Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,46 +1,17 @@
|
|
| 1 |
-
import os
|
| 2 |
-
import re
|
| 3 |
-
import gradio as gr
|
| 4 |
-
import requests
|
| 5 |
-
import pandas as pd
|
| 6 |
-
|
| 7 |
-
from smolagents import CodeAgent, DuckDuckGoSearchTool
|
| 8 |
-
from smolagents.models import OpenAIServerModel
|
| 9 |
-
|
| 10 |
-
# ---------------------------------------------------
|
| 11 |
-
# Constants
|
| 12 |
-
# ---------------------------------------------------
|
| 13 |
-
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 14 |
-
|
| 15 |
-
# ---------------------------------------------------
|
| 16 |
-
# Agent
|
| 17 |
-
# ---------------------------------------------------
|
| 18 |
class BasicAgent:
|
| 19 |
|
| 20 |
def __init__(self):
|
| 21 |
|
| 22 |
-
print("Initializing Groq Agent...")
|
| 23 |
|
| 24 |
self.model = OpenAIServerModel(
|
| 25 |
-
model_id="llama-3.
|
| 26 |
api_base="https://api.groq.com/openai/v1",
|
| 27 |
api_key=os.getenv("GROQ_API_KEY")
|
| 28 |
)
|
| 29 |
|
| 30 |
-
|
| 31 |
-
tools=[
|
| 32 |
-
DuckDuckGoSearchTool()
|
| 33 |
-
],
|
| 34 |
-
model=self.model,
|
| 35 |
-
max_steps=3,
|
| 36 |
-
verbosity_level=0
|
| 37 |
-
)
|
| 38 |
-
|
| 39 |
-
print("Groq Agent initialized.")
|
| 40 |
|
| 41 |
-
# ---------------------------------------------------
|
| 42 |
-
# Clean outputs for exact-match grading
|
| 43 |
-
# ---------------------------------------------------
|
| 44 |
def clean_answer(self, text):
|
| 45 |
|
| 46 |
if text is None:
|
|
@@ -48,38 +19,24 @@ class BasicAgent:
|
|
| 48 |
|
| 49 |
text = str(text)
|
| 50 |
|
| 51 |
-
# remove markdown/code
|
| 52 |
text = text.replace("```", "")
|
| 53 |
text = text.replace("FINAL ANSWER:", "")
|
| 54 |
text = text.replace("Answer:", "")
|
| 55 |
|
| 56 |
-
|
| 57 |
-
text = re.sub(r"\s+", " ", text)
|
| 58 |
|
| 59 |
-
|
| 60 |
-
text = text.split("\n")[0]
|
| 61 |
|
| 62 |
-
# concise
|
| 63 |
-
text = text.strip()
|
| 64 |
-
|
| 65 |
-
return text[:300]
|
| 66 |
-
|
| 67 |
-
# ---------------------------------------------------
|
| 68 |
-
# Run agent
|
| 69 |
-
# ---------------------------------------------------
|
| 70 |
def __call__(self, question: str) -> str:
|
| 71 |
|
| 72 |
prompt = f"""
|
| 73 |
-
|
| 74 |
|
| 75 |
IMPORTANT:
|
| 76 |
- Return ONLY the final answer.
|
| 77 |
- No reasoning.
|
| 78 |
-
- No
|
| 79 |
-
-
|
| 80 |
-
- No bullet points.
|
| 81 |
-
- Be concise and accurate.
|
| 82 |
-
- Use web search if needed.
|
| 83 |
|
| 84 |
Question:
|
| 85 |
{question}
|
|
@@ -87,220 +44,19 @@ Question:
|
|
| 87 |
|
| 88 |
try:
|
| 89 |
|
| 90 |
-
|
|
|
|
|
|
|
|
|
|
| 91 |
|
| 92 |
-
cleaned = self.clean_answer(
|
| 93 |
|
| 94 |
-
print(
|
| 95 |
-
print(f"\nANSWER:\n{cleaned}")
|
| 96 |
|
| 97 |
return cleaned
|
| 98 |
|
| 99 |
except Exception as e:
|
| 100 |
|
| 101 |
-
print(
|
| 102 |
-
|
| 103 |
-
return ""
|
| 104 |
-
|
| 105 |
-
|
| 106 |
-
# ---------------------------------------------------
|
| 107 |
-
# Evaluation + Submission
|
| 108 |
-
# ---------------------------------------------------
|
| 109 |
-
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 110 |
-
|
| 111 |
-
space_id = os.getenv("SPACE_ID")
|
| 112 |
-
|
| 113 |
-
# Login check
|
| 114 |
-
if profile:
|
| 115 |
-
username = profile.username
|
| 116 |
-
print(f"Logged in as: {username}")
|
| 117 |
-
else:
|
| 118 |
-
return "Please login with Hugging Face.", None
|
| 119 |
-
|
| 120 |
-
api_url = DEFAULT_API_URL
|
| 121 |
-
questions_url = f"{api_url}/questions"
|
| 122 |
-
submit_url = f"{api_url}/submit"
|
| 123 |
-
|
| 124 |
-
# ---------------------------------------------------
|
| 125 |
-
# Initialize Agent
|
| 126 |
-
# ---------------------------------------------------
|
| 127 |
-
try:
|
| 128 |
-
|
| 129 |
-
agent = BasicAgent()
|
| 130 |
-
|
| 131 |
-
except Exception as e:
|
| 132 |
-
|
| 133 |
-
print(f"Initialization error: {e}")
|
| 134 |
-
|
| 135 |
-
return f"Error initializing agent: {e}", None
|
| 136 |
-
|
| 137 |
-
# Space repo link
|
| 138 |
-
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
| 139 |
-
|
| 140 |
-
# ---------------------------------------------------
|
| 141 |
-
# Fetch Questions
|
| 142 |
-
# ---------------------------------------------------
|
| 143 |
-
try:
|
| 144 |
-
|
| 145 |
-
print("Fetching questions...")
