Update agent.py
Browse files
agent.py
CHANGED
@@ -1,76 +1,37 @@
|
|
1 |
import os
|
2 |
-
import requests
|
3 |
import base64
|
4 |
-
|
5 |
-
from
|
6 |
-
from
|
7 |
-
from langchain.agents.agent_types import AgentType
|
8 |
-
from langchain.memory import ConversationBufferMemory
|
9 |
-
from langchain_core.messages import HumanMessage
|
10 |
|
11 |
-
|
12 |
|
13 |
class BasicAgent:
|
14 |
def __init__(self):
|
|
|
15 |
print("BasicAgent initialized.")
|
16 |
-
# Use DuckDuckGo Search only
|
17 |
-
tools = [
|
18 |
-
Tool(
|
19 |
-
name="DuckDuckGo Search",
|
20 |
-
func=DuckDuckGoSearchRun().run,
|
21 |
-
description="Use this tool to find factual information or recent events."
|
22 |
-
),
|
23 |
-
Tool(
|
24 |
-
name="Image Analyzer",
|
25 |
-
func=self.describe_image,
|
26 |
-
description="Analyzes and describes what's in an image. Input is an image path."
|
27 |
-
)
|
28 |
-
]
|
29 |
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
self.agent = initialize_agent(
|
34 |
-
tools=tools,
|
35 |
-
llm=self.model,
|
36 |
-
agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION,
|
37 |
-
verbose=True,
|
38 |
-
memory=memory
|
39 |
-
)
|
40 |
-
|
41 |
-
def describe_image(self, img_path: str) -> str:
|
42 |
-
all_text = ""
|
43 |
try:
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
"type": "image_url",
|
58 |
-
"image_url": {
|
59 |
-
"url": f"data:image/png;base64,{image_base64}"
|
60 |
-
},
|
61 |
-
},
|
62 |
-
]
|
63 |
-
)
|
64 |
-
]
|
65 |
-
response = self.model.invoke(message)
|
66 |
-
all_text += response.content + "\n\n"
|
67 |
-
return all_text.strip()
|
68 |
except Exception as e:
|
69 |
-
|
70 |
-
print(error_msg)
|
71 |
-
return ""
|
72 |
|
73 |
def fetch_file(self, task_id):
|
|
|
74 |
try:
|
75 |
url = f"{DEFAULT_API_URL}/files/{task_id}"
|
76 |
r = requests.get(url, timeout=10)
|
@@ -78,13 +39,45 @@ class BasicAgent:
|
|
78 |
return url, r.content, r.headers.get("Content-Type", "")
|
79 |
except:
|
80 |
return None, None, None
|
81 |
-
|
82 |
-
def
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import os
|
|
|
2 |
import base64
|
3 |
+
import requests
|
4 |
+
from openai import OpenAI
|
5 |
+
from duckduckgo_search import DDGS
|
|
|
|
|
|
|
6 |
|
7 |
+
# You can expand with image/audio/etc tools as needed
|
8 |
|
9 |
class BasicAgent:
|
10 |
def __init__(self):
|
11 |
+
self.llm = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
|
12 |
print("BasicAgent initialized.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
|
14 |
+
def web_search(self, query: str, max_results: int = 5) -> str:
|
15 |
+
"""Search the web using DuckDuckGo for current information."""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
16 |
try:
|
17 |
+
with DDGS() as ddgs:
|
18 |
+
results = list(ddgs.text(query, max_results=max_results))
|
19 |
+
if not results:
|
20 |
+
return f"No results found for query: {query}"
|
21 |
+
formatted_results = f"Web search results for '{query}':\n\n"
|
22 |
+
for i, result in enumerate(results, 1):
|
23 |
+
title = result.get('title', 'No title')
|
24 |
+
body = result.get('body', 'No description')
|
25 |
+
href = result.get('href', 'No URL')
|
26 |
+
formatted_results += f"{i}. {title}\n"
|
27 |
+
formatted_results += f" URL: {href}\n"
|
28 |
+
formatted_results += f" Description: {body}\n\n"
|
29 |
+
return formatted_results
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
30 |
except Exception as e:
|
31 |
+
return f"Error performing web search: {str(e)}"
|
|
|
|
|
32 |
|
33 |
def fetch_file(self, task_id):
|
34 |
+
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
35 |
try:
|
36 |
url = f"{DEFAULT_API_URL}/files/{task_id}"
|
37 |
r = requests.get(url, timeout=10)
|
|
|
39 |
return url, r.content, r.headers.get("Content-Type", "")
|
40 |
except:
|
41 |
return None, None, None
|
42 |
+
|
43 |
+
def describe_image(self, img_path: str) -> str:
|
44 |
+
# This is optional: image-based chess move recognition
|
45 |
+
try:
|
46 |
+
r = requests.get(img_path, timeout=10)
|
47 |
+
image_bytes = r.content
|
48 |
+
image_base64 = base64.b64encode(image_bytes).decode("utf-8")
|
49 |
+
prompt = (
|
50 |
+
"You're a chess assistant. Answer only with the best move in algebraic notation (e.g., Qd1#)."
|
51 |
+
)
|
52 |
+
# In this template, you'd need your LLM to accept vision input.
|
53 |
+
# If not, just return an error or a stub.
|
54 |
+
return "[Image analysis not implemented in this agent.]"
|
55 |
+
except Exception as e:
|
56 |
+
return f"Error extracting text: {str(e)}"
|
57 |
+
|
58 |
+
def __call__(self, question: str, task_id: str = None) -> str:
|
59 |
+
# Step 1: Always web search and build prompt
|
60 |
+
search_snippet = self.web_search(question)
|
61 |
+
# Step 2: Compose the mandated prompt exactly as specified
|
62 |
+
full_prompt = (
|
63 |
+
"You are a general AI assistant. I will ask you a question. "
|
64 |
+
"Report your thoughts, and finish your answer with the following template: "
|
65 |
+
"FINAL ANSWER: [YOUR FINAL ANSWER]. YOUR FINAL ANSWER should be a number OR as few words as possible OR a comma separated list of numbers and/or strings. "
|
66 |
+
"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. "
|
67 |
+
"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. "
|
68 |
+
"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.\n\n"
|
69 |
+
f"Here are web search results and the question:\n{search_snippet}\n\nQuestion: {question}"
|
70 |
+
)
|
71 |
+
|
72 |
+
# Step 3: Get answer from OpenAI, temperature 0 for max determinism
|
73 |
+
response = self.llm.chat.completions.create(
|
74 |
+
model="gpt-4o",
|
75 |
+
messages=[
|
76 |
+
{"role": "system", "content": full_prompt},
|
77 |
+
],
|
78 |
+
temperature=0.0,
|
79 |
+
max_tokens=512,
|
80 |
+
)
|
81 |
+
answer = response.choices[0].message.content.strip()
|
82 |
+
# You may want to extract just the "FINAL ANSWER: ..." line, or you can return as is
|
83 |
+
return answer
|