Andrei Sferdian commited on
Commit ·
1c7a9a0
1
Parent(s): 8b65802
Add python code execution tool
Browse files- agent.py +24 -11
- app.py +5 -0
- python_code.py +35 -0
- tools/__init__.py +4 -1
- tools/execute_code.py +27 -0
- tools/files.py +30 -1
agent.py
CHANGED
|
@@ -1,18 +1,21 @@
|
|
| 1 |
from typing import Annotated, TypedDict
|
| 2 |
|
| 3 |
-
from langchain_community.tools import
|
| 4 |
from langchain_core.messages import AnyMessage, HumanMessage
|
| 5 |
from langchain_together import ChatTogether
|
| 6 |
from langgraph.graph import START, StateGraph
|
| 7 |
from langgraph.graph.message import add_messages
|
| 8 |
from langgraph.prebuilt import create_react_agent
|
| 9 |
|
|
|
|
| 10 |
from tools import (
|
| 11 |
GetYouTubeTranscript,
|
| 12 |
add,
|
| 13 |
divide,
|
|
|
|
| 14 |
find_wiki_page,
|
| 15 |
floor_divide,
|
|
|
|
| 16 |
get_wiki_page_section_by_title,
|
| 17 |
load_excel_file,
|
| 18 |
modulus,
|
|
@@ -22,11 +25,13 @@ from tools import (
|
|
| 22 |
)
|
| 23 |
|
| 24 |
tools = [
|
| 25 |
-
DuckDuckGoSearchRun(),
|
| 26 |
GetYouTubeTranscript(),
|
|
|
|
|
|
|
| 27 |
find_wiki_page,
|
| 28 |
get_wiki_page_section_by_title,
|
| 29 |
load_excel_file,
|
|
|
|
| 30 |
add,
|
| 31 |
subtract,
|
| 32 |
multiply,
|
|
@@ -46,12 +51,14 @@ model = ChatTogether(
|
|
| 46 |
prompt = (
|
| 47 |
"You are an agent that will undergo the GAIA test.\n"
|
| 48 |
"You will be given a question and you will have to answer it using the tools provided.\n"
|
| 49 |
-
"Respond in the requested format, if specified, otherwise use the short answer
|
| 50 |
-
"When searching for
|
| 51 |
-
" a. Use
|
|
|
|
|
|
|
|
|
|
| 52 |
" The response will be a JSON string with the page's summary and sections.\n"
|
| 53 |
-
"
|
| 54 |
-
"If the user asks for a YouTube transcript, use the GetYouTubeTranscript tool.\n"
|
| 55 |
"If the user asks for an Excel file, use the load_excel_file tool.\n"
|
| 56 |
"When you need to perform calculations, use the provided calculator tools (add, subtract, multiply, divide).\n"
|
| 57 |
"Do not performa calculations by yourself!\n"
|
|
@@ -103,12 +110,11 @@ def prepare(state: GraphState):
|
|
| 103 |
print("Question is in reverse order, reversing it...")
