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
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
|
import glob
import json
import os
from typing import Dict, List, Optional, Union
import pandas as pd
from gql import Client, gql
from gql.transport.aiohttp import AIOHTTPTransport
from pydantic import BaseModel, Field
# from agbenchmark.reports.processing.report_types import Report, SuiteTest
class Metrics(BaseModel):
difficulty: str
success: bool
success_percent: float = Field(..., alias="success_%")
run_time: Optional[str] = None
fail_reason: Optional[str] = None
attempted: Optional[bool] = None
class MetricsOverall(BaseModel):
run_time: str
highest_difficulty: str
percentage: Optional[float] = None
class Test(BaseModel):
data_path: str
is_regression: bool
answer: str
description: str
metrics: Metrics
category: List[str]
task: Optional[str] = None
reached_cutoff: Optional[bool] = None
class SuiteTest(BaseModel):
data_path: str
metrics: MetricsOverall
tests: Dict[str, Test]
category: Optional[List[str]] = None
task: Optional[str] = None
reached_cutoff: Optional[bool] = None
class Report(BaseModel):
command: str
completion_time: str
benchmark_start_time: str
metrics: MetricsOverall
tests: Dict[str, Union[Test, SuiteTest]]
config: Dict[str, str | dict[str, str]]
def get_reports():
# Initialize an empty list to store the report data
report_data = []
# Get the current working directory
current_dir = os.getcwd()
# Check if the current directory ends with 'reports'
if current_dir.endswith("reports"):
reports_dir = "/"
else:
reports_dir = "reports"
# Iterate over all agent directories in the reports directory
for agent_name in os.listdir(reports_dir):
if agent_name is None:
continue
agent_dir = os.path.join(reports_dir, agent_name)
# Check if the item is a directory (an agent directory)
if os.path.isdir(agent_dir):
# Construct the path to the report.json file
# Get all directories and files, but note that this will also include any file, not just directories.
run_dirs = glob.glob(os.path.join(agent_dir, "*"))
# Get all json files starting with 'file'
# old_report_files = glob.glob(os.path.join(agent_dir, "file*.json"))
# For each run directory, add the report.json to the end
# Only include the path if it's actually a directory
report_files = [
os.path.join(run_dir, "report.json")
for run_dir in run_dirs
if os.path.isdir(run_dir)
]
# old_report_files already contains the full paths, so no need to join again
# report_files = report_files + old_report_files
for report_file in report_files:
# Check if the report.json file exists
if os.path.isfile(report_file):
# Open the report.json file
with open(report_file, "r") as f:
# Load the JSON data from the file
json_data = json.load(f)
print(f"Processing {report_file}")
report = Report.parse_obj(json_data)
for test_name, test_data in report.tests.items():
test_json = {
"agent": agent_name.lower(),
"benchmark_start_time": report.benchmark_start_time,
}
if isinstance(test_data, SuiteTest):
if (
test_data.category
): # this means it's a same task test
test_json["challenge"] = test_name
test_json["attempted"] = test_data.tests[
list(test_data.tests.keys())[0]
].metrics.attempted
test_json["categories"] = ", ".join(
test_data.category
)
test_json["task"] = test_data.task
test_json["success"] = test_data.metrics.percentage
test_json[
"difficulty"
] = test_data.metrics.highest_difficulty
test_json[
"success_%"
] = test_data.metrics.percentage
test_json["run_time"] = test_data.metrics.run_time
test_json["is_regression"] = test_data.tests[
list(test_data.tests.keys())[0]
].is_regression
else: # separate tasks in 1 suite
for (
suite_test_name,
suite_data,
) in test_data.tests.items():
test_json["challenge"] = suite_test_name
test_json[
"attempted"
] = suite_data.metrics.attempted
test_json["categories"] = ", ".join(
suite_data.category
)
test_json["task"] = suite_data.task
test_json["success"] = (
100.0 if suite_data.metrics.success else 0
)
test_json[
"difficulty"
] = suite_data.metrics.difficulty
test_json[
"success_%"
] = suite_data.metrics.success_percentage
