fixes for gaia submission
Browse files
app.py
CHANGED
|
@@ -2,35 +2,20 @@ import os
|
|
| 2 |
import gradio as gr
|
| 3 |
import requests
|
| 4 |
import pandas as pd
|
| 5 |
-
from langchain.agents import create_agent
|
| 6 |
-
from langchain_google_genai import ChatGoogleGenerativeAI
|
| 7 |
-
# Agent implementation is moved to gaia_agent.py
|
| 8 |
from gaia_agent import GAIAAgent
|
| 9 |
from dotenv import load_dotenv
|
| 10 |
|
| 11 |
# Load environment variables
|
| 12 |
-
#load_dotenv()
|
| 13 |
|
| 14 |
# --- Constants ---
|
| 15 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 16 |
|
| 17 |
# --- Agent Setup ---
|
| 18 |
-
openai_key = os.getenv("OPENAI_API_KEY")
|
| 19 |
-
googleai_key = os.getenv("GOOGLE_API_KEY")
|
| 20 |
-
|
| 21 |
-
# Use OpenRouter via LangChain's ChatOpenAI
|
| 22 |
openrouter_key = os.getenv("OPENROUTER_API_KEY")
|
| 23 |
if not openrouter_key:
|
| 24 |
raise RuntimeError("Set OPENROUTER_API_KEY in your .env (OpenRouter API key)")
|
| 25 |
|
| 26 |
-
# model is created inside gaia_agent module
|
| 27 |
-
# (gaia_agent.py will initialize the ChatOpenAI model using OPENROUTER_API_KEY)
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
# The tools and GAIAAgent implementation live in gaia_agent.py now. This file
|
| 32 |
-
# imports GAIAAgent and uses it in run_and_submit_all.
|
| 33 |
-
|
| 34 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 35 |
"""
|
| 36 |
Fetches all questions, runs the GAIAAgent on them, submits all answers,
|
|
@@ -38,18 +23,18 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 38 |
"""
|
| 39 |
# --- Determine HF Space Runtime URL and Repo URL ---
|
| 40 |
space_id = os.getenv("SPACE_ID") # Get the SPACE_ID for sending link to the code
|
| 41 |
-
|
| 42 |
if profile:
|
| 43 |
username = f"{profile.username}"
|
| 44 |
print(f"User logged in: {username}")
|
| 45 |
else:
|
| 46 |
print("User not logged in.")
|
| 47 |
return "Please Login to Hugging Face with the button.", None
|
| 48 |
-
|
| 49 |
api_url = DEFAULT_API_URL
|
| 50 |
questions_url = f"{api_url}/questions"
|
| 51 |
submit_url = f"{api_url}/submit"
|
| 52 |
-
|
| 53 |
# 1. Instantiate Agent
|
| 54 |
try:
|
| 55 |
agent = GAIAAgent()
|
|
@@ -60,16 +45,18 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 60 |
# In the case of an app running as a Hugging Face space, this link points toward your codebase
|
| 61 |
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main" if space_id else "Local"
|
| 62 |
print(f"Agent code location: {agent_code}")
|
| 63 |
-
|
| 64 |
# 2. Fetch Questions
|
| 65 |
print(f"Fetching questions from: {questions_url}")
|
| 66 |
try:
|
| 67 |
response = requests.get(questions_url, timeout=15)
|
| 68 |
response.raise_for_status()
|
| 69 |
questions_data = response.json()
|
|
|
|
| 70 |
if not questions_data:
|
| 71 |
print("Fetched questions list is empty.")
|
| 72 |
return "Fetched questions list is empty or invalid format.", None
|
|
|
|
| 73 |
print(f"Fetched {len(questions_data)} questions.")
|
| 74 |
except requests.exceptions.RequestException as e:
|
| 75 |
print(f"Error fetching questions: {e}")
|
|
@@ -81,40 +68,52 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 81 |
except Exception as e:
|
| 82 |
print(f"An unexpected error occurred fetching questions: {e}")
|
| 83 |
return f"An unexpected error occurred fetching questions: {e}", None
|
| 84 |
-
|
| 85 |
-
|
| 86 |
results_log = []
|
| 87 |
answers_payload = []
|
|
|
|
| 88 |
print(f"Running agent on {len(questions_data)} questions...")
