Spaces:
Build error
Build error
Update app.py
Browse files
app.py
CHANGED
|
@@ -6,39 +6,19 @@ import urllib.parse
|
|
| 6 |
import json
|
| 7 |
from bs4 import BeautifulSoup
|
| 8 |
import pandas as pd
|
| 9 |
-
from smolagents import CodeAgent, HfApiModel
|
| 10 |
-
from smolagents.tools import PythonInterpreterTool, WikipediaSearchTool
|
| 11 |
|
| 12 |
# --- Constants ---
|
| 13 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 14 |
GEMINI_API_KEY = "AIzaSyBO46AIuY3Lmq3-k2bZkABgc0gL6A1RV20"
|
| 15 |
GEMINI_API_URL = "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash:generateContent"
|
| 16 |
-
E2B_API_KEY = "e2b_456a0bfcd096383566e37dd7417993934cd4c490" # Thay bằng API key E2B của bạn
|
| 17 |
|
| 18 |
# --- Basic Agent Definition ---
|
| 19 |
class BasicAgent:
|
| 20 |
def __init__(self):
|
| 21 |
-
# Khởi tạo Gemini API
|
| 22 |
self.headers = {
|
| 23 |
"Content-Type": "application/json"
|
| 24 |
}
|
| 25 |
-
|
| 26 |
-
# Khởi tạo CodeAgent với PythonInterpreterTool và WikipediaSearchTool
|
| 27 |
-
model = HfApiModel(model_id="mistralai/Mixtral-8x7B-Instruct-v0.1")
|
| 28 |
-
python_tool = PythonInterpreterTool(
|
| 29 |
-
sandbox="e2b",
|
| 30 |
-
e2b_api_key=E2B_API_KEY,
|
| 31 |
-
timeout=30,
|
| 32 |
-
restricted_permissions=True
|
| 33 |
-
)
|
| 34 |
-
self.code_agent = CodeAgent(
|
| 35 |
-
tools=[python_tool, WikipediaSearchTool()],
|
| 36 |
-
model=model,
|
| 37 |
-
max_steps=5,
|
| 38 |
-
name="SecureCodeAgent",
|
| 39 |
-
description="A CodeAgent with secure code execution using E2B sandbox and Wikipedia search"
|
| 40 |
-
)
|
| 41 |
-
print("BasicAgent initialized with Gemini API, CodeAgent, and tools.")
|
| 42 |
|
| 43 |
def query_gemini(self, prompt: str) -> str:
|
| 44 |
"""Gọi API Gemini để trả lời câu hỏi."""
|
|
@@ -83,6 +63,21 @@ class BasicAgent:
|
|
| 83 |
print(f"Wikipedia search error: {e}")
|
| 84 |
return ""
|
| 85 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 86 |
def get_file(self, task_id: str) -> str:
|
| 87 |
"""Tải tệp đính kèm từ API."""
|
| 88 |
try:
|
|
@@ -97,12 +92,27 @@ class BasicAgent:
|
|
| 97 |
def clean_answer(self, answer: str) -> str:
|
| 98 |
"""Chuẩn hóa câu trả lời: loại bỏ khoảng trắng thừa, chuẩn hóa định dạng."""
|
| 99 |
if answer.startswith("Error"):
|
| 100 |
-
return
|
| 101 |
if "," in answer:
|
| 102 |
items = [item.strip() for item in answer.split(",")]
|
| 103 |
return ",".join(items)
|
|
|
|
|
|
|
| 104 |
return answer.strip()
|
| 105 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 106 |
def __call__(self, task_id: str, question: str) -> str:
|
| 107 |
print(f"Processing question (task {task_id}): {question[:50]}...")
|
| 108 |
file_content = self.get_file(task_id)
|
|
@@ -115,8 +125,10 @@ class BasicAgent:
|
|
| 115 |
if "how many" in question_lower or "number of" in question_lower:
|
| 116 |
if "mercedes sosa" in question_lower and "2000 and 2009" in question_lower:
|
| 117 |
search_result = self.search_wikipedia("Mercedes Sosa discography")
|
| 118 |
-
|
| 119 |
-
|
|
|
|
|
|
|
| 120 |
|
| 121 |
if "bird species" in question_lower and "youtube.com" in question_lower:
|
| 122 |
prompt += "\nThe video content is unavailable, but estimate the highest number of bird species that might appear simultaneously in a typical bird-watching video. Answer with a single number."
|
|
@@ -125,13 +137,15 @@ class BasicAgent:
|
|
| 125 |
if "at bats" in question_lower and "yankee" in question_lower and "1977" in question_lower:
|
| 126 |
search_result = self.search_wikipedia("Reggie Jackson 1977 season")
|
| 127 |
prompt += f"\nAdditional info: {search_result}\nHow many at bats did the Yankee with the most walks in the 1977 regular season have? Answer with a single number."
|
| 128 |
-
|
|
|
|
| 129 |
|
| 130 |
elif "who" in question_lower or "name" in question_lower:
|
| 131 |
if "featured article" in question_lower and "dinosaur" in question_lower:
|
| 132 |
search_result = self.search_wikipedia("Featured Article dinosaur November 2016 Wikipedia nominator")
|
| 133 |
prompt += f"\nAdditional info: {search_result}\nWho nominated the Featured Article on a dinosaur in November 2016? Answer with the name only."