|
| 146 |
-
|
| 147 |
-
response = requests.get(
|
| 148 |
-
questions_url,
|
| 149 |
-
timeout=30
|
| 150 |
-
)
|
| 151 |
-
|
| 152 |
-
response.raise_for_status()
|
| 153 |
-
|
| 154 |
-
questions_data = response.json()
|
| 155 |
-
|
| 156 |
-
print(f"Fetched {len(questions_data)} questions.")
|
| 157 |
-
|
| 158 |
-
except Exception as e:
|
| 159 |
-
|
| 160 |
-
print(f"Question fetch error: {e}")
|
| 161 |
-
|
| 162 |
-
return f"Error fetching questions: {e}", None
|
| 163 |
-
|
| 164 |
-
# ---------------------------------------------------
|
| 165 |
-
# Run Agent
|
| 166 |
-
# ---------------------------------------------------
|
| 167 |
-
answers_payload = []
|
| 168 |
-
results_log = []
|
| 169 |
-
|
| 170 |
-
for item in questions_data:
|
| 171 |
-
|
| 172 |
-
task_id = item.get("task_id")
|
| 173 |
-
question_text = item.get("question")
|
| 174 |
-
|
| 175 |
-
if not task_id or question_text is None:
|
| 176 |
-
continue
|
| 177 |
-
|
| 178 |
-
print(f"\nRunning task {task_id}")
|
| 179 |
-
|
| 180 |
-
try:
|
| 181 |
-
|
| 182 |
-
submitted_answer = agent(question_text)
|
| 183 |
-
|
| 184 |
-
answers_payload.append({
|
| 185 |
-
"task_id": task_id,
|
| 186 |
-
"submitted_answer": submitted_answer
|
| 187 |
-
})
|
| 188 |
-
|
| 189 |
-
results_log.append({
|
| 190 |
-
"Task ID": task_id,
|
| 191 |
-
"Question": question_text,
|
| 192 |
-
"Submitted Answer": submitted_answer
|
| 193 |
-
})
|
| 194 |
-
|
| 195 |
-
except Exception as e:
|
| 196 |
-
|
| 197 |
-
print(f"Task error: {e}")
|
| 198 |
-
|
| 199 |
-
results_log.append({
|
| 200 |
-
"Task ID": task_id,
|
| 201 |
-
"Question": question_text,
|
| 202 |
-
"Submitted Answer": f"ERROR: {e}"
|
| 203 |
-
})
|
| 204 |
-
|
| 205 |
-
# ---------------------------------------------------
|
| 206 |
-
# Submit
|
| 207 |
-
# ---------------------------------------------------
|
| 208 |
-
submission_data = {
|
| 209 |
-
"username": username.strip(),
|
| 210 |
-
"agent_code": agent_code,
|
| 211 |
-
"answers": answers_payload
|
| 212 |
-
}
|
| 213 |
-
|
| 214 |
-
try:
|
| 215 |
-
|
| 216 |
-
print("Submitting answers...")
