Spaces:
Runtime error
Runtime error
File size: 14,749 Bytes
82130cb a732fe2 843d9d3 a732fe2 843d9d3 a732fe2 82130cb a732fe2 82130cb a732fe2 82130cb a732fe2 e912278 a732fe2 82130cb a732fe2 82130cb 1fb12dc 82130cb e912278 843d9d3 82130cb e912278 82130cb e912278 82130cb e912278 82130cb e912278 82130cb e912278 82130cb e912278 82130cb 1fb12dc e912278 1fb12dc 82130cb e912278 82130cb 1fb12dc 82130cb e912278 82130cb 843d9d3 e912278 843d9d3 e912278 843d9d3 e912278 843d9d3 e912278 843d9d3 e912278 843d9d3 e912278 843d9d3 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 |
from __future__ import annotations # For self-referencing annotations
import json
import os
import shutil
import sqlite3
import sys
from huggingface_hub import Repository
from random import choices
from typing import List, Dict, Optional
from src.architectures import Architecture
from src.common import data_dir
class TestGenerator:
"""
Wrapper class to hold testing questions and serve up examples
"""
questions: List[str] = None
@classmethod
def load_questions(cls, reload=False) -> None:
"""
Load the available questions from the json file.
Default to not re-loading if already done, but allow for the option to do so
"""
if cls.questions is not None and not reload:
return
question_file = os.path.join(data_dir, 'json', 'test_questions.json')
with open(question_file, 'r') as f:
question_json = json.load(f)
cls.questions = question_json['questions']
@classmethod
def question_count(cls) -> int:
"""
The total number of questions in the question set
"""
cls.load_questions()
return len(cls.questions)
@classmethod
def get_random_questions(cls, n: int):
"""
Return n random questions
"""
cls.load_questions()
return choices(cls.questions, k=n)
class ArchitectureRequestRecord:
"""
Representation of the test data associated with each invocation of an architecture
"""
all: List[ArchitectureRequestRecord] = None
class ArchStep:
"""
Inner class to just hold this data
"""
def __init__(self, name: str, start: int, end: int):
self.name = name
self.start = start
self.end = end
self.elapsed = end - start
def __init__(self, arch: str, response_len: int, start: int, end: int,
elapsed: int, tags: List[str], test_group: Optional[str],
comment: str, steps: List[ArchitectureRequestRecord.ArchStep]):
self.arch = arch
self.response_len = response_len
self.start = start
self.end = end
self.elapsed = elapsed
self.tags = tags
self.test_group = test_group
self.comment = comment
self.steps = steps
@classmethod
def from_dict(cls, test: Dict) -> ArchitectureRequestRecord:
arch = test['architecture']
response_len = len(test['request']['response_evolution'][-1])
start = test['trace']['steps'][0]['start_ms']
end = test['trace']['steps'][-1]['end_ms']
elapsed = end - start
tags = test['test_tags']
test_group = None
for tag in tags:
if tag.startswith("TestGroup"):
test_group = tag
comment = test['test_comment']
steps = []
for s in test['trace']['steps']:
steps.append(ArchitectureRequestRecord.ArchStep(s['name'], s['start_ms'], s['end_ms']))
return ArchitectureRequestRecord( arch, response_len, start, end, elapsed, tags, test_group, comment, steps)
@classmethod
def load_all(cls, reload=False) -> None:
"""
Load all the traces from json trace log
"""
if cls.all is None or reload:
records = []
test_traces = Architecture.get_trace_records()
for trace in test_traces:
records.append(ArchitectureRequestRecord.from_dict(trace))
cls.all = records
class TestGroup:
"""
A class representing a single batch run of tests from the UI. Identified by the tag
which was assigned from the UI when the test was run, and including summary items
(start, end, elapse) for convenience
"""
all: Dict[str, TestGroup] = None
def __init__(self, test_group:str):
self.arch_request_records: List[ArchitectureRequestRecord] = []
self.test_group = test_group
self.comment = None
self.start = None
self.end = None
self.elapsed = None
self.architectures = set()
@property
def num_archs(self) -> int:
"""
The number of LLM Architectures which were included in this test run from the UI
"""
return len(self.architectures)
@property
def num_tests(self) -> int:
"""
The total number of Architecture tests (inferences) done in this test run from the UI
"""
return len(self.arch_request_records)
@property
def num_tests_per_arch(self) -> int:
"""
The calculated number of tests run through each architecture (simple divide as the UI
forces each architecture to get the same number of requests)
"""
# Should always be an even number but cast to int just in case
return int(self.num_tests / self.num_archs)
def arch_request_records_by_arch(self) -> Dict[List[ArchitectureRequestRecord]]:
"""
Get all the tests ArchitectureRequestRecords grouped by the architecture.