test_json[
"run_time"
] = suite_data.metrics.run_time
test_json[
"is_regression"
] = suite_data.is_regression
else:
test_json["challenge"] = test_name
test_json["attempted"] = test_data.metrics.attempted
test_json["categories"] = ", ".join(test_data.category)
test_json["task"] = test_data.task
test_json["success"] = (
100.0 if test_data.metrics.success else 0
)
test_json["difficulty"] = test_data.metrics.difficulty
test_json[
"success_%"
] = test_data.metrics.success_percentage
test_json["run_time"] = test_data.metrics.run_time
test_json["is_regression"] = test_data.is_regression
report_data.append(test_json)
return pd.DataFrame(report_data)
def get_helicone_data():
helicone_api_key = os.getenv("HELICONE_API_KEY")
url = "https://www.helicone.ai/api/graphql"
# Replace <KEY> with your personal access key
transport = AIOHTTPTransport(
url=url, headers={"authorization": f"Bearer {helicone_api_key}"}
)
client = Client(transport=transport, fetch_schema_from_transport=True)
SIZE = 250
i = 0
data = []
print("Fetching data from Helicone")
while True:
query = gql(
"""
query ExampleQuery($limit: Int, $offset: Int){
heliconeRequest(
limit: $limit
offset: $offset
) {
costUSD
prompt
properties{
name
value
}
requestBody
response
createdAt
}
}
"""
)
print(f"Fetching {i * SIZE} to {(i + 1) * SIZE} records")
try:
result = client.execute(
query, variable_values={"limit": SIZE, "offset": i * SIZE}
)
except Exception as e:
print(f"Error occurred: {e}")
result = None
i += 1
if result:
for item in result["heliconeRequest"]:
properties = {
prop["name"]: prop["value"] for prop in item["properties"]
}
data.append(
{
"createdAt": item["createdAt"],
"agent": properties.get("agent"),
"costUSD": item["costUSD"],
"job_id": properties.get("job_id"),
"challenge": properties.get("challenge"),
"benchmark_start_time": properties.get("benchmark_start_time"),
"prompt": item["prompt"],
"response": item["response"],
"model": item["requestBody"].get("model"),
"request": item["requestBody"].get("messages"),
}
)
if not result or (len(result["heliconeRequest"]) == 0):
print("No more results")
break
df = pd.DataFrame(data)
# Drop rows where agent is None
df = df.dropna(subset=["agent"])
# Convert the remaining agent names to lowercase
df["agent"] = df["agent"].str.lower()
return df
if os.path.exists("raw_reports.pkl") and os.path.exists("raw_helicone.pkl"):
reports_df = pd.read_pickle("raw_reports.pkl")
helicone_df = pd.read_pickle("raw_helicone.pkl")
else:
reports_df = get_reports()
reports_df.to_pickle("raw_reports.pkl")
helicone_df = get_helicone_data()
helicone_df.to_pickle("raw_helicone.pkl")
def try_formats(date_str):
formats = ["%Y-%m-%d-%H:%M", "%Y-%m-%dT%H:%M:%S%z"]
for fmt in formats:
try:
return pd.to_datetime(date_str, format=fmt)
except ValueError:
pass
return None
helicone_df["benchmark_start_time"] = pd.to_datetime(
helicone_df["benchmark_start_time"].apply(try_formats), utc=True
)
helicone_df = helicone_df.dropna(subset=["benchmark_start_time"])
helicone_df["createdAt"] = pd.to_datetime(
helicone_df["createdAt"], unit="ms", origin="unix"
)
reports_df["benchmark_start_time"] = pd.to_datetime(
reports_df["benchmark_start_time"].apply(try_formats), utc=True
)
reports_df = reports_df.dropna(subset=["benchmark_start_time"])
assert pd.api.types.is_datetime64_any_dtype(
helicone_df["benchmark_start_time"]
), "benchmark_start_time in helicone_df is not datetime"
assert pd.api.types.is_datetime64_any_dtype(
reports_df["benchmark_start_time"]
), "benchmark_start_time in reports_df is not datetime"
reports_df["report_time"] = reports_df["benchmark_start_time"]
# df = pd.merge_asof(
# helicone_df.sort_values("benchmark_start_time"),
# reports_df.sort_values("benchmark_start_time"),
# left_on="benchmark_start_time",
# right_on="benchmark_start_time",
# by=["agent", "challenge"],
# direction="backward",
# )
df = pd.merge(
helicone_df,
reports_df,
on=["benchmark_start_time", "agent", "challenge"],
how="inner",
)
df.to_pickle("df.pkl")
print(df.info())
print("Data saved to df.pkl")
print("To load the data use: df = pd.read_pickle('df.pkl')")
|