|
| 89 |
for item in questions_data:
|
| 90 |
task_id = item.get("task_id")
|
| 91 |
question_text = item.get("question")
|
|
|
|
| 92 |
if not task_id or question_text is None:
|
| 93 |
print(f"Skipping item with missing task_id or question: {item}")
|
| 94 |
continue
|
|
|
|
| 95 |
try:
|
|
|
|
|
|
|
|
|
|
|
|
|
| 96 |
submitted_answer = agent(question_text)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 97 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
| 98 |
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
|
| 99 |
except Exception as e:
|
| 100 |
-
|
| 101 |
-
|
| 102 |
-
|
| 103 |
if not answers_payload:
|
| 104 |
print("Agent did not produce any answers to submit.")
|
| 105 |
return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
|
| 106 |
-
|
| 107 |
-
# 4. Prepare Submission
|
| 108 |
submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
|
| 109 |
status_update = f"Agent finished. Submitting {len(answers_payload)} answers for user '{username}'..."
|
| 110 |
print(status_update)
|
| 111 |
-
|
| 112 |
# 5. Submit
|
| 113 |
print(f"Submitting {len(answers_payload)} answers to: {submit_url}")
|
| 114 |
try:
|
| 115 |
response = requests.post(submit_url, json=submission_data, timeout=60)
|
| 116 |
response.raise_for_status()
|
| 117 |
result_data = response.json()
|
|
|
|
| 118 |
final_status = (
|
| 119 |
f"Submission Successful!\n"
|
| 120 |
f"User: {result_data.get('username')}\n"
|
|
@@ -122,9 +121,11 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 122 |
f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
|
| 123 |
f"Message: {result_data.get('message', 'No message received.')}"
|
| 124 |
)
|
|
|
|
| 125 |
print("Submission successful.")
|
| 126 |
results_df = pd.DataFrame(results_log)
|
| 127 |
return final_status, results_df
|
|
|
|
| 128 |
except requests.exceptions.HTTPError as e:
|
| 129 |
error_detail = f"Server responded with status {e.response.status_code}."
|
| 130 |
try:
|
|
@@ -132,55 +133,56 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 132 |
error_detail += f" Detail: {error_json.get('detail', e.response.text)}"
|
| 133 |
except requests.exceptions.JSONDecodeError:
|
| 134 |
error_detail += f" Response: {e.response.text[:500]}"
|
|
|
|
| 135 |
status_message = f"Submission Failed: {error_detail}"
|
| 136 |
print(status_message)
|
| 137 |
results_df = pd.DataFrame(results_log)
|
| 138 |
return status_message, results_df
|
|
|
|
| 139 |
except requests.exceptions.Timeout:
|
| 140 |
status_message = "Submission Failed: The request timed out."
|
| 141 |
print(status_message)
|
| 142 |
results_df = pd.DataFrame(results_log)
|
| 143 |
return status_message, results_df
|
|
|
|
| 144 |
except requests.exceptions.RequestException as e:
|
| 145 |
status_message = f"Submission Failed: Network error - {e}"
|
| 146 |
print(status_message)
|
| 147 |
results_df = pd.DataFrame(results_log)
|
| 148 |
return status_message, results_df
|
|
|
|
| 149 |
except Exception as e:
|
| 150 |
status_message = f"An unexpected error occurred during submission: {e}"
|
| 151 |
print(status_message)
|
| 152 |
results_df = pd.DataFrame(results_log)
|
| 153 |
return status_message, results_df
|
| 154 |
|
| 155 |
-
|
| 156 |
# --- Build Gradio Interface using Blocks ---
|
| 157 |
with gr.Blocks() as demo:
|
| 158 |
gr.Markdown("# GAIA Benchmark Agent Evaluation")
|
| 159 |
gr.Markdown(
|
| 160 |
"""
|
| 161 |
-
|
| 162 |
-
|
| 163 |
-
|
| 164 |
-
|
| 165 |
-
|
| 166 |
-
|
| 167 |
-
|
| 168 |
-
|
| 169 |
-
|
| 170 |
