Spaces:
Sleeping
Sleeping
web
Browse files
agent.py
CHANGED
@@ -11,17 +11,13 @@ from tools import (
|
|
11 |
SYSTEM_PROMPT = """
|
12 |
You are a helpful AI assistant. You will answer questions by thinking step-by-step, taking actions using tools when necessary, and finishing with a final answer.
|
13 |
|
14 |
-
When you want to use a tool, respond *exactly* in the following format:
|
15 |
|
16 |
-
|
17 |
-
|
18 |
Action Input: [input]
|
19 |
|
20 |
When you receive an observation, continue reasoning. Write:
|
21 |
|
22 |
-
Thought: [final reasoning]
|
23 |
-
FINAL ANSWER: [your answer]
|
24 |
-
|
25 |
IMPORTANT:
|
26 |
- YOUR FINAL ANSWER should be a number OR as few words as possible OR a comma separated list of numbers and/or strings. If you are asked for a number, don't use comma to write your number neither use units such as $ or percent sign unless specified otherwise. If you are asked for a string, don't use articles, neither abbreviations (e.g. for cities), and write the digits in plain text unless specified otherwise. If you are asked for a comma separated list, apply the above rules depending of whether the element to be put in the list is a number or a string.
|
27 |
- If using tools that require a `task_id`, only use the value provided.
|
@@ -31,6 +27,10 @@ IMPORTANT:
|
|
31 |
- Try to use the web search tool first before using wiki and arxiv tools.
|
32 |
- If wiki tool doesnt give you enough information, try again with a shorter query, or a more genral query. Dont use a query that is too specific.
|
33 |
- DOnt user wiki tool, web search tool and arxiv tool more than 3 times.
|
|
|
|
|
|
|
|
|
34 |
"""
|
35 |
|
36 |
TOOLS = [
|
|
|
11 |
SYSTEM_PROMPT = """
|
12 |
You are a helpful AI assistant. You will answer questions by thinking step-by-step, taking actions using tools when necessary, and finishing with a final answer.
|
13 |
|
|
|
14 |
|
15 |
+
|
16 |
+
|
17 |
Action Input: [input]
|
18 |
|
19 |
When you receive an observation, continue reasoning. Write:
|
20 |
|
|
|
|
|
|
|
21 |
IMPORTANT:
|
22 |
- YOUR FINAL ANSWER should be a number OR as few words as possible OR a comma separated list of numbers and/or strings. If you are asked for a number, don't use comma to write your number neither use units such as $ or percent sign unless specified otherwise. If you are asked for a string, don't use articles, neither abbreviations (e.g. for cities), and write the digits in plain text unless specified otherwise. If you are asked for a comma separated list, apply the above rules depending of whether the element to be put in the list is a number or a string.
|
23 |
- If using tools that require a `task_id`, only use the value provided.
|
|
|
27 |
- Try to use the web search tool first before using wiki and arxiv tools.
|
28 |
- If wiki tool doesnt give you enough information, try again with a shorter query, or a more genral query. Dont use a query that is too specific.
|
29 |
- DOnt user wiki tool, web search tool and arxiv tool more than 3 times.
|
30 |
+
|
31 |
+
Return the final answer in the following format:
|
32 |
+
Thought: [final reasoning]
|
33 |
+
FINAL ANSWER: [your answer]
|
34 |
"""
|
35 |
|
36 |
TOOLS = [
|
tools.py
CHANGED
@@ -219,12 +219,21 @@ def wikipedia_search_tool(wiki_query: str) -> str:
|
|
219 |
- "What is the capital of France?"
|
220 |
- "Find information about quantum computing"
|
221 |
- "What is the history of the internet?"
