submit
Browse files- __pycache__/crew.cpython-310.pyc +0 -0
- __pycache__/utils.cpython-310.pyc +0 -0
- app.py +92 -146
- app_or.py +153 -0
- config/agents.yaml +14 -9
- config/tasks.yaml +3 -2
- crew.py +170 -125
- tools/__pycache__/__init__.cpython-310.pyc +0 -0
- tools/__pycache__/ai_tools.cpython-310.pyc +0 -0
- tools/__pycache__/arithmetic_tools.cpython-310.pyc +0 -0
- tools/ai_tools.py +50 -1
__pycache__/crew.cpython-310.pyc
ADDED
|
Binary file (8.93 kB). View file
|
|
|
__pycache__/utils.cpython-310.pyc
ADDED
|
Binary file (1.94 kB). View file
|
|
|
app.py
CHANGED
|
@@ -1,152 +1,98 @@
|
|
| 1 |
-
import os
|
| 2 |
import gradio as gr
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
def ask(question, openai_api_key, gemini_api_key, anthropic_api_key, file_name = ""):
|
| 7 |
-
"""
|
| 8 |
-
Ask General AI Assistant a question to answer.
|
| 9 |
-
|
| 10 |
-
Args:
|
| 11 |
-
question (str): The question to answer
|
| 12 |
-
openai_api_key (str): OpenAI API key
|
| 13 |
-
gemini_api_key (str): Gemini API key
|
| 14 |
-
anthropic_api_key (str): Anthropic API key
|
| 15 |
-
file_name (str): Optional file name
|
| 16 |
-
|
| 17 |
-
Returns:
|
| 18 |
-
str: The answer to the question
|
| 19 |
-
"""
|
| 20 |
-
if not question:
|
| 21 |
-
raise gr.Error("Question is required.")
|
| 22 |
-
|
| 23 |
-
if not openai_api_key:
|
| 24 |
-
raise gr.Error("OpenAI API Key is required.")
|
| 25 |
-
|
| 26 |
-
if not gemini_api_key:
|
| 27 |
-
raise gr.Error("Gemini API Key is required.")
|
| 28 |
-
|
| 29 |
-
if not anthropic_api_key:
|
| 30 |
-
raise gr.Error("Anthropic API Key is required.")
|
| 31 |
-
|
| 32 |
-
if file_name:
|
| 33 |
-
file_name = f"data/{file_name}"
|
| 34 |
-
|
| 35 |
-
lock = threading.Lock()
|
| 36 |
-
|
| 37 |
-
with lock:
|
| 38 |
-
answer = ""
|
| 39 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 40 |
try:
|
| 41 |
-
|
| 42 |
-
os.environ["GEMINI_API_KEY"] = gemini_api_key
|
| 43 |
-
os.environ["MODEL_API_KEY"] = anthropic_api_key
|
| 44 |
-
|
| 45 |
-
answer = run_crew(question, file_name)
|
| 46 |
except Exception as e:
|
| 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 |
-
with gr.Row():
|
| 72 |
-
level = gr.Radio(
|
| 73 |
-
choices=[1, 2, 3],
|
| 74 |
-
label="GAIA Benchmark Level",
|
| 75 |
-
interactive=True,
|
| 76 |
-
scale=1
|
| 77 |
-
)
|
| 78 |
-
ground_truth = gr.Textbox(
|
| 79 |
-
label="Ground Truth",
|
| 80 |
-
interactive=True,
|
| 81 |
-
scale=1
|
| 82 |
-
)
|
| 83 |
-
file_name = gr.Textbox(
|
| 84 |
-
label="File Name",
|
| 85 |
-
interactive=True,
|
| 86 |
-
scale=2
|
| 87 |
-
)
|
| 88 |
-
with gr.Row():
|
| 89 |
-
openai_api_key = gr.Textbox(
|
| 90 |
-
label="OpenAI API Key *",
|
| 91 |
-
type="password",
|
| 92 |
-
placeholder="sk‑...",
|
| 93 |
-
interactive=True
|
| 94 |
-
)
|
| 95 |
-
gemini_api_key = gr.Textbox(
|
| 96 |
-
label="Gemini API Key *",
|
| 97 |
-
type="password",
|
| 98 |
-
interactive=True
|
| 99 |
-
)
|
| 100 |
-
anthropic_api_key = gr.Textbox(
|
| 101 |
-
label="Anthropic API Key *",
|
| 102 |
-
type="password",
|
| 103 |
-
placeholder="sk-ant-...",
|
| 104 |
-
interactive=True
|
| 105 |
-
)
|
| 106 |
-
with gr.Row():
|
| 107 |
-
clear_btn = gr.ClearButton(
|
| 108 |
-
components=[question, level, ground_truth, file_name]
|
| 109 |
-
)
|
| 110 |
-
submit_btn = gr.Button("Submit", variant="primary")
|
| 111 |
-
with gr.Column(scale=1):
|
| 112 |
-
answer = gr.Textbox(
|
| 113 |
-
label="Answer",
|
| 114 |
-
lines=1,
|
| 115 |
-
interactive=False
|
| 116 |
-
)
|
| 117 |
-
|
| 118 |
-
submit_btn.click(
|
| 119 |
-
fn=ask,
|
| 120 |
-
inputs=[question, openai_api_key, gemini_api_key, anthropic_api_key, file_name],
|
| 121 |
-
outputs=answer
|
| 122 |
-
)
|
| 123 |
-
|
| 124 |
-
QUESTION_FILE_PATH = "data/gaia_validation.jsonl"
|
| 125 |
-
|
| 126 |
-
gr.Examples(
|
| 127 |
-
label="GAIA Benchmark Level 1 Problems",
|
| 128 |
-
examples=get_questions(QUESTION_FILE_PATH, 1),
|
| 129 |
-
inputs=[question, level, ground_truth, file_name, openai_api_key, gemini_api_key, anthropic_api_key],
|
| 130 |
-
outputs=answer,
|
| 131 |
-
cache_examples=False
|
| 132 |
-
)
|
| 133 |
-
|
| 134 |
-
gr.Examples(
|
| 135 |
-
label="GAIA Benchmark Level 2 Problems",
|
| 136 |
-
examples=get_questions(QUESTION_FILE_PATH, 2),
|
| 137 |
-
inputs=[question, level, ground_truth, file_name, openai_api_key, gemini_api_key, anthropic_api_key],
|
| 138 |
-
outputs=answer,
|
| 139 |
-
cache_examples=False
|
| 140 |
-
)
|
| 141 |
-
|
| 142 |
-
gr.Examples(
|
| 143 |
-
label="GAIA Benchmark Level 3 Problems",
|
| 144 |
-
examples=get_questions(QUESTION_FILE_PATH, 3),
|
| 145 |
-
inputs=[question, level, ground_truth, file_name, openai_api_key, gemini_api_key, anthropic_api_key],
|
| 146 |
-
outputs=answer,
|
| 147 |
-
cache_examples=False
|
| 148 |
)
|
| 149 |
-
|
| 150 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 151 |
|
| 152 |
-
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
import gradio as gr
|
| 3 |
+
import requests
|
| 4 |
+
import pandas as pd
|
| 5 |
+
from crew import run_crew # This is your actual multi-agent logic
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
|
| 7 |
+
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 8 |
+
|
| 9 |
+
# Custom wrapper to integrate your real GAIA agent logic
|
| 10 |
+
class CrewAgent:
|
| 11 |
+
def __call__(self, question: str) -> str:
|
| 12 |
+
return run_crew(question, file_path="") # File support is optional here
|
| 13 |
+
|
| 14 |
+
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 15 |
+
space_id = os.getenv("SPACE_ID")
|
| 16 |
+
|
| 17 |
+
if profile:
|
| 18 |
+
username = f"{profile.username}"
|
| 19 |
+
print(f"User logged in: {username}")
|
| 20 |
+
else:
|
| 21 |
+
print("User not logged in.")