-
|
| 171 |
-
|
| 172 |
-
|
| 173 |
-
|
| 174 |
"""
|
| 175 |
)
|
| 176 |
-
|
| 177 |
gr.LoginButton()
|
| 178 |
-
|
| 179 |
run_button = gr.Button("Run Evaluation & Submit All Answers", variant="primary")
|
| 180 |
-
|
| 181 |
status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
|
| 182 |
results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
|
| 183 |
-
|
| 184 |
run_button.click(
|
| 185 |
fn=run_and_submit_all,
|
| 186 |
outputs=[status_output, results_table]
|
|
@@ -192,27 +194,27 @@ if __name__ == "__main__":
|
|
| 192 |
# Check for required environment variables
|
| 193 |
space_host_startup = os.getenv("SPACE_HOST")
|
| 194 |
space_id_startup = os.getenv("SPACE_ID")
|
| 195 |
-
|
| 196 |
|
| 197 |
if space_host_startup:
|
| 198 |
print(f"✅ SPACE_HOST found: {space_host_startup}")
|
| 199 |
print(f" Runtime URL should be: https://{space_host_startup}.hf.space")
|
| 200 |
else:
|
| 201 |
-
print("ℹ️
|
| 202 |
-
|
| 203 |
if space_id_startup:
|
| 204 |
print(f"✅ SPACE_ID found: {space_id_startup}")
|
| 205 |
print(f" Repo URL: https://huggingface.co/spaces/{space_id_startup}")
|
| 206 |
print(f" Repo Tree URL: https://huggingface.co/spaces/{space_id_startup}/tree/main")
|
| 207 |
else:
|
| 208 |
-
print("ℹ️
|
| 209 |
-
|
| 210 |
-
if
|
| 211 |
-
print("✅
|
| 212 |
else:
|
| 213 |
-
print("⚠️
|
| 214 |
-
|
| 215 |
print("-"*(60 + len(" App Starting ")) + "\n")
|
| 216 |
-
|
| 217 |
print("Launching Gradio Interface for GAIA Agent Evaluation...")
|
| 218 |
-
|
|
|
|
|
|
| 2 |
import gradio as gr
|
| 3 |
import requests
|
| 4 |
import pandas as pd
|
|
|
|
|
|
|
|
|
|
| 5 |
from gaia_agent import GAIAAgent
|
| 6 |
from dotenv import load_dotenv
|
| 7 |
|
| 8 |
# Load environment variables
|
| 9 |
+
# load_dotenv()
|
| 10 |
|
| 11 |
# --- Constants ---
|
| 12 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 13 |
|
| 14 |
# --- Agent Setup ---
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
openrouter_key = os.getenv("OPENROUTER_API_KEY")
|
| 16 |
if not openrouter_key:
|
| 17 |
raise RuntimeError("Set OPENROUTER_API_KEY in your .env (OpenRouter API key)")
|
| 18 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 20 |
"""
|
| 21 |
Fetches all questions, runs the GAIAAgent on them, submits all answers,
|
|
|
|
| 23 |
"""
|
| 24 |
# --- Determine HF Space Runtime URL and Repo URL ---
|
| 25 |
space_id = os.getenv("SPACE_ID") # Get the SPACE_ID for sending link to the code
|
| 26 |
+
|
| 27 |
if profile:
|
| 28 |
username = f"{profile.username}"
|
| 29 |
print(f"User logged in: {username}")
|
| 30 |
else:
|
| 31 |
print("User not logged in.")
|
| 32 |
return "Please Login to Hugging Face with the button.", None
|
| 33 |
+
|
| 34 |
api_url = DEFAULT_API_URL
|
| 35 |
questions_url = f"{api_url}/questions"
|
| 36 |
submit_url = f"{api_url}/submit"
|
| 37 |
+
|
| 38 |
# 1. Instantiate Agent
|
| 39 |
try:
|
| 40 |
agent = GAIAAgent()
|
|
|
|
| 45 |
# In the case of an app running as a Hugging Face space, this link points toward your codebase
|
| 46 |
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main" if space_id else "Local"
|
| 47 |
print(f"Agent code location: {agent_code}")
|
| 48 |
+
|
| 49 |
# 2. Fetch Questions
|
| 50 |
print(f"Fetching questions from: {questions_url}")
|
| 51 |
try:
|
| 52 |
response = requests.get(questions_url, timeout=15)
|
| 53 |
response.raise_for_status()
|
| 54 |
questions_data = response.json()
|
| 55 |
+
|
| 56 |
if not questions_data:
|
| 57 |
print("Fetched questions list is empty.")