|
| 104 |
question = question[::-1]
|
| 105 |
|
| 106 |
-
|
|
|
|
| 107 |
return {
|
| 108 |
"messages": [
|
| 109 |
-
HumanMessage(
|
| 110 |
-
content=f"Question: {question}\nFile: {state['file_name']}"
|
| 111 |
-
)
|
| 112 |
],
|
| 113 |
}
|
| 114 |
|
|
@@ -134,3 +140,10 @@ graph.add_edge("prepare", "react")
|
|
| 134 |
graph.add_edge("prepare", "react")
|
| 135 |
|
| 136 |
agent = graph.compile()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
from typing import Annotated, TypedDict
|
| 2 |
|
| 3 |
+
from langchain_community.tools import GoogleSerperResults
|
| 4 |
from langchain_core.messages import AnyMessage, HumanMessage
|
| 5 |
from langchain_together import ChatTogether
|
| 6 |
from langgraph.graph import START, StateGraph
|
| 7 |
from langgraph.graph.message import add_messages
|
| 8 |
from langgraph.prebuilt import create_react_agent
|
| 9 |
|
| 10 |
+
import questions
|
| 11 |
from tools import (
|
| 12 |
GetYouTubeTranscript,
|
| 13 |
add,
|
| 14 |
divide,
|
| 15 |
+
execute_python_code,
|
| 16 |
find_wiki_page,
|
| 17 |
floor_divide,
|
| 18 |
+
get_text_file,
|
| 19 |
get_wiki_page_section_by_title,
|
| 20 |
load_excel_file,
|
| 21 |
modulus,
|
|
|
|
| 25 |
)
|
| 26 |
|
| 27 |
tools = [
|
|
|
|
| 28 |
GetYouTubeTranscript(),
|
| 29 |
+
GoogleSerperResults(),
|
| 30 |
+
execute_python_code,
|
| 31 |
find_wiki_page,
|
| 32 |
get_wiki_page_section_by_title,
|
| 33 |
load_excel_file,
|
| 34 |
+
get_text_file,
|
| 35 |
add,
|
| 36 |
subtract,
|
| 37 |
multiply,
|
|
|
|
| 51 |
prompt = (
|
| 52 |
"You are an agent that will undergo the GAIA test.\n"
|
| 53 |
"You will be given a question and you will have to answer it using the tools provided.\n"
|
| 54 |
+
"Respond in the requested format, if specified, otherwise use the short answer style. I.E. Question: What is the capital of France? Answer: Paris.\n"
|
| 55 |
+
"When searching for information, always follow this specific sequence:\n"
|
| 56 |
+
" a. Use the 'google_serper_results_json' tool to search for relevant information.\n"
|
| 57 |
+
" Provide the query as input to the tool.\n"
|
| 58 |
+
" The tool will return a JSON string with the search results.\n"
|
| 59 |
+
" b. If the required information is not found in the search results, use the find_wiki_page tool to locate the relevant Wikipedia page.\n"
|
| 60 |
" The response will be a JSON string with the page's summary and sections.\n"
|
| 61 |
+
" c. If the Wikipedia page exists, and the summary does not contain the required info, use get_wiki_page_section_by_title tool to get the content of the desired section.\n"
|
|
|
|
| 62 |
"If the user asks for an Excel file, use the load_excel_file tool.\n"
|
| 63 |
"When you need to perform calculations, use the provided calculator tools (add, subtract, multiply, divide).\n"
|
| 64 |
"Do not performa calculations by yourself!\n"
|
|
|
|
| 110 |
print("Question is in reverse order, reversing it...")
|
| 111 |
question = question[::-1]
|
| 112 |
|
| 113 |
+
file_name = state.get("file_name")
|
| 114 |
+
if file_name:
|
| 115 |
return {
|
| 116 |
"messages": [
|
| 117 |
+
HumanMessage(content=f"Question: {question}\nFile name: {file_name}")
|
|
|
|
|
|
|
| 118 |
],
|
| 119 |
}
|
| 120 |
|
|
|
|
| 140 |
graph.add_edge("prepare", "react")
|
| 141 |
|
| 142 |
agent = graph.compile()
|
| 143 |
+
|
| 144 |
+
if __name__ == "__main__":
|
| 145 |
+
import questions
|
| 146 |
+
|
| 147 |
+
file_bytes = questions.get_file("f918266a-b3e0-4914-865d-4faa564f1aef")
|
| 148 |
+
file_content = file_bytes.decode("utf-8")
|
| 149 |
+
exec(file_content)
|
app.py
CHANGED
|
@@ -288,6 +288,11 @@ def run_locally():
|
|
| 288 |