|
| 22 |
+
return "Please log in with your Hugging Face account.", None
|
| 23 |
+
|
| 24 |
+
agent = CrewAgent()
|
| 25 |
+
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
| 26 |
+
|
| 27 |
+
# Step 1: Fetch Questions
|
| 28 |
+
try:
|
| 29 |
+
response = requests.get(f"{DEFAULT_API_URL}/questions", timeout=15)
|
| 30 |
+
response.raise_for_status()
|
| 31 |
+
questions_data = response.json()
|
| 32 |
+
print(f"✅ Fetched {len(questions_data)} questions.")
|
| 33 |
+
except Exception as e:
|
| 34 |
+
return f"❌ Failed to fetch questions: {e}", None
|
| 35 |
+
|
| 36 |
+
# Step 2: Run Agent
|
| 37 |
+
results_log = []
|
| 38 |
+
answers_payload = []
|
| 39 |
+
for item in questions_data:
|
| 40 |
+
task_id = item.get("task_id")
|
| 41 |
+
question_text = item.get("question")
|
| 42 |
try:
|
| 43 |
+
submitted_answer = agent(question_text)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 44 |
except Exception as e:
|
| 45 |
+
submitted_answer = f"AGENT ERROR: {e}"
|
| 46 |
+
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
| 47 |
+
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
|
| 48 |
+
|
| 49 |
+
if not answers_payload:
|
| 50 |
+
return "⚠️ No answers generated.", pd.DataFrame(results_log)
|
| 51 |
+
|
| 52 |
+
# Step 3: Submit
|
| 53 |
+
submission = {
|
| 54 |
+
"username": username.strip(),
|
| 55 |
+
"agent_code": agent_code,
|
| 56 |
+
"answers": answers_payload,
|
| 57 |
+
}
|
| 58 |
+
|
| 59 |
+
try:
|
| 60 |
+
response = requests.post(f"{DEFAULT_API_URL}/submit", json=submission, timeout=60)
|
| 61 |
+
response.raise_for_status()
|
| 62 |
+
result_data = response.json()
|
| 63 |
+
final_status = (
|
| 64 |
+
f"✅ Submission Successful!\n"
|
| 65 |
+
f"User: {result_data.get('username')}\n"
|
| 66 |
+
f"Score: {result_data.get('score', 'N/A')}% "
|
| 67 |
+
f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
|
| 68 |
+
f"Message: {result_data.get('message', 'No message')}"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 69 |
)
|
| 70 |
+
return final_status, pd.DataFrame(results_log)
|
| 71 |
+
except Exception as e:
|
| 72 |
+
return f"❌ Submission failed: {e}", pd.DataFrame(results_log)
|
| 73 |
+
|
| 74 |
+
# --- Gradio UI ---
|
| 75 |
+
with gr.Blocks() as demo:
|
| 76 |
+
gr.Markdown("# GAIA Agent Evaluation and Submission")
|
| 77 |
+
gr.Markdown(
|
| 78 |
+
"""
|
| 79 |
+
Login and submit your answers for the Hugging Face GAIA Benchmark.
|
| 80 |
+
|
| 81 |
+
**Instructions**:
|
| 82 |
+
- Modify your GAIA agent logic in the `crew.py` file.
|
| 83 |
+
- Log in with Hugging Face.
|
| 84 |
+
- Press **Run Evaluation & Submit** to begin.
|
| 85 |
+
"""
|
| 86 |
+
)
|
| 87 |
+
|
| 88 |
+
gr.LoginButton()
|
| 89 |
+
run_button = gr.Button("Run Evaluation & Submit")
|
| 90 |
+
|
| 91 |
+
status_output = gr.Textbox(label="Status", lines=6, interactive=False)
|
| 92 |
+
results_table = gr.DataFrame(label="Submitted Answers")
|
| 93 |
+
|
| 94 |
+
run_button.click(fn=run_and_submit_all, outputs=[status_output, results_table])
|
| 95 |
|
| 96 |
+
if __name__ == "__main__":
|
| 97 |
+
print("🚀 Starting Submission App...")
|
| 98 |
+
demo.launch(debug=True)
|
app_or.py
ADDED
|
@@ -0,0 +1,153 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os, threading
|
| 2 |
+
import gradio as gr
|
| 3 |
+
from crew import run_parallel_crew
|
| 4 |
+
from crew import run_crew
|
| 5 |
+
from utils import get_questions
|
| 6 |
+
|
| 7 |
+
|
| 8 |
+
def ask(question, openai_api_key, gemini_api_key, anthropic_api_key, file_name = ""):
|
| 9 |
+
"""
|
| 10 |
+
Ask General AI Assistant a question to answer.
