Update app.py
Browse files
app.py
CHANGED
|
@@ -4,33 +4,34 @@ import requests
|
|
| 4 |
import pandas as pd
|
| 5 |
from smolagents import CodeAgent, DuckDuckGoSearchTool, LiteLLMModel
|
| 6 |
|
|
|
|
| 7 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 8 |
|
| 9 |
class BasicAgent:
|
| 10 |
def __init__(self):
|
| 11 |
-
|
| 12 |
self.model = LiteLLMModel(model_id="huggingface/Qwen/Qwen2.5-72B-Instruct")
|
| 13 |
|
|
|
|
|
|
|
|
|
|
| 14 |
self.agent = CodeAgent(
|
| 15 |
-
tools=[
|
| 16 |
model=self.model,
|
| 17 |
add_base_tools=True
|
| 18 |
)
|
| 19 |
print("Advanced smolagent initialized with LiteLLM.")
|
| 20 |
|
| 21 |
def __call__(self, question: str) -> str:
|
| 22 |
-
# Prompting for Exact Match scoring
|
| 23 |
clean_prompt = (
|
| 24 |
f"Question: {question}\n\n"
|
| 25 |
"Instructions: Solve the question above. Provide ONLY the final answer "
|
| 26 |
"value without any explanation, units, or extra text. "
|
| 27 |
-
"Do not include 'FINAL ANSWER' in your response."
|
| 28 |
)
|
| 29 |
|
| 30 |
try:
|
| 31 |
-
# The agent will reason and use tools if necessary
|
| 32 |
result = self.agent.run(clean_prompt)
|
| 33 |
-
# Ensure we return a clean string
|
| 34 |
return str(result).strip()
|
| 35 |
except Exception as e:
|
| 36 |
print(f"Error during agent execution: {e}")
|
|
@@ -46,14 +47,15 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 46 |
api_url = DEFAULT_API_URL
|
| 47 |
questions_url = f"{api_url}/questions"
|
| 48 |
submit_url = f"{api_url}/submit"
|
|
|
|
|
|
|
|
|
|
| 49 |
|
| 50 |
# 1. Instantiate Agent
|
| 51 |
try:
|
| 52 |
agent = BasicAgent()
|
| 53 |
except Exception as e:
|
| 54 |
return f"Error initializing agent: {e}", None
|
| 55 |
-
|
| 56 |
-
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
| 57 |
|
| 58 |
# 2. Fetch Questions
|
| 59 |
try:
|
|
@@ -67,6 +69,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 67 |
results_log = []
|
| 68 |
answers_payload = []
|
| 69 |
|
|
|
|
| 70 |
for item in questions_data:
|
| 71 |
task_id = item.get("task_id")
|
| 72 |
question_text = item.get("question")
|
|
@@ -74,9 +77,6 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 74 |
if not task_id or question_text is None:
|
| 75 |
continue
|
| 76 |
|
| 77 |
-
# Optional: Handle file downloads for specific tasks
|
| 78 |
-
# You could add logic here to download from f"{api_url}/files/{task_id}"
|
| 79 |
-
|
| 80 |
try:
|
| 81 |
submitted_answer = agent(question_text)
|
| 82 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
|
@@ -99,8 +99,8 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 99 |
final_status = (
|
| 100 |
f"Submission Successful!\n"
|
| 101 |
f"User: {result_data.get('username')}\n"
|
| 102 |
-
f"Score: {result_data.get('score')}% "
|
| 103 |
-
f"({result_data.get('correct_count')}/{result_data.get('total_attempted')} correct)"
|
| 104 |
)
|
| 105 |
return final_status, pd.DataFrame(results_log)
|
| 106 |
except Exception as e:
|
|
@@ -109,11 +109,8 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 109 |
# --- Gradio Interface ---
|
| 110 |
with gr.Blocks() as demo:
|
| 111 |
gr.Markdown("# GAIA Agent Evaluation Runner")
|
| 112 |
-
gr.Markdown("This agent uses `smolagents` with a Python interpreter and Web Search to solve tasks.")
