Update app.py
#642
by khm12342 - opened
app.py
CHANGED
|
@@ -1,196 +1,208 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import os
|
|
|
|
|
|
|
|
|
|
| 2 |
import gradio as gr
|
| 3 |
-
import requests
|
| 4 |
-
import inspect
|
| 5 |
import pandas as pd
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
|
| 7 |
-
# (Keep Constants as is)
|
| 8 |
-
# --- Constants ---
|
| 9 |
-
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 10 |
|
| 11 |
-
# --- Basic Agent Definition ---
|
| 12 |
-
# ----- THIS IS WERE YOU CAN BUILD WHAT YOU WANT ------
|
| 13 |
class BasicAgent:
|
| 14 |
def __init__(self):
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 42 |
try:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 43 |
agent = BasicAgent()
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 53 |
try:
|
| 54 |
-
|
|
|
|
| 55 |
response.raise_for_status()
|
| 56 |
-
|
| 57 |
-
if not
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
print(f"Error decoding JSON response from questions endpoint: {e}")
|
| 66 |
-
print(f"Response text: {response.text[:500]}")
|
| 67 |
-
return f"Error decoding server response for questions: {e}", None
|
| 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 |
-
print(f"Running agent on {len(questions_data)} questions...")
|
| 76 |
-
for item in questions_data:
|
| 77 |
task_id = item.get("task_id")
|
| 78 |
-
|
| 79 |
-
if not task_id or
|
| 80 |
-
print(f"Skipping item with missing task_id or question: {item}")
|
| 81 |
continue
|
| 82 |
try:
|
| 83 |
-
|
| 84 |
-
|
| 85 |
-
|
| 86 |
-
|
| 87 |
-
|
| 88 |
-
|
| 89 |
-
|
| 90 |
-
|
| 91 |
-
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
|
| 97 |
-
|
| 98 |
-
|
| 99 |
-
|
| 100 |
-
|
|
|
|
|
|
|
| 101 |
try:
|
| 102 |
-
response = requests.post(
|
| 103 |
response.raise_for_status()
|
| 104 |
-
|
| 105 |
-
|
| 106 |
-
|
| 107 |
-
f"User: {
|
| 108 |
-
f"Overall Score: {
|
| 109 |
-
f"({
|
| 110 |
-
f"
|
|
|
|
| 111 |
)
|
| 112 |
-
|
| 113 |
-
|
| 114 |
-
return
|
| 115 |
-
|
| 116 |
-
|
| 117 |
-
try:
|
| 118 |
-
error_json = e.response.json()
|
| 119 |
-
error_detail += f" Detail: {error_json.get('detail', e.response.text)}"
|
| 120 |
-
except requests.exceptions.JSONDecodeError:
|
| 121 |
-
error_detail += f" Response: {e.response.text[:500]}"
|
| 122 |
-
status_message = f"Submission Failed: {error_detail}"
|
| 123 |
-
print(status_message)
|
| 124 |
-
results_df = pd.DataFrame(results_log)
|
| 125 |
-
return status_message, results_df
|
| 126 |
-
except requests.exceptions.Timeout:
|
| 127 |
-
status_message = "Submission Failed: The request timed out."
|
| 128 |
-
print(status_message)
|
| 129 |
-
results_df = pd.DataFrame(results_log)
|
| 130 |
-
return status_message, results_df
|
| 131 |
-
except requests.exceptions.RequestException as e:
|
| 132 |
-
status_message = f"Submission Failed: Network error - {e}"
|
| 133 |
-
print(status_message)
|
| 134 |
-
results_df = pd.DataFrame(results_log)
|
| 135 |
-
return status_message, results_df
|
| 136 |
-
except Exception as e:
|
| 137 |
-
status_message = f"An unexpected error occurred during submission: {e}"
|
| 138 |
-
print(status_message)
|
| 139 |
-
results_df = pd.DataFrame(results_log)
|
| 140 |
-
return status_message, results_df
|
| 141 |
-
|
| 142 |
-
|
| 143 |
-
# --- Build Gradio Interface using Blocks ---
|
| 144 |
with gr.Blocks() as demo:
|
| 145 |
-
gr.Markdown("#
|
| 146 |
gr.Markdown(
|
| 147 |
-
""