|
| 11 |
+
Args:
|
| 12 |
+
question (str): The question to answer
|
| 13 |
+
openai_api_key (str): OpenAI API key
|
| 14 |
+
gemini_api_key (str): Gemini API key
|
| 15 |
+
anthropic_api_key (str): Anthropic API key
|
| 16 |
+
file_name (str): Optional file name
|
| 17 |
+
Returns:
|
| 18 |
+
str: The answer to the question
|
| 19 |
+
"""
|
| 20 |
+
if not question:
|
| 21 |
+
raise gr.Error("Question is required.")
|
| 22 |
+
|
| 23 |
+
if not openai_api_key:
|
| 24 |
+
raise gr.Error("OpenAI API Key is required.")
|
| 25 |
+
|
| 26 |
+
if not gemini_api_key:
|
| 27 |
+
raise gr.Error("Gemini API Key is required.")
|
| 28 |
+
|
| 29 |
+
if not anthropic_api_key:
|
| 30 |
+
raise gr.Error("Anthropic API Key is required.")
|
| 31 |
+
|
| 32 |
+
if file_name:
|
| 33 |
+
file_name = f"data/{file_name}"
|
| 34 |
+
|
| 35 |
+
lock = threading.Lock()
|
| 36 |
+
|
| 37 |
+
with lock:
|
| 38 |
+
answer = ""
|
| 39 |
+
|
| 40 |
+
try:
|
| 41 |
+
os.environ["OPENAI_API_KEY"] = openai_api_key
|
| 42 |
+
os.environ["GEMINI_API_KEY"] = gemini_api_key
|
| 43 |
+
os.environ["MODEL_API_KEY"] = anthropic_api_key
|
| 44 |
+
|
| 45 |
+
#answer = run_parallel_crew(question, file_name)
|
| 46 |
+
answer = run_crew(question, file_name)
|
| 47 |
+
except Exception as e:
|
| 48 |
+
raise gr.Error(e)
|
| 49 |
+
finally:
|
| 50 |
+
del os.environ["OPENAI_API_KEY"]
|
| 51 |
+
del os.environ["GEMINI_API_KEY"]
|
| 52 |
+
del os.environ["MODEL_API_KEY"]
|
| 53 |
+
|
| 54 |
+
return answer
|
| 55 |
+
|
| 56 |
+
gr.close_all()
|
| 57 |
+
|
| 58 |
+
with gr.Blocks() as grady:
|
| 59 |
+
gr.Markdown("## Grady - General AI Assistant")
|
| 60 |
+
|
| 61 |
+
with gr.Tab("Solution"):
|
| 62 |
+
gr.Markdown(os.environ.get("DESCRIPTION"))
|
| 63 |
+
|
| 64 |
+
with gr.Row():
|
| 65 |
+
with gr.Column(scale=3):
|
| 66 |
+
with gr.Row():
|
| 67 |
+
question = gr.Textbox(
|
| 68 |
+
label="Question *",
|
| 69 |
+
placeholder="In the 2025 Gradio Agents & MCP Hackathon, what percentage of participants submitted a solution during the last 24 hours?",
|
| 70 |
+
interactive=True
|
| 71 |
+
)
|
| 72 |
+
with gr.Row():
|
| 73 |
+
level = gr.Radio(
|
| 74 |
+
choices=[1, 2, 3],
|
| 75 |
+
label="GAIA Benchmark Level",
|
| 76 |
+
interactive=True,
|
| 77 |
+
scale=1
|
| 78 |
+
)
|
| 79 |
+
ground_truth = gr.Textbox(
|
| 80 |
+
label="Ground Truth",
|
| 81 |
+
interactive=True,
|
| 82 |
+
scale=1
|
| 83 |
+
)
|
| 84 |
+
file_name = gr.Textbox(
|
| 85 |
+
label="File Name",
|
| 86 |
+
interactive=True,
|
| 87 |
+
scale=2
|
| 88 |
+
)
|
| 89 |
+
with gr.Row():
|
| 90 |
+
openai_api_key = gr.Textbox(
|
| 91 |
+
label="OpenAI API Key *",
|
| 92 |
+
type="password",
|
| 93 |
+
placeholder="sk‑...",
|
| 94 |
+
interactive=True
|
| 95 |
+
)
|
| 96 |
+
gemini_api_key = gr.Textbox(
|
| 97 |
+
label="Gemini API Key *",
|
| 98 |
+
type="password",
|
| 99 |
+
interactive=True
|
| 100 |
+
)
|
| 101 |
+
anthropic_api_key = gr.Textbox(
|
| 102 |
+
label="Anthropic API Key *",
|
| 103 |
+
type="password",
|
| 104 |
+
placeholder="sk-ant-...",
|
| 105 |
+
interactive=True
|
| 106 |
+
)
|
| 107 |
+
with gr.Row():
|
| 108 |
+
clear_btn = gr.ClearButton(
|
| 109 |
+
components=[question, level, ground_truth, file_name]
|
| 110 |
+
)
|
| 111 |
+
submit_btn = gr.Button("Submit", variant="primary")
|
| 112 |
+
with gr.Column(scale=1):
|
| 113 |
+
answer = gr.Textbox(
|
| 114 |
+
label="Answer",
|
| 115 |
+
lines=1,
|
| 116 |
+
interactive=False
|
| 117 |
+
)
|
| 118 |
+
|
| 119 |
+
submit_btn.click(
|
| 120 |
+
fn=ask,
|
| 121 |
+
inputs=[question, openai_api_key, gemini_api_key, anthropic_api_key, file_name],
|
| 122 |
+
outputs=answer
|
| 123 |
+
)
|
| 124 |
+
|
| 125 |
+
QUESTION_FILE_PATH = "data/gaia_validation.jsonl"
|
| 126 |
+
|
| 127 |
+
gr.Examples(
|
| 128 |
+
label="GAIA Benchmark Level 1 Problems",
|
| 129 |
+
examples=get_questions(QUESTION_FILE_PATH, 1),
|
| 130 |
+
inputs=[question, level, ground_truth, file_name, openai_api_key, gemini_api_key, anthropic_api_key],
|
| 131 |
+
outputs=answer,
|
| 132 |
+
cache_examples=False
|
| 133 |
+
)
|
| 134 |
+
|
| 135 |
+
gr.Examples(
|
| 136 |
+
label="GAIA Benchmark Level 2 Problems",
|
| 137 |
+
examples=get_questions(QUESTION_FILE_PATH, 2),
|
| 138 |
+
inputs=[question, level, ground_truth, file_name, openai_api_key, gemini_api_key, anthropic_api_key],
|
| 139 |
+
outputs=answer,
|
| 140 |
+
cache_examples=False
|
| 141 |
+
)
|
| 142 |
+
|
| 143 |
+
gr.Examples(
|
| 144 |
+
label="GAIA Benchmark Level 3 Problems",
|
| 145 |
+
examples=get_questions(QUESTION_FILE_PATH, 3),
|
| 146 |
+
inputs=[question, level, ground_truth, file_name, openai_api_key, gemini_api_key, anthropic_api_key],
|
| 147 |
+
outputs=answer,
|
| 148 |
+
cache_examples=False
|
| 149 |
+
)
|
| 150 |
+
with gr.Tab("Documentation"):
|
| 151 |
+
gr.Markdown(os.environ.get("DOCUMENTATION"))
|
| 152 |
+
|
| 153 |
+
grady.launch(mcp_server=True)
|
config/agents.yaml
CHANGED
|
@@ -5,7 +5,6 @@ web_search_agent:
|
|
| 5 |
Given a question only, search the web and answer the question: {question}
|
| 6 |
backstory: >
|
| 7 |
As an expert web search assistant, you search the web to answer the question.