|
| 113 |
-
|
| 114 |
gr.LoginButton()
|
| 115 |
run_button = gr.Button("Run Evaluation & Submit All Answers", variant="primary")
|
| 116 |
-
|
| 117 |
status_output = gr.Textbox(label="Status", lines=4)
|
| 118 |
results_table = gr.DataFrame(label="Detailed Results", wrap=True)
|
| 119 |
|
|
|
|
| 4 |
import pandas as pd
|
| 5 |
from smolagents import CodeAgent, DuckDuckGoSearchTool, LiteLLMModel
|
| 6 |
|
| 7 |
+
# --- Constants ---
|
| 8 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 9 |
|
| 10 |
class BasicAgent:
|
| 11 |
def __init__(self):
|
| 12 |
+
# Using LiteLLM to connect to Qwen
|
| 13 |
self.model = LiteLLMModel(model_id="huggingface/Qwen/Qwen2.5-72B-Instruct")
|
| 14 |
|
| 15 |
+
# Initialize search tool
|
| 16 |
+
search_tool = DuckDuckGoSearchTool()
|
| 17 |
+
|
| 18 |
self.agent = CodeAgent(
|
| 19 |
+
tools=[search_tool],
|
| 20 |
model=self.model,
|
| 21 |
add_base_tools=True
|
| 22 |
)
|
| 23 |
print("Advanced smolagent initialized with LiteLLM.")
|
| 24 |
|
| 25 |
def __call__(self, question: str) -> str:
|
|
|
|
| 26 |
clean_prompt = (
|
| 27 |
f"Question: {question}\n\n"
|
| 28 |
"Instructions: Solve the question above. Provide ONLY the final answer "
|
| 29 |
"value without any explanation, units, or extra text. "
|
| 30 |
+
"Do not include the phrase 'FINAL ANSWER' in your response."
|
| 31 |
)
|
| 32 |
|
| 33 |
try:
|
|
|
|
| 34 |
result = self.agent.run(clean_prompt)
|
|
|
|
| 35 |
return str(result).strip()
|
| 36 |
except Exception as e:
|
| 37 |
print(f"Error during agent execution: {e}")
|
|
|
|
| 47 |
api_url = DEFAULT_API_URL
|
| 48 |
questions_url = f"{api_url}/questions"
|
| 49 |
submit_url = f"{api_url}/submit"
|
| 50 |
+
|
| 51 |
+
# Define agent_code link correctly
|
| 52 |
+
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main" if space_id else "https://huggingface.co/spaces"
|
| 53 |
|
| 54 |
# 1. Instantiate Agent
|
| 55 |
try:
|
| 56 |
agent = BasicAgent()
|
| 57 |
except Exception as e:
|
| 58 |
return f"Error initializing agent: {e}", None
|
|
|
|
|
|
|
| 59 |
|
| 60 |
# 2. Fetch Questions
|
| 61 |
try:
|
|
|
|
| 69 |
results_log = []
|
| 70 |
answers_payload = []
|
| 71 |
|
| 72 |
+
# LIMITING TO FIRST 5 FOR TESTING - Remove [:5] to do all
|
| 73 |
for item in questions_data:
|
| 74 |
task_id = item.get("task_id")
|
| 75 |
question_text = item.get("question")
|
|
|
|
| 77 |
if not task_id or question_text is None:
|
| 78 |
continue
|
| 79 |
|
|
|
|
|
|
|
|
|
|
| 80 |
try:
|
| 81 |
submitted_answer = agent(question_text)
|
| 82 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
|
|
|
| 99 |
final_status = (
|
| 100 |
f"Submission Successful!\n"
|
| 101 |
f"User: {result_data.get('username')}\n"
|
| 102 |
+
f"Score: {result_data.get('score', 0)}% "
|
| 103 |
+
f"({result_data.get('correct_count', 0)}/{result_data.get('total_attempted', 0)} correct)"
|
| 104 |
)
|
| 105 |
return final_status, pd.DataFrame(results_log)
|
| 106 |
except Exception as e:
|
|
|
|
| 109 |
# --- Gradio Interface ---
|
| 110 |
with gr.Blocks() as demo:
|
| 111 |
gr.Markdown("# GAIA Agent Evaluation Runner")
|
|
|
|
|
|
|
| 112 |
gr.LoginButton()
|
| 113 |
run_button = gr.Button("Run Evaluation & Submit All Answers", variant="primary")
|
|
|
|
| 114 |
status_output = gr.Textbox(label="Status", lines=4)
|
| 115 |
results_table = gr.DataFrame(label="Detailed Results", wrap=True)
|
| 116 |
|