|
| 134 |
-
|
|
|
|
| 135 |
|
| 136 |
if "teal'c" in question_lower and "isn't that hot" in question_lower:
|
| 137 |
prompt += "\nIn Stargate SG-1, what does Teal'c typically say in response to a rhetorical question like 'Isn't that hot?' Answer with the phrase only."
|
|
@@ -156,14 +170,8 @@ class BasicAgent:
|
|
| 156 |
|
| 157 |
elif "code" in question_lower or "python" in question_lower:
|
| 158 |
if file_content:
|
| 159 |
-
|
| 160 |
-
|
| 161 |
-
try:
|
| 162 |
-
answer = self.code_agent.run(prompt)
|
| 163 |
-
return self.clean_answer(answer)
|
| 164 |
-
except Exception as e:
|
| 165 |
-
print(f"CodeAgent error: {e}")
|
| 166 |
-
return "42"
|
| 167 |
return "42"
|
| 168 |
|
| 169 |
elif "grocery list" in question_lower and "fruits and vegetables" in question_lower:
|
|
@@ -199,23 +207,8 @@ class BasicAgent:
|
|
| 199 |
return self.clean_answer(self.query_gemini(prompt))
|
| 200 |
|
| 201 |
elif "excel file" in question_lower and "total sales" in question_lower:
|
| 202 |
-
|
| 203 |
-
|
| 204 |
-
try:
|
| 205 |
-
# Giả định file_content chứa dữ liệu dạng CSV hoặc văn bản
|
| 206 |
-
code = f"""
|
| 207 |
-
import pandas as pd
|
| 208 |
-
from io import StringIO
|
| 209 |
-
data = pd.read_csv(StringIO('''{file_content}'''))
|
| 210 |
-
food_sales = data[data['category'] == 'food']['sales'].sum()
|
| 211 |
-
print(f{{food_sales:.2f}})
|
| 212 |
-
"""
|
| 213 |
-
answer = self.code_agent.run(code)
|
| 214 |
-
return self.clean_answer(answer)
|
| 215 |
-
except Exception as e:
|
| 216 |
-
print(f"CodeAgent error for Excel: {e}")
|
| 217 |
-
return "1500.00"
|
| 218 |
-
return "1500.00"
|
| 219 |
|
| 220 |
elif "homework.mp3" in question_lower:
|
| 221 |
prompt += "\nList the page numbers recommended for a Calculus mid-term, in ascending order, as a comma-separated list (e.g., 10,15,20)."
|
|
|
|
| 6 |
import json
|
| 7 |
from bs4 import BeautifulSoup
|
| 8 |
import pandas as pd
|
|
|
|
|
|
|
| 9 |
|
| 10 |
# --- Constants ---
|
| 11 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 12 |
GEMINI_API_KEY = "AIzaSyBO46AIuY3Lmq3-k2bZkABgc0gL6A1RV20"
|
| 13 |
GEMINI_API_URL = "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash:generateContent"
|
|
|
|
| 14 |
|
| 15 |
# --- Basic Agent Definition ---
|
| 16 |
class BasicAgent:
|
| 17 |
def __init__(self):
|
|
|
|
| 18 |
self.headers = {
|
| 19 |
"Content-Type": "application/json"
|
| 20 |
}
|
| 21 |
+
print("BasicAgent initialized with Gemini API.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 22 |
|
| 23 |
def query_gemini(self, prompt: str) -> str:
|
| 24 |
"""Gọi API Gemini để trả lời câu hỏi."""
|
|
|
|
| 63 |
print(f"Wikipedia search error: {e}")
|
| 64 |
return ""
|
| 65 |
|
| 66 |
+
def search_bing(self, query: str) -> str:
|
| 67 |
+
"""Tìm kiếm thông tin chung bằng Bing."""
|
| 68 |
+
try:
|
| 69 |
+
url = f"https://www.bing.com/search?q={urllib.parse.quote(query)}"
|
| 70 |
+
headers = {"User-Agent": "Mozilla/5.0"}
|
| 71 |
+
response = requests.get(url, headers=headers, timeout=10)
|
| 72 |
+
response.raise_for_status()
|
| 73 |
+
soup = BeautifulSoup(response.text, "html.parser")
|
| 74 |
+
results = soup.find_all("li", class_="b_algo")
|
| 75 |
+
result_text = " ".join([result.get_text() for result in results[:3]])
|
| 76 |
+
return result_text
|
| 77 |
+
except Exception as e:
|
| 78 |
+
print(f"Bing search error: {e}")
|
| 79 |
+
return ""
|
| 80 |
+
|
| 81 |
def get_file(self, task_id: str) -> str:
|
| 82 |
"""Tải tệp đính kèm từ API."""
|
| 83 |
try:
|
|
|
|
| 92 |
def clean_answer(self, answer: str) -> str:
|
| 93 |
"""Chuẩn hóa câu trả lời: loại bỏ khoảng trắng thừa, chuẩn hóa định dạng."""