|
| 8 |
-
|
| 9 |
web_browser_agent:
|
| 10 |
role: >
|
| 11 |
Web Browser Agent
|
|
@@ -13,7 +12,6 @@ web_browser_agent:
|
|
| 13 |
Given a question, URL, and action, load the URL and act, extract, or observe, and answer the question: {question}
|
| 14 |
backstory: >
|
| 15 |
As an expert browser assistant, you load the URL and act, extract, or observe to answer the question.
|
| 16 |
-
|
| 17 |
image_analysis_agent:
|
| 18 |
role: >
|
| 19 |
Image Analysis Agent
|
|
@@ -21,7 +19,6 @@ image_analysis_agent:
|
|
| 21 |
Given a question and image file, analyze the image and answer the question: {question}
|
| 22 |
backstory: >
|
| 23 |
As an expert image analysis assistant, you analyze the image to answer the question.
|
| 24 |
-
|
| 25 |
audio_analysis_agent:
|
| 26 |
role: >
|
| 27 |
Audio Analysis Agent
|
|
@@ -29,7 +26,6 @@ audio_analysis_agent:
|
|
| 29 |
Given a question and audio file, analyze the audio and answer the question: {question}
|
| 30 |
backstory: >
|
| 31 |
As an expert audio analysis assistant, you analyze the audio to answer the question.
|
| 32 |
-
|
| 33 |
video_analysis_agent:
|
| 34 |
role: >
|
| 35 |
Video Analysis Agent
|
|
@@ -37,7 +33,6 @@ video_analysis_agent:
|
|
| 37 |
Given a question and video file, analyze the video and answer the question: {question}
|
| 38 |
backstory: >
|
| 39 |
As an expert video analysis assistant, you analyze the video file to answer the question.
|
| 40 |
-
|
| 41 |
youtube_analysis_agent:
|
| 42 |
role: >
|
| 43 |
YouTube Analysis Agent
|
|
@@ -45,7 +40,6 @@ youtube_analysis_agent:
|
|
| 45 |
Given a question and YouTube URL, analyze the video and answer the question: {question}
|
| 46 |
backstory: >
|
| 47 |
As an expert YouTube analysis assistant, you analyze the video to answer the question.
|
| 48 |
-
|
| 49 |
document_analysis_agent:
|
| 50 |
role: >
|
| 51 |
Document Analysis Agent
|
|
@@ -53,7 +47,6 @@ document_analysis_agent:
|
|
| 53 |
Given a question and document file, analyze the document and answer the question: {question}
|
| 54 |
backstory: >
|
| 55 |
As an expert document analysis assistant, you analyze the document to answer the question.
|
| 56 |
-
|
| 57 |
arithmetic_agent:
|
| 58 |
role: >
|
| 59 |
Arithmetic Agent
|
|
@@ -69,7 +62,6 @@ code_generation_agent:
|
|
| 69 |
Given a question and JSON data, generate and execute code to answer the question: {question}
|
| 70 |
backstory: >
|
| 71 |
As an expert Python code generation assistant, you generate and execute code to answer the question.
|
| 72 |
-
|
| 73 |
code_execution_agent:
|
| 74 |
role: >
|
| 75 |
Code Execution Agent
|
|
@@ -77,7 +69,20 @@ code_execution_agent:
|
|
| 77 |
Given a question and Python file, execute the file to answer the question: {question}
|
| 78 |
backstory: >
|
| 79 |
As an expert Python code execution assistant, you execute code to answer the question.
|
| 80 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 81 |
manager_agent:
|
| 82 |
role: >
|
| 83 |
Manager Agent
|
|
|
|
| 5 |
Given a question only, search the web and answer the question: {question}
|
| 6 |
backstory: >
|
| 7 |
As an expert web search assistant, you search the web to answer the question.
|
|
|
|
| 8 |
web_browser_agent:
|
| 9 |
role: >
|
| 10 |
Web Browser Agent
|
|
|
|
| 12 |
Given a question, URL, and action, load the URL and act, extract, or observe, and answer the question: {question}
|
| 13 |
backstory: >
|
| 14 |
As an expert browser assistant, you load the URL and act, extract, or observe to answer the question.
|
|
|
|
| 15 |
image_analysis_agent:
|
| 16 |
role: >
|
| 17 |
Image Analysis Agent
|
|
|
|
| 19 |
Given a question and image file, analyze the image and answer the question: {question}
|
| 20 |
backstory: >
|
| 21 |
As an expert image analysis assistant, you analyze the image to answer the question.
|
|
|
|
| 22 |
audio_analysis_agent:
|
| 23 |
role: >
|
| 24 |
Audio Analysis Agent
|
|
|
|
| 26 |
Given a question and audio file, analyze the audio and answer the question: {question}
|
| 27 |
backstory: >
|
| 28 |
As an expert audio analysis assistant, you analyze the audio to answer the question.