"question": "The attached Excel file contains the sales of menu items for a local fast-food chain. What were the total sales that the chain made from food (not including drinks)? Express your answer in USD with two decimal places.",
|
| 289 |
"file_name": "7bd855d8-463d-4ed5-93ca-5fe35145f733.xlsx",
|
| 290 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 291 |
]
|
| 292 |
|
| 293 |
max_concurrency = int(os.getenv("MAX_CONCURRENCY", 1))
|
|
|
|
| 288 |
"question": "The attached Excel file contains the sales of menu items for a local fast-food chain. What were the total sales that the chain made from food (not including drinks)? Express your answer in USD with two decimal places.",
|
| 289 |
"file_name": "7bd855d8-463d-4ed5-93ca-5fe35145f733.xlsx",
|
| 290 |
},
|
| 291 |
+
# {
|
| 292 |
+
# "task_id": "f918266a-b3e0-4914-865d-4faa564f1aef",
|
| 293 |
+
# "question": "What is the final numeric output from the attached Python code?",
|
| 294 |
+
# "file_name": "f918266a-b3e0-4914-865d-4faa564f1aef.py",
|
| 295 |
+
# },
|
| 296 |
]
|
| 297 |
|
| 298 |
max_concurrency = int(os.getenv("MAX_CONCURRENCY", 1))
|
python_code.py
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from random import randint
|
| 2 |
+
import time
|
| 3 |
+
|
| 4 |
+
class UhOh(Exception):
|
| 5 |
+
pass
|
| 6 |
+
|
| 7 |
+
class Hmm:
|
| 8 |
+
def __init__(self):
|
| 9 |
+
self.value = randint(-100, 100)
|
| 10 |
+
|
| 11 |
+
def Yeah(self):
|
| 12 |
+
if self.value == 0:
|
| 13 |
+
return True
|
| 14 |
+
else:
|
| 15 |
+
raise UhOh()
|
| 16 |
+
|
| 17 |
+
def Okay():
|
| 18 |
+
while True:
|
| 19 |
+
yield Hmm()
|
| 20 |
+
|
| 21 |
+
def keep_trying(go, first_try=True):
|
| 22 |
+
maybe = next(go)
|
| 23 |
+
try:
|
| 24 |
+
if maybe.Yeah():
|
| 25 |
+
return maybe.value
|
| 26 |
+
except UhOh:
|
| 27 |
+
if first_try:
|
| 28 |
+
print("Working...")
|
| 29 |
+
print("Please wait patiently...")
|
| 30 |
+
time.sleep(0.1)
|
| 31 |
+
return keep_trying(go, first_try=False)
|
| 32 |
+
|
| 33 |
+
if __name__ == "__main__":
|
| 34 |
+
go = Okay()
|
| 35 |
+
print(f"{keep_trying(go)}")
|
tools/__init__.py
CHANGED
|
@@ -7,7 +7,8 @@ from .calculator import (
|
|
| 7 |
power,
|
| 8 |
subtract,
|
| 9 |
)
|
| 10 |
-
from .
|
|
|
|
| 11 |
from .get_youtube_transcript import GetYouTubeTranscript
|
| 12 |
from .wikipedia import (
|
| 13 |
find_wiki_page,
|
|
@@ -19,6 +20,7 @@ __all__ = [
|
|
| 19 |
"find_wiki_page",
|
| 20 |
"get_wiki_page_section_by_title",
|
| 21 |
"load_excel_file",
|
|
|
|
| 22 |
"add",
|
| 23 |
"subtract",
|
| 24 |
"multiply",
|
|
@@ -26,4 +28,5 @@ __all__ = [
|
|
| 26 |
"power",
|
| 27 |
"modulus",
|
| 28 |
"floor_divide",
|
|
|
|
| 29 |
]
|
|
|
|
| 7 |
power,
|
| 8 |
subtract,
|
| 9 |
)
|
| 10 |
+
from .execute_code import execute_python_code
|
| 11 |
+
from .files import get_text_file, load_excel_file
|
| 12 |
from .get_youtube_transcript import GetYouTubeTranscript
|
| 13 |
from .wikipedia import (
|
| 14 |
find_wiki_page,
|
|
|
|
| 20 |
"find_wiki_page",
|
| 21 |
"get_wiki_page_section_by_title",
|
| 22 |
"load_excel_file",
|
| 23 |
+
"get_text_file",
|
| 24 |
"add",
|
| 25 |
"subtract",
|
| 26 |
"multiply",
|
|
|
|
| 28 |
"power",
|
| 29 |
"modulus",
|
| 30 |
"floor_divide",
|
| 31 |
+
"execute_python_code",
|
| 32 |
]
|
tools/execute_code.py
ADDED
|
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import subprocess
|
| 2 |
+
|
| 3 |
+
from langchain.tools import tool
|
| 4 |
+
|
| 5 |
+
|
| 6 |
+
@tool(parse_docstring=True)
|
| 7 |
+
def execute_python_code(code: str) -> str:
|
| 8 |
+
"""
|
| 9 |
+
Executes a given Python code snippet in a subprocess and returns its output.