|
| 58 |
return "Fetched questions list is empty or invalid format.", None
|
| 59 |
+
|
| 60 |
print(f"Fetched {len(questions_data)} questions.")
|
| 61 |
except requests.exceptions.RequestException as e:
|
| 62 |
print(f"Error fetching questions: {e}")
|
|
|
|
| 68 |
except Exception as e:
|
| 69 |
print(f"An unexpected error occurred fetching questions: {e}")
|
| 70 |
return f"An unexpected error occurred fetching questions: {e}", None
|
| 71 |
+
|
| 72 |
+
# 3. Run your Agent
|
| 73 |
results_log = []
|
| 74 |
answers_payload = []
|
| 75 |
+
|
| 76 |
print(f"Running agent on {len(questions_data)} questions...")
|
| 77 |
for item in questions_data:
|
| 78 |
task_id = item.get("task_id")
|
| 79 |
question_text = item.get("question")
|
| 80 |
+
|
| 81 |
if not task_id or question_text is None:
|
| 82 |
print(f"Skipping item with missing task_id or question: {item}")
|
| 83 |
continue
|
| 84 |
+
|
| 85 |
try:
|
| 86 |
+
print(f"\n{'='*50}")
|
| 87 |
+
print(f"Processing Task ID: {task_id}")
|
| 88 |
+
print(f"Question: {question_text}")
|
| 89 |
+
|
| 90 |
submitted_answer = agent(question_text)
|
| 91 |
+
|
| 92 |
+
print(f"Answer: {submitted_answer}")
|
| 93 |
+
print(f"{'='*50}\n")
|
| 94 |
+
|
| 95 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
| 96 |
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
|
| 97 |
except Exception as e:
|
| 98 |
+
print(f"Error running agent on task {task_id}: {e}")
|
| 99 |
+
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})
|
| 100 |
+
|
| 101 |
if not answers_payload:
|
| 102 |
print("Agent did not produce any answers to submit.")
|
| 103 |
return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
|
| 104 |
+
|
| 105 |
+
# 4. Prepare Submission
|
| 106 |
submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
|
| 107 |
status_update = f"Agent finished. Submitting {len(answers_payload)} answers for user '{username}'..."
|
| 108 |
print(status_update)
|
| 109 |
+
|
| 110 |
# 5. Submit
|
| 111 |
print(f"Submitting {len(answers_payload)} answers to: {submit_url}")
|
| 112 |
try:
|
| 113 |
response = requests.post(submit_url, json=submission_data, timeout=60)
|
| 114 |
response.raise_for_status()
|
| 115 |
result_data = response.json()
|
| 116 |
+
|
| 117 |
final_status = (
|
| 118 |
f"Submission Successful!\n"
|
| 119 |
f"User: {result_data.get('username')}\n"
|
|
|
|
| 121 |
f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
|
| 122 |
f"Message: {result_data.get('message', 'No message received.')}"
|
| 123 |
)
|
| 124 |
+
|
| 125 |
print("Submission successful.")
|
| 126 |
results_df = pd.DataFrame(results_log)
|
| 127 |
return final_status, results_df
|
| 128 |
+
|
| 129 |
except requests.exceptions.HTTPError as e:
|
| 130 |
error_detail = f"Server responded with status {e.response.status_code}."
|
| 131 |
try:
|
|
|
|
| 133 |
error_detail += f" Detail: {error_json.get('detail', e.response.text)}"
|
| 134 |
except requests.exceptions.JSONDecodeError:
|
| 135 |
error_detail += f" Response: {e.response.text[:500]}"
|
| 136 |
+
|
| 137 |
status_message = f"Submission Failed: {error_detail}"
|
| 138 |
print(status_message)
|
| 139 |
results_df = pd.DataFrame(results_log)
|
| 140 |
return status_message, results_df
|
| 141 |
+
|
| 142 |
except requests.exceptions.Timeout:
|
| 143 |
status_message = "Submission Failed: The request timed out."