|
| 217 |
-
|
| 218 |
-
response = requests.post(
|
| 219 |
-
submit_url,
|
| 220 |
-
json=submission_data,
|
| 221 |
-
timeout=120
|
| 222 |
-
)
|
| 223 |
-
|
| 224 |
-
response.raise_for_status()
|
| 225 |
-
|
| 226 |
-
result_data = response.json()
|
| 227 |
-
|
| 228 |
-
final_status = (
|
| 229 |
-
f"Submission Successful!\n\n"
|
| 230 |
-
f"User: {result_data.get('username')}\n"
|
| 231 |
-
f"Score: {result_data.get('score', 'N/A')}%\n"
|
| 232 |
-
f"Correct: "
|
| 233 |
-
f"{result_data.get('correct_count', '?')}/"
|
| 234 |
-
f"{result_data.get('total_attempted', '?')}\n\n"
|
| 235 |
-
f"Message: {result_data.get('message', '')}"
|
| 236 |
-
)
|
| 237 |
-
|
| 238 |
-
print(final_status)
|
| 239 |
-
|
| 240 |
-
return final_status, pd.DataFrame(results_log)
|
| 241 |
-
|
| 242 |
-
except Exception as e:
|
| 243 |
-
|
| 244 |
-
error_message = f"Submission failed: {e}"
|
| 245 |
-
|
| 246 |
-
print(error_message)
|
| 247 |
-
|
| 248 |
-
return error_message, pd.DataFrame(results_log)
|
| 249 |
-
|
| 250 |
-
|
| 251 |
-
# ---------------------------------------------------
|
| 252 |
-
# UI
|
| 253 |
-
# ---------------------------------------------------
|
| 254 |
-
with gr.Blocks() as demo:
|
| 255 |
-
|
| 256 |
-
gr.Markdown("# Hugging Face Agents Course - Final Assignment")
|
| 257 |
-
|
| 258 |
-
gr.Markdown(
|
| 259 |
-
"""
|
| 260 |
-
This agent uses:
|
| 261 |
-
|
| 262 |
-
- Groq API
|
| 263 |
-
- Llama 3.3 70B
|
| 264 |
-
- smolagents
|
| 265 |
-
- DuckDuckGo Search
|
| 266 |
-
"""
|
| 267 |
-
)
|
| 268 |
-
|
| 269 |
-
gr.LoginButton()
|
| 270 |
-
|
| 271 |
-
run_button = gr.Button(
|
| 272 |
-
"Run Evaluation & Submit"
|
| 273 |
-
)
|
| 274 |
-
|
| 275 |
-
status_output = gr.Textbox(
|
| 276 |
-
label="Submission Result",
|
| 277 |
-
lines=8,
|
| 278 |
-
interactive=False
|
| 279 |
-
)
|
| 280 |
-
|
| 281 |
-
results_table = gr.DataFrame(
|
| 282 |
-
label="Agent Answers",
|
| 283 |
-
wrap=True
|
| 284 |
-
)
|
| 285 |
-
|
| 286 |
-
run_button.click(
|
| 287 |
-
fn=run_and_submit_all,
|
| 288 |
-
outputs=[
|
| 289 |
-
status_output,
|
| 290 |
-
results_table
|
| 291 |
-
]
|
| 292 |
-
)
|
| 293 |
-
|
| 294 |
-
# ---------------------------------------------------
|
| 295 |
-
# Launch
|
| 296 |
-
# ---------------------------------------------------
|
| 297 |
-
if __name__ == "__main__":
|
| 298 |
-
|
| 299 |
-
print("\n==============================")
|
| 300 |
-
print("Starting Groq Agent...")
|
| 301 |
-
print("==============================\n")
|
| 302 |
|
| 303 |
-
|
| 304 |
-
debug=True,
|
| 305 |
-
share=False
|
| 306 |
-
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
class BasicAgent:
|
| 2 |
|
| 3 |
def __init__(self):
|
| 4 |
|
| 5 |
+
print("Initializing Fast Groq Agent...")
|
| 6 |
|
| 7 |
self.model = OpenAIServerModel(
|
| 8 |
+
model_id="llama-3.1-8b-instant",
|
| 9 |
api_base="https://api.groq.com/openai/v1",
|
| 10 |
api_key=os.getenv("GROQ_API_KEY")
|
| 11 |
)
|
| 12 |
|
| 13 |
+
print("Fast agent ready.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
|
|
|
|
|
|
|
|
|
|
| 15 |
def clean_answer(self, text):
|
| 16 |
|
| 17 |
if text is None:
|
|
|
|
| 19 |
|
| 20 |
text = str(text)
|
| 21 |
|
|
|
|
| 22 |
text = text.replace("```", "")
|
| 23 |
text = text.replace("FINAL ANSWER:", "")
|
| 24 |
text = text.replace("Answer:", "")
|
| 25 |
|
| 26 |
+
text = text.strip().split("\n")[0]
|
|
|
|
| 27 |
|
| 28 |
+
return text[:200]
|
|
|
|
| 29 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 30 |
def __call__(self, question: str) -> str:
|
| 31 |
|
| 32 |
prompt = f"""
|
| 33 |
+
Answer the question.
|
| 34 |
|
| 35 |
IMPORTANT:
|
| 36 |
- Return ONLY the final answer.
|
| 37 |
- No reasoning.
|
| 38 |
+
- No explanation.
|
| 39 |
+
- Keep answers concise.
|
|
|
|
|
|
|
|
|
|
| 40 |
|
| 41 |
Question:
|
| 42 |
{question}
|
|
|
|
| 44 |
|
| 45 |
try:
|
| 46 |
|
| 47 |
+
response = self.model(
|
| 48 |
+
prompt,
|
| 49 |
+
max_tokens=80
|
| 50 |
+
)
|
| 51 |
|
| 52 |
+
cleaned = self.clean_answer(response)
|
| 53 |
|
| 54 |
+
print(cleaned)
|
|
|
|
| 55 |
|
| 56 |
return cleaned
|
| 57 |
|
| 58 |
except Exception as e:
|
| 59 |
|
| 60 |
+
print(e)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 61 |
|
| 62 |
+
return ""
|
|
|
|
|
|
|
|
|