kpbotla commited on
Commit
421536a
·
verified ·
1 Parent(s): a4d9f42

Upload app.py

Browse files
Files changed (1) hide show
  1. app.py +91 -112
app.py CHANGED
@@ -1,127 +1,106 @@
1
- import os
2
- import gradio as gr
3
- import requests
4
- import pandas as pd
5
- #from smolagents import Agent, tool
6
- from duckduckgo_search import DDGS
7
- from transformers import pipeline
8
-
9
- # --- Tool Definitions ---
10
- from tools.search_tool import web_search
11
- from tools.citation_tool import cite
12
- from tools.summarizer_tool import summarize
13
- from tools.math_tool import python
14
- from tools.fallback_tool import fallback
15
-
16
- class BasicAgent:
17
- def __init__(self):
18
- pass
19
-
20
- def __call__(self, question: str) -> str:
21
- question_lower = question.lower()
22
-
23
- try:
24
- if any(keyword in question_lower for keyword in ["latest", "news", "who", "what", "where", "when"]):
25
- return web_search(question)
26
- elif any(keyword in question_lower for keyword in ["summarize", "summary", "explain"]):
27
- return summarize(question)
28
- elif any(char.isdigit() for char in question) and any(op in question for op in ["+", "-", "*", "/"]):
29
- return python(question)
30
- elif "|||" in question:
31
- return cite(question)
32
- else:
33
- return fallback(question)
34
- except Exception as e:
35
- return f"Error during processing: {str(e)}"
36
-
37
-
38
-
39
-
40
-
41
-
42
- # --- Evaluation Logic ---
43
- DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
44
 
45
- def run_and_submit_all(profile: gr.OAuthProfile | None):
46
- space_id = os.getenv("SPACE_ID")
47
- if profile:
48
- username = profile.username
49
- else:
50
- return "Please Login to Hugging Face with the button.", None
51
-
52
- api_url = DEFAULT_API_URL
53
- questions_url = f"{api_url}/questions"
54
- submit_url = f"{api_url}/submit"
55
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
56
  try:
57
- agent = BasicAgent()
58
- except Exception as e:
59
- return f"Error initializing agent: {e}", None
60
-
61
- agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
62
-
 
 
 
 
 
 
 
 
 
 
 
63
  try:
64
- response = requests.get(questions_url, timeout=15)
65
- response.raise_for_status()
66
- questions_data = response.json()
67
  except Exception as e:
68
- return f"Error fetching questions: {e}", None
69
 
70
- results_log = []
71
- answers_payload = []
 
 
 
72
 
73
- for item in questions_data:
74
- task_id = item.get("task_id")
75
- question_text = item.get("question")
76
- if not task_id or question_text is None:
77
- continue
78
- try:
79
- submitted_answer = agent(question_text)
80
- answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
81
- results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
82
- except Exception as e:
83
- results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})
84
 
85
- if not answers_payload:
86
- return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
 
 
87
 
88
- submission_data = {
89
- "username": username.strip(),
90
- "agent_code": agent_code,
91
- "answers": answers_payload
92
- }
93
 
94
- try:
95
- response = requests.post(submit_url, json=submission_data, timeout=60)
96
- response.raise_for_status()
97
- result_data = response.json()
98
- final_status = (
99
- f"Submission Successful!\n"
100
- f"User: {result_data.get('username')}\n"
101
- f"Overall Score: {result_data.get('score', 'N/A')}% "
102
- f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
103
- f"Message: {result_data.get('message', 'No message received.')}"
104
- )
105
- results_df = pd.DataFrame(results_log)
106
- return final_status, results_df
107
- except Exception as e:
108
- return f"Submission Failed: {e}", pd.DataFrame(results_log)
109
 
110
- # --- Gradio UI ---
111
- with gr.Blocks() as demo:
112
- gr.Markdown("# Smart Agent Evaluation Runner")
113
- gr.Markdown("""
114
- **Instructions:**
115
- 1. Login to your HF account using the button.
116
- 2. Click 'Run Evaluation & Submit All Answers' to test your agent.
117
- """)
118
 