|
| 148 |
-
|
| 149 |
-
|
| 150 |
-
|
| 151 |
-
2. Log in to your Hugging Face account using the button below. This uses your HF username for submission.
|
| 152 |
-
3. Click 'Run Evaluation & Submit All Answers' to fetch questions, run your agent, submit answers, and see the score.
|
| 153 |
-
|
| 154 |
-
---
|
| 155 |
-
**Disclaimers:**
|
| 156 |
-
Once clicking on the "submit button, it can take quite some time ( this is the time for the agent to go through all the questions).
|
| 157 |
-
This space provides a basic setup and is intentionally sub-optimal to encourage you to develop your own, more robust solution. For instance for the delay process of the submit button, a solution could be to cache the answers and submit in a seperate action or even to answer the questions in async.
|
| 158 |
-
"""
|
| 159 |
)
|
| 160 |
-
|
| 161 |
gr.LoginButton()
|
|
|
|
|
|
|
|
|
|
| 162 |
|
| 163 |
-
run_button = gr.Button("Run Evaluation & Submit All Answers")
|
| 164 |
-
|
| 165 |
-
status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
|
| 166 |
-
# Removed max_rows=10 from DataFrame constructor
|
| 167 |
results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
|
| 168 |
-
|
| 169 |
-
run_button.click(
|
| 170 |
-
fn=run_and_submit_all,
|
| 171 |
-
outputs=[status_output, results_table]
|
| 172 |
-
)
|
| 173 |
|
| 174 |
if __name__ == "__main__":
|
| 175 |
-
|
| 176 |
-
# Check for SPACE_HOST and SPACE_ID at startup for information
|
| 177 |
-
space_host_startup = os.getenv("SPACE_HOST")
|
| 178 |
-
space_id_startup = os.getenv("SPACE_ID") # Get SPACE_ID at startup
|
| 179 |
-
|
| 180 |
-
if space_host_startup:
|
| 181 |
-
print(f"✅ SPACE_HOST found: {space_host_startup}")
|
| 182 |
-
print(f" Runtime URL should be: https://{space_host_startup}.hf.space")
|
| 183 |
-
else:
|
| 184 |
-
print("ℹ️ SPACE_HOST environment variable not found (running locally?).")
|
| 185 |
-
|
| 186 |
-
if space_id_startup: # Print repo URLs if SPACE_ID is found
|
| 187 |
-
print(f"✅ SPACE_ID found: {space_id_startup}")
|
| 188 |
-
print(f" Repo URL: https://huggingface.co/spaces/{space_id_startup}")
|
| 189 |
-
print(f" Repo Tree URL: https://huggingface.co/spaces/{space_id_startup}/tree/main")
|
| 190 |
-
else:
|
| 191 |
-
print("ℹ️ SPACE_ID environment variable not found (running locally?). Repo URL cannot be determined.")
|
| 192 |
-
|
| 193 |
-
print("-"*(60 + len(" App Starting ")) + "\n")
|
| 194 |
-
|
| 195 |
-
print("Launching Gradio Interface for Basic Agent Evaluation...")
|
| 196 |
-
demo.launch(debug=True, share=False)
|
|
|
|
| 1 |
+
"""Hugging Face Agents Course Unit 4 - GAIA evaluation Space.
|
| 2 |
+
|
| 3 |
+
Based on the public Final_Assignment_Template API contract. It is a starter
|
| 4 |
+
agent, not a guaranteed passing solution. Keep your Space public for scoring.
