forked from aiplanethub/openagi
-
Notifications
You must be signed in to change notification settings - Fork 0
/
benchmark.py
152 lines (132 loc) · 4.66 KB
/
benchmark.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
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
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
from openagi.llms.azure import AzureChatOpenAIModel
from openagi.agent import Admin
from openagi.memory import Memory
from openagi.worker import Worker
from openagi.planner.task_decomposer import TaskPlanner
from openagi.actions.base import BaseAction
import wikipedia
import os
import joblib
import requests
import re
import string
from tqdm import tqdm
from collections import Counter
from pydantic import Field, validator
import numpy as np
class WikiSearchAction(BaseAction):
"""
Use this Action to get the information from Wikipedia Search
"""
query: str = Field(
default_factory=str,
description="The search string. Be simple."
)
@validator('query')
def validate_query(cls, v):
if not v or not isinstance(v, str):
raise ValueError('Query must be a non-empty string.')
return v
def execute(self):
try:
search_res = wikipedia.search(self.query)
if not search_res:
return 'No results found.'
article = wikipedia.page(search_res[0])
return article.summary
except wikipedia.exceptions.DisambiguationError as e:
return f"Disambiguation error: {str(e)}"
except wikipedia.exceptions.PageError as e:
return f"Page error: {str(e)}"
except Exception as e:
return f"Error: {str(e)}"
def download_file(url, filename):
"""
Download a file from a URL and save it locally.
"""
try:
response = requests.get(url, timeout=10)
response.raise_for_status()
with open(filename, "wb") as f:
f.write(response.content)
print(f"Downloaded {filename}")
except requests.exceptions.RequestException as e:
print(f"Failed to download {filename}: {str(e)}")
def load_hotpot_qa_data(level):
"""
Load HotpotQA data for a given level. If data doesn't exist, download it.
"""
file_path = f"./data/{level}.joblib"
data_url = f"https://github.com/salesforce/BOLAA/raw/main/hotpotqa_run/data/{level}.joblib"
if not os.path.exists(file_path):
print(f"{level} data not found, downloading...")
os.makedirs(os.path.dirname(file_path), exist_ok=True)
download_file(data_url, file_path)
return joblib.load(file_path)
def normalize_answer(s):
"""
Normalize answers for evaluation.
"""
def remove_articles(text):
return re.sub(r"\b(a|an|the)\b", " ", text)
def white_space_fix(text):
return " ".join(text.split())
def remove_punc(text):
return "".join(ch for ch in text if ch not in string.punctuation)
def lower(text):
return text.lower()
return white_space_fix(remove_articles(remove_punc(lower(s))))
def f1_score(prediction, ground_truth):
"""
Compute the F1 score between prediction and ground truth answers.
"""
prediction_tokens = normalize_answer(prediction).split()
ground_truth_tokens = normalize_answer(ground_truth).split()
common = Counter(prediction_tokens) & Counter(ground_truth_tokens)
num_same = sum(common.values())
if num_same == 0:
return 0, 0, 0
precision = num_same / len(prediction_tokens)
recall = num_same / len(ground_truth_tokens)
f1 = (2 * precision * recall) / (precision + recall)
return f1, precision, recall
def agent(query, llm):
planner = TaskPlanner(autonomous=True)
admin = Admin(
planner=planner,
memory=Memory(),
actions=[WikiSearchAction],
llm=llm,
)
res = admin.run(
query=query,
description="Provide answer for the query. You should decompose your task into executable actions.",
)
return res
def run_agent(level = 'easy'):
os.environ["AZURE_BASE_URL"] = ""
os.environ["AZURE_DEPLOYMENT_NAME"] = ""
os.environ["AZURE_MODEL_NAME"]="gpt4"
os.environ["AZURE_OPENAI_API_VERSION"]=""
os.environ["AZURE_OPENAI_API_KEY"]= ""
config = AzureChatOpenAIModel.load_from_env_config()
llm = AzureChatOpenAIModel(config=config)
hotpot_data = load_hotpot_qa_data(level)
hotpot_data = hotpot_data.reset_index(drop=True)
task_instructions = [
(row["question"], row["answer"]) for _, row in hotpot_data.iterrows()
]
f1_list = []
correct = 0
results = {}
for task , answer in tqdm(task_instructions[0:30]):
response = agent(task , llm)
f1 , _ ,_ = f1_score(response,answer)
f1_list.append(f1)
correct += int(response == answer)
avg_f1 = np.mean(f1_list)
acc = correct / len(task_instructions[0:30])
return avg_f1, acc
# levels are 'easy', 'medium', 'hard'
f1, acc = run_agent(level='easy')
print(f"F1 score: {f1}, Accuracy: {acc}")