|
|
|
|
| 29 |
video_analysis_agent:
|
| 30 |
role: >
|
| 31 |
Video Analysis Agent
|
|
|
|
| 33 |
Given a question and video file, analyze the video and answer the question: {question}
|
| 34 |
backstory: >
|
| 35 |
As an expert video analysis assistant, you analyze the video file to answer the question.
|
|
|
|
| 36 |
youtube_analysis_agent:
|
| 37 |
role: >
|
| 38 |
YouTube Analysis Agent
|
|
|
|
| 40 |
Given a question and YouTube URL, analyze the video and answer the question: {question}
|
| 41 |
backstory: >
|
| 42 |
As an expert YouTube analysis assistant, you analyze the video to answer the question.
|
|
|
|
| 43 |
document_analysis_agent:
|
| 44 |
role: >
|
| 45 |
Document Analysis Agent
|
|
|
|
| 47 |
Given a question and document file, analyze the document and answer the question: {question}
|
| 48 |
backstory: >
|
| 49 |
As an expert document analysis assistant, you analyze the document to answer the question.
|
|
|
|
| 50 |
arithmetic_agent:
|
| 51 |
role: >
|
| 52 |
Arithmetic Agent
|
|
|
|
| 62 |
Given a question and JSON data, generate and execute code to answer the question: {question}
|
| 63 |
backstory: >
|
| 64 |
As an expert Python code generation assistant, you generate and execute code to answer the question.
|
|
|
|
| 65 |
code_execution_agent:
|
| 66 |
role: >
|
| 67 |
Code Execution Agent
|
|
|
|
| 69 |
Given a question and Python file, execute the file to answer the question: {question}
|
| 70 |
backstory: >
|
| 71 |
As an expert Python code execution assistant, you execute code to answer the question.
|
| 72 |
+
translation_agent:
|
| 73 |
+
role: >
|
| 74 |
+
A language translator agent that converts text from one language to another.
|
| 75 |
+
goal: >
|
| 76 |
+
Translate any given text into the requested target language.
|
| 77 |
+
backstory: >
|
| 78 |
+
An expert polyglot AI fluent in hundreds of languages.
|
| 79 |
+
summarization_agent:
|
| 80 |
+
role: >
|
| 81 |
+
A summarizer agent that condenses long text into clear and concise summaries.
|
| 82 |
+
goal: >
|
| 83 |
+
Summarize long articles, documents, or extracted content upon request.
|
| 84 |
+
backstory: >
|
| 85 |
+
A highly trained AI editor capable of identifying the core meaning of complex passages, documents, and technical content.
|
| 86 |
manager_agent:
|
| 87 |
role: >
|
| 88 |
Manager Agent
|
config/tasks.yaml
CHANGED
|
@@ -11,8 +11,9 @@ manager_task:
|
|
| 11 |
- Arithmetic Agent requires a question and **two numbers to add, subtract, multiply, divide, or get the modulus**. In case there are more than two numbers, use the Code Generation Agent instead.
|
| 12 |
- Code Generation Agent requires a question and **JSON data**.
|
| 13 |
- Code Execution Agent requires a question and **Python file**.
|
|
|
|
|
|
|
| 14 |
In case you cannot answer the question and there is not a good coworker, delegate to the Code Generation Agent.
|
| 15 |
Question: {question}
|
| 16 |
expected_output: >
|
| 17 |
-
The answer to the question.
|
| 18 |
-
agent: manager_agent
|
|
|
|
| 11 |
- Arithmetic Agent requires a question and **two numbers to add, subtract, multiply, divide, or get the modulus**. In case there are more than two numbers, use the Code Generation Agent instead.
|
| 12 |
- Code Generation Agent requires a question and **JSON data**.
|
| 13 |
- Code Execution Agent requires a question and **Python file**.
|
| 14 |
+
- Translation Agent requires a question and **a target language** (e.g. Translate this to French: 'Hello').
|
| 15 |
+
- Summarization Agent requires a question and **a long text to summarize** (e.g. Summarize this: 'The history of AI began in...').
|
| 16 |
In case you cannot answer the question and there is not a good coworker, delegate to the Code Generation Agent.
|
| 17 |
Question: {question}
|
| 18 |
expected_output: >
|
| 19 |
+
The answer to the question.
|
|
|
crew.py
CHANGED
|
@@ -34,160 +34,145 @@ tracer_provider = register(
|
|
| 34 |
project_name="gaia"
|
| 35 |
)
|
| 36 |
|
| 37 |
-
|
| 38 |
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
|
|
|
| 43 |
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
max_iter=2,
|
| 51 |
-
tools=[AITools.web_search_tool],
|
| 52 |
-
verbose=True
|
| 53 |
-
)
|
| 54 |
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
allow_delegation=False,
|
| 60 |
-
llm=AGENT_MODEL,
|
| 61 |
-
max_iter=3,
|
| 62 |
-
tools=[AITools.web_browser_tool],
|
| 63 |
-
verbose=True
|
| 64 |
-
)
|
| 65 |
-
|
| 66 |
-
@agent
|
| 67 |
-
def image_analysis_agent(self) -> Agent:
|
| 68 |
-
return Agent(
|
| 69 |
-
config=self.agents_config["image_analysis_agent"],
|
| 70 |
-
allow_delegation=False,
|
| 71 |
-
llm=AGENT_MODEL,
|
| 72 |
-
max_iter=2,
|
| 73 |
-
tools=[AITools.image_analysis_tool],
|
| 74 |
-
verbose=True
|
| 75 |
-
)
|
| 76 |
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
|
| 81 |
-
|
| 82 |
-
|
| 83 |
-
|
| 84 |
-
tools=[AITools.audio_analysis_tool],
|
| 85 |
-
verbose=True
|
| 86 |
-
)
|
| 87 |
|
| 88 |
-
|
| 89 |
-
|
| 90 |
-
|
| 91 |
-
|
| 92 |
-
allow_delegation=False,
|
| 93 |
-
llm=AGENT_MODEL,
|
| 94 |
-
max_iter=2,
|
| 95 |
-
tools=[AITools.video_analysis_tool],
|
| 96 |
-
verbose=True
|
| 97 |
-
)
|
| 98 |
|
| 99 |
-
|
| 100 |
-
|
| 101 |
-
|
| 102 |
-
|
| 103 |
-
|
| 104 |
-
|
| 105 |
-
|
| 106 |
-
|
| 107 |
-
|
| 108 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 109 |
|
| 110 |
-
@agent
|
| 111 |
-
def document_analysis_agent(self) -> Agent:
|
| 112 |
-
return Agent(
|
| 113 |
-
config=self.agents_config["document_analysis_agent"],
|
| 114 |
-
allow_delegation=False,
|
| 115 |
-
llm=AGENT_MODEL,
|
| 116 |
-
max_iter=2,