|
| 5 |
+
"""
|
| 6 |
+
|
| 7 |
+
import io
|
| 8 |
import os
|
| 9 |
+
import re
|
| 10 |
+
from pathlib import Path
|
| 11 |
+
|
| 12 |
import gradio as gr
|
|
|
|
|
|
|
| 13 |
import pandas as pd
|
| 14 |
+
import requests
|
| 15 |
+
from smolagents import (
|
| 16 |
+
CodeAgent,
|
| 17 |
+
DuckDuckGoSearchTool,
|
| 18 |
+
InferenceClientModel,
|
| 19 |
+
VisitWebpageTool,
|
| 20 |
+
)
|
| 21 |
+
|
| 22 |
+
API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 23 |
+
MODEL_ID = os.getenv("MODEL_ID", "Qwen/Qwen3-Next-80B-A3B-Instruct")
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
def read_attachment(task_id: str, file_name: str) -> str:
|
| 27 |
+
"""Download an official task attachment and extract readable text."""
|
| 28 |
+
if not file_name:
|
| 29 |
+
return ""
|
| 30 |
+
try:
|
| 31 |
+
response = requests.get(f"{API_URL}/files/{task_id}", timeout=30)
|
| 32 |
+
response.raise_for_status()
|
| 33 |
+
if len(response.content) > 15_000_000:
|
| 34 |
+
return "Attachment is too large to read with this starter."
|
| 35 |
+
suffix = Path(file_name).suffix.lower()
|
| 36 |
+
data = io.BytesIO(response.content)
|
| 37 |
+
|
| 38 |
+
if suffix in {".txt", ".md", ".py", ".json", ".html", ".xml"}:
|
| 39 |
+
contents = response.content.decode("utf-8", errors="replace")
|
| 40 |
+
elif suffix == ".pdf":
|
| 41 |
+
from pypdf import PdfReader
|
| 42 |
+
pdf = PdfReader(data)
|
| 43 |
+
contents = "\n".join(page.extract_text() or "" for page in pdf.pages[:40])
|
| 44 |
+
elif suffix == ".csv":
|
| 45 |
+
contents = pd.read_csv(data).head(100).to_string(index=False)
|
| 46 |
+
elif suffix in {".xlsx", ".xls"}:
|
| 47 |
+
sheets = pd.read_excel(data, sheet_name=None)
|
| 48 |
+
contents = "\n".join(
|
| 49 |
+
f"SHEET {name}:\n{frame.head(60).to_string(index=False)}"
|
| 50 |
+
for name, frame in sheets.items()
|
| 51 |
+
)
|
| 52 |
+
elif suffix == ".docx":
|
| 53 |
+
from docx import Document
|
| 54 |
+
doc = Document(data)
|
| 55 |
+
contents = "\n".join(p.text for p in doc.paragraphs)
|
| 56 |
+
else:
|
| 57 |
+
return (
|
| 58 |
+
f"Attachment '{file_name}' is present, but this starter cannot "
|
| 59 |
+
"interpret its media format. Do not guess its contents."
|
| 60 |
+
)
|
| 61 |
+
return f"Attachment ({file_name}):\n{contents[:14000]}"
|
| 62 |
+
except Exception as exc:
|
| 63 |
+
return f"Could not read attachment {file_name}: {type(exc).__name__}: {exc}"
|
| 64 |
|
|
|
|
|
|
|
|
|
|
| 65 |
|
|
|
|
|
|
|
| 66 |
class BasicAgent:
|
| 67 |
def __init__(self):
|
| 68 |
+
token = os.getenv("HF_TOKEN")
|
| 69 |
+
if not token:
|
| 70 |
+
raise RuntimeError(
|
| 71 |
+
"HF_TOKEN Secret is missing. Add it in Space Settings "
|
| 72 |
+
"-> Variables and secrets -> New secret."