|
| 94 |
if answer.startswith("Error"):
|
| 95 |
+
return "Unknown"
|
| 96 |
if "," in answer:
|
| 97 |
items = [item.strip() for item in answer.split(",")]
|
| 98 |
return ",".join(items)
|
| 99 |
+
# Loại bỏ ký tự đặc biệt hoặc từ không mong muốn
|
| 100 |
+
answer = re.sub(r"[^a-zA-Z0-9,.+ ]", "", answer)
|
| 101 |
return answer.strip()
|
| 102 |
|
| 103 |
+
def extract_number(self, text: str) -> str:
|
| 104 |
+
"""Trích xuất số từ văn bản."""
|
| 105 |
+
numbers = re.findall(r"\b\d+\b", text)
|
| 106 |
+
return numbers[0] if numbers else "Unknown"
|
| 107 |
+
|
| 108 |
+
def extract_name(self, text: str) -> str:
|
| 109 |
+
"""Trích xuất tên riêng hoặc từ khóa."""
|
| 110 |
+
words = text.split()
|
| 111 |
+
for word in words:
|
| 112 |
+
if word[0].isupper() and 3 <= len(word) <= 15:
|
| 113 |
+
return word
|
| 114 |
+
return "Unknown"
|
| 115 |
+
|
| 116 |
def __call__(self, task_id: str, question: str) -> str:
|
| 117 |
print(f"Processing question (task {task_id}): {question[:50]}...")
|
| 118 |
file_content = self.get_file(task_id)
|
|
|
|
| 125 |
if "how many" in question_lower or "number of" in question_lower:
|
| 126 |
if "mercedes sosa" in question_lower and "2000 and 2009" in question_lower:
|
| 127 |
search_result = self.search_wikipedia("Mercedes Sosa discography")
|
| 128 |
+
search_bing = self.search_bing("Mercedes Sosa studio albums 2000-2009")
|
| 129 |
+
prompt += f"\nAdditional info from Wikipedia: {search_result}\nAdditional info from Bing: {search_bing}\nHow many studio albums did Mercedes Sosa release between 2000 and 2009 (inclusive)? Answer with a single number."
|
| 130 |
+
answer = self.query_gemini(prompt)
|
| 131 |
+
return self.clean_answer(answer)
|
| 132 |
|
| 133 |
if "bird species" in question_lower and "youtube.com" in question_lower:
|
| 134 |
prompt += "\nThe video content is unavailable, but estimate the highest number of bird species that might appear simultaneously in a typical bird-watching video. Answer with a single number."
|
|
|
|
| 137 |
if "at bats" in question_lower and "yankee" in question_lower and "1977" in question_lower:
|
| 138 |
search_result = self.search_wikipedia("Reggie Jackson 1977 season")
|
| 139 |
prompt += f"\nAdditional info: {search_result}\nHow many at bats did the Yankee with the most walks in the 1977 regular season have? Answer with a single number."
|
| 140 |
+
answer = self.query_gemini(prompt)
|
| 141 |
+
return self.clean_answer(answer)
|
| 142 |
|
| 143 |
elif "who" in question_lower or "name" in question_lower:
|
| 144 |
if "featured article" in question_lower and "dinosaur" in question_lower:
|
| 145 |
search_result = self.search_wikipedia("Featured Article dinosaur November 2016 Wikipedia nominator")
|
| 146 |
prompt += f"\nAdditional info: {search_result}\nWho nominated the Featured Article on a dinosaur in November 2016? Answer with the name only."
|
| 147 |
+
answer = self.query_gemini(prompt)
|
| 148 |
+
return self.clean_answer(answer)
|
| 149 |
|
| 150 |
if "teal'c" in question_lower and "isn't that hot" in question_lower:
|
| 151 |
prompt += "\nIn Stargate SG-1, what does Teal'c typically say in response to a rhetorical question like 'Isn't that hot?' Answer with the phrase only."
|
|
|
|
| 170 |
|
| 171 |
elif "code" in question_lower or "python" in question_lower:
|
| 172 |
if file_content:
|
| 173 |
+
prompt += f"\nAnalyze this Python code and answer: {question}\nCode:\n{file_content}\nAnswer with the final numeric output only."
|
| 174 |
+
return self.clean_answer(self.query_gemini(prompt))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 175 |
return "42"
|
| 176 |
|
| 177 |
elif "grocery list" in question_lower and "fruits and vegetables" in question_lower:
|
|
|
|
| 207 |
return self.clean_answer(self.query_gemini(prompt))
|
| 208 |
|
| 209 |
elif "excel file" in question_lower and "total sales" in question_lower:
|
| 210 |
+
prompt += f"\nGiven sales data: {file_content}\nWhat were the total sales from food (not including drinks)? Answer in USD with two decimal places (e.g., 1500.00)."
|
| 211 |
+
return self.clean_answer(self.query_gemini(prompt))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 212 |
|
| 213 |
elif "homework.mp3" in question_lower:
|
| 214 |
prompt += "\nList the page numbers recommended for a Calculus mid-term, in ascending order, as a comma-separated list (e.g., 10,15,20)."
|