|
| 117 |
-
tools=[AITools.document_analysis_tool],
|
| 118 |
-
verbose=True
|
| 119 |
-
)
|
| 120 |
|
| 121 |
-
|
| 122 |
-
def arithmetic_agent(self) -> Agent:
|
| 123 |
-
return Agent(
|
| 124 |
-
config=self.agents_config["document_analysis_agent"],
|
| 125 |
-
allow_delegation=False,
|
| 126 |
-
llm=AGENT_MODEL,
|
| 127 |
-
max_iter=2,
|
| 128 |
-
tools=[ArithmeticTools.add, ArithmeticTools.subtract, ArithmeticTools.multiply, ArithmeticTools.divide, ArithmeticTools.modulus],
|
| 129 |
-
verbose=True
|
| 130 |
-
)
|
| 131 |
|
| 132 |
-
|
| 133 |
-
|
| 134 |
-
|
| 135 |
-
config=self.agents_config["code_generation_agent"],
|
| 136 |
-
allow_delegation=False,
|
| 137 |
-
llm=AGENT_MODEL,
|
| 138 |
-
max_iter=3,
|
| 139 |
-
tools=[AITools.code_generation_tool],
|
| 140 |
-
verbose=True
|
| 141 |
-
)
|
| 142 |
|
| 143 |
-
|
| 144 |
-
|
| 145 |
-
|
| 146 |
-
config=self.agents_config["code_execution_agent"],
|
| 147 |
-
allow_delegation=False,
|
| 148 |
-
llm=AGENT_MODEL,
|
| 149 |
-
max_iter=3,
|
| 150 |
-
tools=[AITools.code_execution_tool],
|
| 151 |
-
verbose=True
|
| 152 |
-
)
|
| 153 |
|
| 154 |
-
|
| 155 |
-
|
| 156 |
-
|
| 157 |
-
|
| 158 |
-
|
| 159 |
-
|
| 160 |
-
|
| 161 |
-
|
| 162 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 163 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 164 |
@task
|
| 165 |
def manager_task(self) -> Task:
|
| 166 |
-
|
| 167 |
-
|
| 168 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 169 |
|
| 170 |
-
|
| 171 |
-
def crew(self) -> Crew:
|
| 172 |
return Crew(
|
| 173 |
agents=self.agents,
|
| 174 |
-
tasks=self.
|
| 175 |
verbose=True
|
| 176 |
)
|
| 177 |
|
| 178 |
def run_crew(question, file_path):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 179 |
final_question = question
|
| 180 |
-
|
| 181 |
if file_path:
|
| 182 |
-
if is_ext(file_path, ".csv") or is_ext(file_path, ".xls")
|
|
|
|
|
|
|
| 183 |
json_data = read_file_json(file_path)
|
| 184 |
final_question = f"{question} JSON data:\n{json_data}."
|
| 185 |
else:
|
| 186 |
final_question = f"{question} File path: {file_path}."
|
| 187 |
-
|
| 188 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 189 |
final_answer = get_final_answer(FINAL_ANSWER_MODEL, question, str(answer))
|
| 190 |
|
|
|
|
| 191 |
print(f"=> Initial question: {question}")
|
| 192 |
print(f"=> Final question: {final_question}")
|
| 193 |
print(f"=> Initial answer: {answer}")
|
|
@@ -195,6 +180,66 @@ def run_crew(question, file_path):
|
|
| 195 |
|
| 196 |
return final_answer
|
| 197 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 198 |
def get_final_answer(model, question, answer):
|
| 199 |
prompt_template = """
|
| 200 |
You are an expert question answering assistant. Given a question and an initial answer, your task is to provide the final answer.
|
|
|
|
| 34 |
project_name="gaia"
|
| 35 |
)
|
| 36 |
|
| 37 |
+
## Tools
|
| 38 |
|
| 39 |
+
DOCUMENT_TOOLS = [
|
| 40 |
+
AITools.document_analysis_tool,
|
| 41 |
+
AITools.summarize_tool,
|
| 42 |
+
AITools.translate_tool
|
| 43 |
+
]
|
| 44 |
|
| 45 |
+
MEDIA_TOOLS = [
|
| 46 |
+
AITools.image_analysis_tool,
|
| 47 |
+
AITools.audio_analysis_tool,
|
| 48 |
+
AITools.video_analysis_tool,
|
| 49 |
+
AITools.youtube_analysis_tool
|
| 50 |
+
]
|
|
|
|
|
|
|
|
|
|
|
|
|
| 51 |
|
| 52 |
+
WEB_TOOLS = [
|
| 53 |
+
AITools.web_search_tool,
|
| 54 |
+
AITools.web_browser_tool
|
| 55 |
+
]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 56 |
|
| 57 |
+
ARITHMETIC_TOOLS = [
|
| 58 |
+
ArithmeticTools.add,
|
| 59 |
+
ArithmeticTools.subtract,
|
| 60 |
+
ArithmeticTools.multiply,
|
| 61 |
+
ArithmeticTools.divide,
|
| 62 |
+
ArithmeticTools.modulus
|
| 63 |
+
]
|
|
|
|
|
|
|
|
|
|
| 64 |
|
| 65 |
+
CODE_TOOLS = [
|
| 66 |
+
AITools.code_generation_tool,
|
| 67 |
+
AITools.code_execution_tool
|
| 68 |
+
]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 69 |
|
| 70 |
+
#Get specific tools
|
| 71 |
+
def get_tools_for(agent_name: str):
|
| 72 |
+
if "document" in agent_name or "translation" in agent_name or "summarization" in agent_name:
|
| 73 |
+
return DOCUMENT_TOOLS
|
| 74 |
+
elif any(keyword in agent_name for keyword in ["image", "audio", "video", "youtube"]):
|
| 75 |
+
return MEDIA_TOOLS
|
| 76 |
+
elif "web_search" in agent_name or "web_browser" in agent_name:
|
| 77 |
+
return WEB_TOOLS
|
| 78 |
+
elif "code_generation" in agent_name or "code_execution" in agent_name:
|
| 79 |
+
return CODE_TOOLS
|
| 80 |
+
elif "arithmetic" in agent_name:
|
| 81 |
+
return ARITHMETIC_TOOLS
|
| 82 |
+
elif "manager" in agent_name:
|
| 83 |
+
return []
|
| 84 |
+
else:
|
| 85 |
+
return []
|
| 86 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 87 |
|
| 88 |
+
#CrewAIInstrumentor().instrument(tracer_provider=tracer_provider)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 89 |
|
| 90 |
+
#@CrewBase
|
| 91 |
+
class GAIACrew():
|
| 92 |
+
tasks: List[Task]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 93 |
|
| 94 |
+
def __init__(self):
|
| 95 |
+
self.agents_config = self._load_yaml("config/agents.yaml")
|
| 96 |
+
self.tasks_config = self._load_yaml("config/tasks.yaml")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 97 |
|
| 98 |
+
def _load_yaml(self, path):
|
| 99 |
+
import yaml
|
| 100 |
+
with open(path, "r") as f:
|
| 101 |
+
return yaml.safe_load(f)
|
| 102 |
+
|
| 103 |
+
@property
|
| 104 |
+
def agents(self) -> List[Agent]:
|
| 105 |
+
agents = []
|
| 106 |
+
for name in self.agents_config:
|
| 107 |
+
config = self.agents_config[name]
|
| 108 |
+
if config is None:
|
| 109 |
+
print(f"❌ Agent config for '{name}' is None!")