|
| 73 |
+
)
|
| 74 |
+
model = InferenceClientModel(
|
| 75 |
+
model_id=MODEL_ID,
|
| 76 |
+
token=token,
|
| 77 |
+
timeout=120,
|
| 78 |
+
max_tokens=1800,
|
| 79 |
+
)
|
| 80 |
+
self.agent = CodeAgent(
|
| 81 |
+
tools=[DuckDuckGoSearchTool(max_results=5), VisitWebpageTool()],
|
| 82 |
+
model=model,
|
| 83 |
+
max_steps=6,
|
| 84 |
+
additional_authorized_imports=[
|
| 85 |
+
"math", "statistics", "datetime", "re", "json", "itertools"
|
| 86 |
+
],
|
| 87 |
+
)
|
| 88 |
+
|
| 89 |
+
def __call__(self, question: str, attachment: str = "") -> str:
|
| 90 |
+
prompt = (
|
| 91 |
+
"You are solving a GAIA benchmark question. Use web search and "
|
| 92 |
+
"calculations if needed. Check evidence and do not invent facts. "
|
| 93 |
+
"Your final_answer must contain ONLY the shortest precise answer "
|
| 94 |
+
"requested by the question. No explanation, citations, markdown, "
|
| 95 |
+
"or the words FINAL ANSWER. Preserve requested units and format.\n\n"
|
| 96 |
+
f"QUESTION:\n{question}\n\n"
|
| 97 |
+
f"{attachment}"
|
| 98 |
+
)
|
| 99 |
+
answer = str(self.agent.run(prompt, reset=True)).strip()
|
| 100 |
+
answer = re.sub(r"^\s*(?:final answer|answer)\s*:\s*", "", answer, flags=re.I)
|
| 101 |
+
answer = re.sub(r"<think>.*?</think>", "", answer, flags=re.I | re.S).strip()
|
| 102 |
+
return answer.strip("` \n")
|
| 103 |
+
|
| 104 |
+
|
| 105 |
+
def answer_task(agent: BasicAgent, item: dict) -> str:
|
| 106 |
+
file_context = read_attachment(item["task_id"], item.get("file_name") or "")
|
| 107 |
+
return agent(item["question"], file_context)
|
| 108 |
+
|
| 109 |
+
|
| 110 |
+
def test_one_question():
|
| 111 |
+
"""Try one official sample before running all 20 questions."""
|
| 112 |
try:
|
| 113 |
+
response = requests.get(f"{API_URL}/random-question", timeout=20)
|
| 114 |
+
response.raise_for_status()
|
| 115 |
+
item = response.json()
|
| 116 |
+
if not isinstance(item, dict) or not item.get("task_id"):
|
| 117 |
+
return "The sample API returned an unexpected response."
|
| 118 |
agent = BasicAgent()
|
| 119 |
+
answer = answer_task(agent, item)
|
| 120 |
+
return f"Question: {item['question']}\n\nAgent answer: {answer}"
|
| 121 |
+
except Exception as exc:
|
| 122 |
+
return f"Test failed: {type(exc).__name__}: {exc}"
|
| 123 |
+
|
| 124 |
+
|
| 125 |
+
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 126 |
+
"""Run the official GAIA question set and submit the generated answers."""
