
250 |
+
youtube_video_url: A youtube video url (string).
|
251 |
+
Returns:
|
252 |
+
A single string containing the file path of the downloaded youtube video.
|
253 |
+
answer_youtube_video_question(file_path: str, question: str) -> str:
|
254 |
+
Answer the question based on file path of the downloaded youtube video
|
255 |
+
Args:
|
256 |
+
file_path: File path of the downloaded youtube video (string).
|
257 |
+
question: Question to answer (string).
|
258 |
+
Returns:
|
259 |
+
A single string containing the answer to the question.
|
260 |
+
|
261 |
+
read_excel_file(file_path: str) -> str:
|
262 |
+
Read the excel file.
|
263 |
+
Args:
|
264 |
+
file_path: File path of the excel file (string).
|
265 |
+
Returns:
|
266 |
+
A markdown formatted string containing the contents of the excel file.
|
267 |
"""
|
268 |
+
file_path=state["file_path"]
|
269 |
+
prompt = f"""
|
270 |
+
You are a helpful assistant that can analyse images, videos, excel files and Python scripts and run computations with provided tools:
|
271 |
+
{textual_description_of_tools}
|
272 |
+
You have access to the file path of the attached file in case it's informed. Currently the file path is: {file_path}
|
273 |
+
Be direct and specific. GAIA benchmark requires exact matching answers.
|
274 |
+
For example, if asked "What is the capital of France?", respond simply with "Paris".
|
275 |
+
Do not include any prefixes, dots, enumerations, explanations, or quotation marks.
|
276 |
+
Do not include any additional text or formatting.
|
277 |
+
If you are required a number, return a number, not the items.
|
278 |
"""
|
279 |
+
sys_msg = SystemMessage(content=prompt)
|
280 |
+
|
281 |
+
return {
|
282 |
+
"messages": [self.llm_with_tools.invoke([sys_msg] + state["messages"], config={"configurable": {"file_path": state["file_path"]}})],
|
283 |
+
"file_path": state["file_path"]
|
284 |
+
}
|
285 |
+
'''return {
|
286 |
+
"messages": [self.llm_with_tools.invoke(
|
287 |
+
state["messages"],
|
288 |
+
config={"configurable": {"file_path": state["file_path"]}} # Aquí pasas el task_id
|
289 |
+
)],
|
290 |
+
"file_path": state["file_path"]
|
291 |
+
}'''
|
292 |
+
|
293 |
+
def __call__(self, question: str, task_id: str, file_name: str) -> str:
|
294 |
+
print(f"######################### Agent received question (first 50 chars): {question[:50]}... with file_name: {file_name}")
|
295 |
+
|
296 |
+
# Get the file path
|
297 |
+
tmp_file_path = None
|
298 |
+
if file_name is not None and file_name != "":
|
299 |
+
file_content = retrieve_task_file(task_id)
|
300 |
+
if file_content is not None:
|
301 |
+
print(f"Saving file {file_name} to tmp folder")
|
302 |
+
tmp_file_path = f"tmp/{file_name}"
|
303 |
+
with open(tmp_file_path, "wb") as f:
|
304 |
+
f.write(file_content)
|
305 |
+
# Show the file path
|
306 |
+
print(f"File path: {tmp_file_path}")
|
307 |
+
|
308 |
+
messages = self.agent.invoke({"messages": [HumanMessage(question)], "file_path": tmp_file_path})
|
309 |
+
# Show the messages
|
310 |
+
for m in messages['messages']:
|
311 |
+
m.pretty_print()
|
312 |
+
answer = messages["messages"][-1].content
|
313 |
+
answer = format_gaia_answer(answer)
|
314 |
+
print(f"######################### Agent returning answer: {answer}\n")
|
315 |
+
# Delete the file
|
316 |
+
if tmp_file_path is not None:
|
317 |
+
os.remove(tmp_file_path)
|
318 |
+
return answer
|
319 |
|
320 |
|
321 |
|