|
| 110 |
+
continue
|
| 111 |
|
| 112 |
+
full_config = {**config, "name": name}
|
| 113 |
+
print(f"✅ Creating agent: {name}")
|
| 114 |
+
|
| 115 |
+
agents.append(Agent(
|
| 116 |
+
config=full_config,
|
| 117 |
+
allow_delegation=("manager" in name),
|
| 118 |
+
llm=MANAGER_MODEL if "manager" in name else AGENT_MODEL,
|
| 119 |
+
max_iter=5 if "manager" in name else 2,
|
| 120 |
+
tools=get_tools_for(name),
|
| 121 |
+
verbose=True
|
| 122 |
+
))
|
| 123 |
+
return agents
|
| 124 |
+
|
| 125 |
@task
|
| 126 |
def manager_task(self) -> Task:
|
| 127 |
+
# Build the Task object from your YAML
|
| 128 |
+
task = Task(config=self.tasks_config["manager_task"])
|
| 129 |
+
|
| 130 |
+
# Find the Agent instance whose YAML key is "manager_agent"
|
| 131 |
+
agent_list = self.agents
|
| 132 |
+
name_list = list(self.agents_config.keys())
|
| 133 |
+
for idx, name in enumerate(name_list):
|
| 134 |
+
if name == "manager_agent":
|
| 135 |
+
task.agent = agent_list[idx]
|
| 136 |
+
break
|
| 137 |
+
|
| 138 |
+
return task
|
| 139 |
|
| 140 |
+
def get_crew(self) -> Crew:
|
|
|
|
| 141 |
return Crew(
|
| 142 |
agents=self.agents,
|
| 143 |
+
tasks=[self.manager_task()],
|
| 144 |
verbose=True
|
| 145 |
)
|
| 146 |
|
| 147 |
def run_crew(question, file_path):
|
| 148 |
+
"""
|
| 149 |
+
Orchestrates the GAIA crew to answer a question, optionally with a file.
|
| 150 |
+
Args:
|
| 151 |
+
question (str): The user's question.
|
| 152 |
+
file_path (str): Optional path to a data file to include in the prompt.
|
| 153 |
+
Returns:
|
| 154 |
+
str: The final answer from the manager agent.
|
| 155 |
+
"""
|
| 156 |
+
# Build the final prompt, including file JSON if needed
|
| 157 |
final_question = question
|
|
|
|
| 158 |
if file_path:
|
| 159 |
+
if is_ext(file_path, ".csv") or is_ext(file_path, ".xls") \
|
| 160 |
+
or is_ext(file_path, ".xlsx") or is_ext(file_path, ".json") \
|
| 161 |
+
or is_ext(file_path, ".jsonl"):
|
| 162 |
json_data = read_file_json(file_path)
|
| 163 |
final_question = f"{question} JSON data:\n{json_data}."
|
| 164 |
else:
|
| 165 |
final_question = f"{question} File path: {file_path}."
|
| 166 |
+
|
| 167 |
+
# Instantiate the crew and kick off the workflow
|
| 168 |
+
crew_instance = GAIACrew()
|
| 169 |
+
crew = crew_instance.get_crew()
|
| 170 |
+
answer = crew.kickoff(inputs={"question": final_question})
|
| 171 |
+
|
| 172 |
+
# Post-process through the final-answer model
|
| 173 |
final_answer = get_final_answer(FINAL_ANSWER_MODEL, question, str(answer))
|
| 174 |
|
| 175 |
+
# Debug logging
|
| 176 |
print(f"=> Initial question: {question}")
|
| 177 |
print(f"=> Final question: {final_question}")
|
| 178 |
print(f"=> Initial answer: {answer}")
|
|
|
|
| 180 |
|
| 181 |
return final_answer
|
| 182 |
|
| 183 |
+
import concurrent.futures
|
| 184 |
+
|
| 185 |
+
def run_parallel_crew(question: str, file_path: str):
|
| 186 |
+
"""
|
| 187 |
+
1) Prepares the prompt (including file data if any).
|
| 188 |
+
2) Runs every non-manager agent in parallel on that prompt.
|
| 189 |
+
3) Gathers their raw outputs.
|
| 190 |
+
4) Sends a combined prompt to the manager_agent for the final answer.
|
| 191 |
+
"""
|
| 192 |
+
# 1) Build the final prompt
|
| 193 |
+
final_question = question
|
| 194 |
+
if file_path:
|
| 195 |
+
if is_ext(file_path, ".csv") or is_ext(file_path, ".xls") \
|
| 196 |
+
or is_ext(file_path, ".xlsx") or is_ext(file_path, ".json") \
|
| 197 |
+
or is_ext(file_path, ".jsonl"):
|
| 198 |
+
json_data = read_file_json(file_path)
|
| 199 |
+
final_question = f"{question} JSON data:\n{json_data}."