119
- gr.LoginButton()
120
- run_button = gr.Button("Run Evaluation & Submit All Answers")
121
- status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
122
- results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
123
 
124
- run_button.click(fn=run_and_submit_all, outputs=[status_output, results_table])
 
125
 
 
126
  if __name__ == "__main__":
127
- demo.launch()
 
 
 
 
1
+ from smolagents import tool
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2
 
3
+ # TOOL 1: Wikipedia-based Search Tool
4
+ from duckduckgo_search import DDGS
 
 
 
 
 
 
 
 
5
 
6
+ @tool
7
+ def web_search(query: str) -> str:
8
+ """
9
+ Searches for up-to-date facts, biased toward Wikipedia for accuracy.
10
+
11
+ Args:
12
+ query (str): The user's factual question.
13
+
14
+ Returns:
15
+ str: Best matching fact and URL.
16
+ """
17
+ refined = f"{query} site:en.wikipedia.org"
18
+ with DDGS() as ddgs:
19
+ results = ddgs.text(refined)
20
+ for r in results[:5]:
21
+ if "wikipedia.org" in r["href"].lower():
22
+ snippet = r.get("body") or r.get("content") or r.get("snippet", "")
23
+ if snippet:
24
+ return f"{snippet}\n\nSource: [{r['href']}]({r['href']})"
25
+ return "Could not find a direct answer from Wikipedia."
26
+
27
+ # TOOL 2: Citation Tool
28
+ @tool
29
+ def cite(input: str) -> str:
30
+ """
31
+ Formats a response and URL into a markdown citation.
32
+
33
+ Args:
34
+ input (str): A string like 'answer ||| source-url'.
35
+
36
+ Returns:
37
+ str: Answer followed by markdown citation.
38
+ """
39
  try:
40
+ answer, url = input.split("|||")
41
+ return f"{answer.strip()}\n\nSource: [{url.strip()}]({url.strip()})"
42
+ except:
43
+ return "Could not format citation."
44
+
45
+ # TOOL 3: Math Eval
46
+ @tool
47
+ def python(code: str) -> str:
48
+ """
49
+ Evaluates math expressions using Python sandboxed eval.
50
+
51
+ Args:
52
+ code (str): A math expression or calculation.
53
+
54
+ Returns:
55
+ str: The result or error.
56
+ """
57
  try:
58
+ result = str(eval(code, {"__builtins__": {}}))
59
+ return f"Answer: {result}"
 
60
  except Exception as e:
61
+ return f"Error: {str(e)}"
62
 
63
+ # TOOL 4: Fallback
64
+ @tool
65
+ def fallback(_: str) -> str:
66
+ """
67
+ Handles unclear or unanswerable queries politely.
68
 
69
+ Args:
70
+ _ (str): Unused.
 
 
 
 
 
 
 
 
 
71
 
72
+ Returns:
73
+ str: A polite fallback message.
74
+ """
75
+ return "Sorry, I couldn't confidently answer that. Could you rephrase?"
76
 
77
+ # MAIN AGENT
78
+ class BasicAgent:
79
+ def __call__(self, question: str) -> str:
80
+ q = question.lower()
 
81
 
82
+ try:
83
+ # Wikipedia-focused factual lookups
84
+ if any(x in q for x in ["who", "what", "when", "where", "how many", "how much", "did", "which", "name"]):
85
+ return web_search(question)
 
 
 
 
 
 
 
 
 
 
 
86
 
87
+ # Inline citation formatting
88
+ elif "|||" in q:
89
+ return cite(question)
90
+
91
+ # Math evaluation
92
+ elif any(op in q for op in ["+", "-", "*", "/"]):
93
+ return python(question)
 
94
 
95
+ # Default catch-all
96
+ return fallback(question)
 
 
97
 
98
+ except Exception as e:
99
+ return f"Agent error: {str(e)}"
100
 
101
+ # CLI test loop
102
  if __name__ == "__main__":
103
+ agent = BasicAgent()
104
+ while True:
105
+ q = input("Ask GAIA: ")
106
+ print(agent(q))