|
| 10 |
+
|
| 11 |
+
Args:
|
| 12 |
+
code (str): The Python code to execute as a string.
|
| 13 |
+
|
| 14 |
+
Returns:
|
| 15 |
+
str: The standard output of the executed code if successful, or an error message
|
| 16 |
+
containing the standard error output if the execution fails.
|
| 17 |
+
Raises:
|
| 18 |
+
None: Any exceptions during execution are caught and handled internally.
|
| 19 |
+
"""
|
| 20 |
+
|
| 21 |
+
try:
|
| 22 |
+
process = subprocess.run(
|
| 23 |
+
["python3", "-c", code], text=True, capture_output=True, check=True
|
| 24 |
+
)
|
| 25 |
+
return process.stdout
|
| 26 |
+
except subprocess.CalledProcessError as e:
|
| 27 |
+
return f"Error: {e.stderr}"
|
tools/files.py
CHANGED
|
@@ -1,7 +1,36 @@
|
|
| 1 |
from langchain_core.tools import tool
|
| 2 |
|
| 3 |
|
| 4 |
-
@tool()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5 |
def load_excel_file(file_name: str) -> str:
|
| 6 |
"""
|
| 7 |
Loads an Excel file by its name and returns its content as a string.
|
|
|
|
| 1 |
from langchain_core.tools import tool
|
| 2 |
|
| 3 |
|
| 4 |
+
@tool(parse_docstring=True)
|
| 5 |
+
def get_text_file(file_name: str) -> str:
|
| 6 |
+
"""
|
| 7 |
+
Loads a text file by its name and returns its content as a string.
|
| 8 |
+
|
| 9 |
+
Args:
|
| 10 |
+
file_name (str): The name of the text file to load. The file name should include
|
| 11 |
+
a task ID followed by a file extension (e.g., "12345.txt").
|
| 12 |
+
|
| 13 |
+
Returns:
|
| 14 |
+
str: The content of the text file as a string.
|
| 15 |
+
"""
|
| 16 |
+
print(f"Loading text file: {file_name}")
|
| 17 |
+
try:
|
| 18 |
+
import questions
|
| 19 |
+
|
| 20 |
+
task_id, ext = file_name.rsplit(".", 1)
|
| 21 |
+
if not task_id or not ext:
|
| 22 |
+
return "Invalid file name format. Please provide a valid file name with task ID and extension."
|
| 23 |
+
|
| 24 |
+
file_content = questions.get_file(task_id)
|
| 25 |
+
if not file_content:
|
| 26 |
+
return f"File with task ID {task_id} not found."
|
| 27 |
+
|
| 28 |
+
return file_content.decode("utf-8")
|
| 29 |
+
except Exception as e:
|
| 30 |
+
return f"Error loading text file: {e}"
|
| 31 |
+
|
| 32 |
+
|
| 33 |
+
@tool(parse_docstring=True)
|
| 34 |
def load_excel_file(file_name: str) -> str:
|
| 35 |
"""
|
| 36 |
Loads an Excel file by its name and returns its content as a string.
|