|
| 200 |
+
else:
|
| 201 |
+
final_question = f"{question} File path: {file_path}."
|
| 202 |
+
|
| 203 |
+
# 2) Instantiate your crew and split manager vs workers
|
| 204 |
+
crew_instance = GAIACrew()
|
| 205 |
+
names = list(crew_instance.agents_config.keys())
|
| 206 |
+
agents = crew_instance.agents
|
| 207 |
+
pairs = list(zip(names, agents))
|
| 208 |
+
|
| 209 |
+
workers = [(n, a) for n, a in pairs if n != "manager_agent"]
|
| 210 |
+
manager_name, manager = next((n, a) for n, a in pairs if n == "manager_agent")
|
| 211 |
+
|
| 212 |
+
# 3) Run workers in parallel, giving each the plain-string prompt
|
| 213 |
+
results = {}
|
| 214 |
+
with concurrent.futures.ThreadPoolExecutor(max_workers=len(workers)) as pool:
|
| 215 |
+
future_to_name = {
|
| 216 |
+
pool.submit(agent.kickoff, final_question): name
|
| 217 |
+
for name, agent in workers
|
| 218 |
+
}
|
| 219 |
+
for fut in concurrent.futures.as_completed(future_to_name):
|
| 220 |
+
name = future_to_name[fut]
|
| 221 |
+
try:
|
| 222 |
+
results[name] = fut.result()
|
| 223 |
+
except Exception as e:
|
| 224 |
+
results[name] = f"<error: {e}>"
|
| 225 |
+
|
| 226 |
+
# 4) Compose a manager prompt with all the raw outputs
|
| 227 |
+
combined = "\n\n".join(f"--- {n} output ---\n{out}"
|
| 228 |
+
for n, out in results.items())
|
| 229 |
+
manager_prompt = (
|
| 230 |
+
f"You have received these reports from your coworkers:\n\n"
|
| 231 |
+
f"{combined}\n\n"
|
| 232 |
+
f"Now, based on the original question, provide the final answer.\n"
|
| 233 |
+
f"Original question: {question}"
|
| 234 |
+
)
|
| 235 |
+
|
| 236 |
+
# 5) Run the manager agent on the combined prompt
|
| 237 |
+
final = manager.kickoff(manager_prompt)
|
| 238 |
+
|
| 239 |
+
# 6) Post-process via your final-answer model
|
| 240 |
+
return get_final_answer(FINAL_ANSWER_MODEL, question, str(final))
|
| 241 |
+
|
| 242 |
+
|
| 243 |
def get_final_answer(model, question, answer):
|
| 244 |
prompt_template = """
|
| 245 |
You are an expert question answering assistant. Given a question and an initial answer, your task is to provide the final answer.
|
tools/__pycache__/__init__.cpython-310.pyc
ADDED
|
Binary file (125 Bytes). View file
|
|
|
tools/__pycache__/ai_tools.cpython-310.pyc
ADDED
|
Binary file (10.9 kB). View file
|
|
|
tools/__pycache__/arithmetic_tools.cpython-310.pyc
ADDED
|
Binary file (2.01 kB). View file
|
|
|
tools/ai_tools.py
CHANGED
|
@@ -319,4 +319,53 @@ class AITools():
|
|
| 319 |
if part.code_execution_result is not None:
|
| 320 |
return part.code_execution_result.output
|
| 321 |
except Exception as e:
|
| 322 |
-
raise RuntimeError(f"Processing failed: {str(e)}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 319 |
if part.code_execution_result is not None:
|
| 320 |
return part.code_execution_result.output
|
| 321 |
except Exception as e:
|
| 322 |
+
raise RuntimeError(f"Processing failed: {str(e)}")
|
| 323 |
+
|
| 324 |
+
@tool("Translation Tool")
|
| 325 |
+
def translate_tool(text: str, target_lang: str) -> str:
|
| 326 |
+
"""Translate a given text into the target language.
|
| 327 |
+
|
| 328 |
+
Args:
|
| 329 |
+
text (str): The text to translate.
|
| 330 |
+
target_lang (str): The target language (e.g., 'french', 'de', 'zh').
|
| 331 |
+
|
| 332 |
+
Returns:
|
| 333 |
+
str: Translated text.
|
| 334 |
+
"""
|
| 335 |
+
try:
|
| 336 |
+
client = genai.Client(api_key=os.environ["GEMINI_API_KEY"])
|
| 337 |
+
|
| 338 |
+
prompt = f"Translate this text to {target_lang}: '{text}'"
|
| 339 |
+
|
| 340 |
+
response = client.models.generate_content(
|
| 341 |
+
model="gemini-2.5-flash-preview-04-17",
|
| 342 |
+
contents=prompt
|
| 343 |
+
)
|
| 344 |
+
|
| 345 |
+
return response.text
|
| 346 |
+
except Exception as e:
|
| 347 |
+
raise RuntimeError(f"Translation failed: {str(e)}")
|
| 348 |
+
|
| 349 |
+
@tool("Summarization Tool")
|
| 350 |
+
def summarize_tool(text: str) -> str:
|
| 351 |
+
"""Summarize the given text input.
|
| 352 |
+
|
| 353 |
+
Args:
|
| 354 |
+
text (str): Long content to summarize.
|
| 355 |
+
|
| 356 |
+
Returns:
|
| 357 |
+
str: A concise summary.
|
| 358 |
+
"""
|
| 359 |
+
try:
|
| 360 |
+
client = genai.Client(api_key=os.environ["GEMINI_API_KEY"])
|
| 361 |
+
|
| 362 |
+
prompt = f"Summarize this content: {text}"
|
| 363 |
+
|
| 364 |
+
response = client.models.generate_content(
|
| 365 |
+
model="gemini-2.5-flash-preview-04-17",
|
| 366 |
+
contents=prompt
|
| 367 |
+
)
|
| 368 |
+
|
| 369 |
+
return response.text
|
| 370 |
+
except Exception as e:
|
| 371 |
+
raise RuntimeError(f"Summarization failed: {str(e)}")
|