|
222 |
-
If no valid wiki_query is provided, returns
|
223 |
"""
|
224 |
print("reached wikipedia search tool")
|
225 |
-
|
|
|
|
|
|
|
|
|
|
|
226 |
if not query:
|
227 |
-
return
|
|
|
|
|
|
|
|
|
228 |
|
229 |
try:
|
230 |
# 1) Use the MediaWiki API to search for page titles matching the query
|
@@ -240,16 +249,17 @@ def wikipedia_search_tool(wiki_query: str) -> str:
|
|
240 |
search_data = search_resp.json()
|
241 |
|
242 |
search_results = search_data.get("query", {}).get("search", [])
|
243 |
-
# print("wikipedia: search_results",search_results)
|
244 |
if not search_results:
|
245 |
-
|
246 |
-
|
|
|
247 |
|
248 |
# 2) Take the first search result's title
|
249 |
first_title = search_results[0].get("title", "")
|
250 |
if not first_title:
|
251 |
-
|
252 |
-
|
|
|
253 |
|
254 |
# 3) Fetch the page summary for that title via the REST summary endpoint
|
255 |
title_for_url = requests.utils.requote_uri(first_title)
|
@@ -262,16 +272,18 @@ def wikipedia_search_tool(wiki_query: str) -> str:
|
|
262 |
summary_text = summary_data.get("extract")
|
263 |
if not summary_text:
|
264 |
summary_text = summary_data.get("description", "No summary available.")
|
|
|
|
|
|
|
265 |
print("Submitted wiki successfully")
|
266 |
-
return
|
267 |
-
|
268 |
|
269 |
except requests.exceptions.RequestException as e:
|
270 |
print("Wikipedia search error: ", e)
|
271 |
-
return f"Wikipedia search error: {e}"
|
272 |
except Exception as e:
|
273 |
print("Unexpected error in wikipedia_search_tool: ", e)
|
274 |
-
return f"Unexpected error in wikipedia_search_tool: {e}"
|
275 |
|
276 |
@tool
|
277 |
def arxiv_search_tool(query: str) -> str:
|
@@ -452,16 +464,20 @@ def web_search_tool(query: str) -> str:
|
|
452 |
Purpose: When the user asks for current information, recent news, or topics not covered by Wikipedia, use this tool.
|
453 |
|
454 |
Input: A string describing what to search for on the web.
|
455 |
-
|
456 |
-
Example usage:
|
457 |
-
- "What's the latest news about AI?"
|
458 |
-
- "Current stock price of Tesla"
|
459 |
-
- "Recent developments in renewable energy"
|
460 |
"""
|
461 |
print("reached web_search_tool")
|
|
|
|
|
|
|
|
|
|
|
462 |
if not query:
|
463 |
return "No search query provided."
|
464 |
|
|
|
|
|
|
|
|
|
465 |
ddg = DDGS()
|
466 |
max_retries = 5
|
467 |
result_text = ""
|
@@ -470,26 +486,23 @@ def web_search_tool(query: str) -> str:
|
|
470 |
try:
|
471 |
result_text = str(ddg.text(query, max_results=5))
|
472 |
except Exception as e:
|
473 |
-
# Network error or timeout—retry up to max_retries
|
474 |
if attempt < max_retries:
|
475 |
print(f"web_search_tool: exception '{e}', retrying in 4 seconds ({attempt}/{max_retries})")
|
476 |
time.sleep(4)
|
477 |
continue
|
478 |
else:
|
479 |
-
|
480 |
-
return f"Error during DuckDuckGo search: {e}"
|
481 |
|
482 |
-
# Check for DuckDuckGo rate‐limit indicator
|
483 |
if "202 Ratelimit" in result_text:
|
484 |
if attempt < max_retries:
|
485 |
print(f"web_search_tool: received '202 Ratelimit', retrying in 4 seconds ({attempt}/{max_retries})")
|
486 |
time.sleep(4)
|
487 |
continue
|
488 |
else:
|
489 |
-
# Final attempt still rate‐limited
|
490 |
break
|
|
|
491 |
|
492 |
-
|
493 |
-
|
494 |
print("Submitted web search successfully")
|
495 |
return result_text
|
|
|
219 |
- "What is the capital of France?"
|
220 |
- "Find information about quantum computing"
|
221 |
- "What is the history of the internet?"
|
222 |
+
If no valid wiki_query is provided, returns an empty string.