|
| 144 |
print(status_message)
|
| 145 |
results_df = pd.DataFrame(results_log)
|
| 146 |
return status_message, results_df
|
| 147 |
+
|
| 148 |
except requests.exceptions.RequestException as e:
|
| 149 |
status_message = f"Submission Failed: Network error - {e}"
|
| 150 |
print(status_message)
|
| 151 |
results_df = pd.DataFrame(results_log)
|
| 152 |
return status_message, results_df
|
| 153 |
+
|
| 154 |
except Exception as e:
|
| 155 |
status_message = f"An unexpected error occurred during submission: {e}"
|
| 156 |
print(status_message)
|
| 157 |
results_df = pd.DataFrame(results_log)
|
| 158 |
return status_message, results_df
|
| 159 |
|
|
|
|
| 160 |
# --- Build Gradio Interface using Blocks ---
|
| 161 |
with gr.Blocks() as demo:
|
| 162 |
gr.Markdown("# GAIA Benchmark Agent Evaluation")
|
| 163 |
gr.Markdown(
|
| 164 |
"""
|
| 165 |
+
**Instructions:**
|
| 166 |
+
1. This app integrates a LangChain ReAct agent with multiple tools (calculator, Wikipedia, web search, Arxiv).
|
| 167 |
+
2. Log in to your Hugging Face account using the button below.
|
| 168 |
+
3. Click 'Run Evaluation & Submit All Answers' to fetch GAIA questions, run your agent, and submit answers.
|
| 169 |
+
|
| 170 |
+
**Agent Tools:**
|
| 171 |
+
- Mathematical operations (add, subtract, multiply, divide, modulus)
|
| 172 |
+
- Wikipedia search
|
| 173 |
+
- Web search (DuckDuckGo)
|
| 174 |
+
- Arxiv academic paper search
|
| 175 |
+
- Web Scraping Tool using BeautifulSoup
|
| 176 |
+
|
| 177 |
+
**Note:** Processing all questions may take several minutes depending on the number of questions and API response times.
|
| 178 |
"""
|
| 179 |
)
|
| 180 |
+
|
| 181 |
gr.LoginButton()
|
|
|
|
| 182 |
run_button = gr.Button("Run Evaluation & Submit All Answers", variant="primary")
|
|
|
|
| 183 |
status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
|
| 184 |
results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
|
| 185 |
+
|
| 186 |
run_button.click(
|
| 187 |
fn=run_and_submit_all,
|
| 188 |
outputs=[status_output, results_table]
|
|
|
|
| 194 |
# Check for required environment variables
|
| 195 |
space_host_startup = os.getenv("SPACE_HOST")
|
| 196 |
space_id_startup = os.getenv("SPACE_ID")
|
| 197 |
+
openrouter_key_startup = os.getenv("OPENROUTER_API_KEY")
|
| 198 |
|
| 199 |
if space_host_startup:
|
| 200 |
print(f"✅ SPACE_HOST found: {space_host_startup}")
|
| 201 |
print(f" Runtime URL should be: https://{space_host_startup}.hf.space")
|
| 202 |
else:
|
| 203 |
+
print("ℹ️ SPACE_HOST environment variable not found (running locally?).")
|
| 204 |
+
|
| 205 |
if space_id_startup:
|
| 206 |
print(f"✅ SPACE_ID found: {space_id_startup}")
|
| 207 |
print(f" Repo URL: https://huggingface.co/spaces/{space_id_startup}")
|
| 208 |
print(f" Repo Tree URL: https://huggingface.co/spaces/{space_id_startup}/tree/main")
|
| 209 |
else:
|
| 210 |
+
print("ℹ️ SPACE_ID environment variable not found (running locally?).")
|
| 211 |
+
|
| 212 |
+
if openrouter_key_startup:
|
| 213 |
+
print("✅ OPENROUTER_API_KEY found")
|
| 214 |
else:
|
| 215 |
+
print("⚠️ OPENROUTER_API_KEY not found - agent will not work without it!")
|
| 216 |
+
|
| 217 |
print("-"*(60 + len(" App Starting ")) + "\n")
|
|
|
|
| 218 |
print("Launching Gradio Interface for GAIA Agent Evaluation...")
|
| 219 |
+
|
| 220 |
+
demo.launch(debug=True, share=False)
|