|
| 127 |
+
if not profile:
|
| 128 |
+
return "Please log in with the Hugging Face button first.", None
|
| 129 |
+
space_id = os.getenv("SPACE_ID")
|
| 130 |
+
if not space_id:
|
| 131 |
+
return "SPACE_ID is missing. Run this code inside your own Hugging Face Space.", None
|
| 132 |
+
username = profile.username
|
| 133 |
+
|
| 134 |
try:
|
| 135 |
+
agent = BasicAgent()
|
| 136 |
+
response = requests.get(f"{API_URL}/questions", timeout=30)
|
| 137 |
response.raise_for_status()
|
| 138 |
+
questions = response.json()
|
| 139 |
+
if not isinstance(questions, list) or not questions:
|
| 140 |
+
return "The question API did not return a question list.", None
|
| 141 |
+
except Exception as exc:
|
| 142 |
+
return f"Initialization failed: {type(exc).__name__}: {exc}", None
|
| 143 |
+
|
| 144 |
+
payload = []
|
| 145 |
+
rows = []
|
| 146 |
+
for index, item in enumerate(questions, 1):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 147 |
task_id = item.get("task_id")
|
| 148 |
+
question = item.get("question")
|
| 149 |
+
if not task_id or not question:
|
|
|
|
| 150 |
continue
|
| 151 |
try:
|
| 152 |
+
answer = answer_task(agent, item)
|
| 153 |
+
if not answer:
|
| 154 |
+
raise ValueError("The agent returned an empty answer")
|
| 155 |
+
payload.append({"task_id": task_id, "submitted_answer": answer})
|
| 156 |
+
rows.append({"Task ID": task_id, "Question": question, "Submitted Answer": answer})
|
| 157 |
+
print(f"Task {index}/{len(questions)} completed")
|
| 158 |
+
except Exception as exc:
|
| 159 |
+
error = f"ERROR: {type(exc).__name__}: {exc}"
|
| 160 |
+
print(f"Task {index}/{len(questions)} failed: {error}")
|
| 161 |
+
rows.append({"Task ID": task_id, "Question": question, "Submitted Answer": error})
|
| 162 |
+
|
| 163 |
+
table = pd.DataFrame(rows)
|
| 164 |
+
if not payload:
|
| 165 |
+
return "No answers generated. Check the Space logs and HF_TOKEN / model quota.", table
|
| 166 |
+
|
| 167 |
+
request_data = {
|
| 168 |
+
"username": username,
|
| 169 |
+
"agent_code": f"https://huggingface.co/spaces/{space_id}/tree/main",
|
| 170 |
+
"answers": payload,
|
| 171 |
+
}
|
| 172 |
try:
|
| 173 |
+
response = requests.post(f"{API_URL}/submit", json=request_data, timeout=90)
|
| 174 |
response.raise_for_status()
|
| 175 |
+
result = response.json()
|
| 176 |
+
status = (
|
| 177 |
+
"Submission Successful!\n"
|
| 178 |
+
f"User: {result.get('username', username)}\n"
|
| 179 |
+
f"Overall Score: {result.get('score', '?')}% "
|
| 180 |
+
f"({result.get('correct_count', '?')}/{result.get('total_attempted', '?')} correct)\n"
|
| 181 |
+
f"Submitted answers: {len(payload)}/{len(questions)}\n"
|
| 182 |
+
f"Message: {result.get('message', '')}"
|
| 183 |
)
|
| 184 |
+
return status, table
|
| 185 |
+
except Exception as exc:
|
| 186 |
+
return f"Submission failed: {type(exc).__name__}: {exc}", table
|
| 187 |
+
|
| 188 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 189 |
with gr.Blocks() as demo:
|
| 190 |
+
gr.Markdown("# Hugging Face Agents Course - Unit 4")
|
| 191 |
gr.Markdown(
|
| 192 |
+
"First configure the **HF_TOKEN** Secret in your Space Settings. "
|
| 193 |
+
"Log in below for the leaderboard. Test one question before submitting all 20. "
|
| 194 |
+
"This starter handles web search and text-like attachments; image/audio "
|
| 195 |
+
"tasks need additional tools. Model calls may consume paid credits."
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 196 |
)
|
|
|
|
| 197 |
gr.LoginButton()
|
| 198 |
+
test_button = gr.Button("1. Test one question")
|
| 199 |
+
sample_output = gr.Textbox(label="Sample test result", lines=8)
|
| 200 |
+
test_button.click(fn=test_one_question, outputs=sample_output)
|
| 201 |
|
| 202 |
+
run_button = gr.Button("2. Run Evaluation & Submit All Answers", variant="primary")
|
| 203 |
+
status_output = gr.Textbox(label="Submission result", lines=6)
|
|
|
|
|
|
|
| 204 |
results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
|
| 205 |
+
run_button.click(fn=run_and_submit_all, outputs=[status_output, results_table])
|
|
|
|
|
|
|
|
|
|
|
|
|
| 206 |
|
| 207 |
if __name__ == "__main__":
|
| 208 |
+
demo.launch()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|