|
223 |
"""
|
224 |
print("reached wikipedia search tool")
|
225 |
+
|
226 |
+
# --- Simple in-memory cache to avoid repeated look-ups in a single session
|
227 |
+
if not hasattr(wikipedia_search_tool, "_cache"):
|
228 |
+
wikipedia_search_tool._cache = {}
|
229 |
+
|
230 |
+
query = wiki_query.strip()
|
231 |
if not query:
|
232 |
+
return ""
|
233 |
+
|
234 |
+
if query in wikipedia_search_tool._cache:
|
235 |
+
print("Returning cached Wikipedia result for query:", query)
|
236 |
+
return wikipedia_search_tool._cache[query]
|
237 |
|
238 |
try:
|
239 |
# 1) Use the MediaWiki API to search for page titles matching the query
|
|
|
249 |
search_data = search_resp.json()
|
250 |
|
251 |
search_results = search_data.get("query", {}).get("search", [])
|
|
|
252 |
if not search_results:
|
253 |
+
msg = f"No Wikipedia page found for '{query}'. [END_OF_SEARCH]"
|
254 |
+
wikipedia_search_tool._cache[query] = msg
|
255 |
+
return msg
|
256 |
|
257 |
# 2) Take the first search result's title
|
258 |
first_title = search_results[0].get("title", "")
|
259 |
if not first_title:
|
260 |
+
msg = "Unexpected format from Wikipedia search. [END_OF_SEARCH]"
|
261 |
+
wikipedia_search_tool._cache[query] = msg
|
262 |
+
return msg
|
263 |
|
264 |
# 3) Fetch the page summary for that title via the REST summary endpoint
|
265 |
title_for_url = requests.utils.requote_uri(first_title)
|
|
|
272 |
summary_text = summary_data.get("extract")
|
273 |
if not summary_text:
|
274 |
summary_text = summary_data.get("description", "No summary available.")
|
275 |
+
|
276 |
+
result = f"Title: {first_title}\n\n{summary_text}\n\n[END_OF_SEARCH]"
|
277 |
+
wikipedia_search_tool._cache[query] = result
|
278 |
print("Submitted wiki successfully")
|
279 |
+
return result
|
|
|
280 |
|
281 |
except requests.exceptions.RequestException as e:
|
282 |
print("Wikipedia search error: ", e)
|
283 |
+
return f"Wikipedia search error: {e} [END_OF_SEARCH]"
|
284 |
except Exception as e:
|
285 |
print("Unexpected error in wikipedia_search_tool: ", e)
|
286 |
+
return f"Unexpected error in wikipedia_search_tool: {e} [END_OF_SEARCH]"
|
287 |
|
288 |
@tool
|
289 |
def arxiv_search_tool(query: str) -> str:
|
|
|
464 |
Purpose: When the user asks for current information, recent news, or topics not covered by Wikipedia, use this tool.
|
465 |
|
466 |
Input: A string describing what to search for on the web.
|
|
|
|
|
|
|
|
|
|
|
467 |
"""
|
468 |
print("reached web_search_tool")
|
469 |
+
|
470 |
+
if not hasattr(web_search_tool, "_cache"):
|
471 |
+
web_search_tool._cache = {}
|
472 |
+
|
473 |
+
query = query.strip()
|
474 |
if not query:
|
475 |
return "No search query provided."
|
476 |
|
477 |
+
if query in web_search_tool._cache:
|
478 |
+
print("Returning cached web search result for query:", query)
|
479 |
+
return web_search_tool._cache[query]
|
480 |
+
|
481 |
ddg = DDGS()
|
482 |
max_retries = 5
|
483 |
result_text = ""
|
|
|
486 |
try:
|
487 |
result_text = str(ddg.text(query, max_results=5))
|
488 |
except Exception as e:
|
|
|
489 |
if attempt < max_retries:
|
490 |
print(f"web_search_tool: exception '{e}', retrying in 4 seconds ({attempt}/{max_retries})")
|
491 |
time.sleep(4)
|
492 |
continue
|
493 |
else:
|
494 |
+
return f"Error during DuckDuckGo search: {e} [END_OF_SEARCH]"
|
|
|
495 |
|
|
|
496 |
if "202 Ratelimit" in result_text:
|
497 |
if attempt < max_retries:
|
498 |
print(f"web_search_tool: received '202 Ratelimit', retrying in 4 seconds ({attempt}/{max_retries})")
|
499 |
time.sleep(4)
|
500 |
continue
|
501 |
else:
|
|
|
502 |
break
|
503 |
+
break # Successful
|
504 |
|
505 |
+
result_text += "\n\n[END_OF_SEARCH]"
|
506 |
+
web_search_tool._cache[query] = result_text
|
507 |
print("Submitted web search successfully")
|
508 |
return result_text
|