:return: dict keyed by the architecture name containing a list of ArchitectureRequestRecords
detailing the tests run through that architecture. Note - the keys are intended to be used for
display purposes - attempting to use them to load the original architecture will be
dependent on the availability of that architecture at look up time and changes in architecture
config could cause that lookup to fail (i.e. the tested architecture is no longer configured).
"""
grouped = {}
for arr in self.arch_request_records:
if arr.arch not in grouped:
grouped[arr.arch] = []
grouped[arr.arch].append(arr)
return grouped
def summary_stats_by_arch(self) -> List[Dict]:
"""
Get a pack of statistics for use in the UI, detailing this TestGroup.
:return: a list, sorted by architecture name, of statistics per architecture. Each list item
is a dict of information (arch_name, elapsed[list of elapsed times in ms],
response_len[list of the lengthe of the final response in characters], steps[list of the
individual architecture steps each containing dict of name, mean_elapsed(ms)])
"""
arch_records = self.arch_request_records_by_arch()
arch_names = list(arch_records.keys())
arch_names.sort()
stats = []
for a in arch_names:
stat_pack = {'arch_name': a, 'elapsed': [rec.elapsed for rec in arch_records[a]],
'response_len': [rec.response_len for rec in arch_records[a]], 'steps': []}
for i in range(len(arch_records[a][0].steps)):
stat_pack['steps'].append({'step_name': arch_records[a][0].steps[i].name})
num_recs = len(arch_records[a])
total_elapsed = 0
for j in range(num_recs):
total_elapsed += arch_records[a][j].steps[i].elapsed
stat_pack['steps'][-1]['mean_elapsed'] = total_elapsed / num_recs
stats.append(stat_pack)
return stats
def add_record(self, arr: ArchitectureRequestRecord) -> None:
"""
Add an ArchitectureRequestRecord into this test group. Update the
TestGroup level start, end and elapsed with the new data
"""
if arr.test_group != self.test_group:
raise ValueError("Attempted to group a test record into the wrong group")
self.arch_request_records.append(arr)
self.architectures.add(arr.arch)
if self.comment is None:
self.comment = arr.comment
if self.start is None or self.start > arr.start:
self.start = arr.start
if self.end is None or self.end < arr.end:
self.end = arr.end
self.elapsed = self.end - self.start
@classmethod
def load_db_records(cls) -> List[TestGroup]:
"""
Load all the test groups from the DataBase
"""
db_file = os.path.join(data_dir, 'sqlite', 'test_records.db')
con = sqlite3.connect(db_file)
cur = con.cursor()
sql = "SELECT id, test_group from test_groups"
cur.execute(sql)
tg_id_names = [(r[0], r[1]) for r in cur.fetchall()]
test_groups: List[TestGroup] = []
for tg_id, tg_name in tg_id_names:
tg = TestGroup(tg_name)
sql = f"SELECT id, arch_name, response_len, start, end, comment FROM arch_requests WHERE test_group_id={tg_id}"
cur.execute(sql)
arch_requests = [(r[0], r[1], r[2], r[3], r[4], r[5]) for r in cur.fetchall()]
for ar_id, ar_arch_name, ar_resp_len, ar_start, ar_end, ar_comment in arch_requests:
sql = f"SELECT name, start, end FROM arch_req_steps WHERE arch_req_id={ar_id}"
cur.execute(sql)
steps = [ArchitectureRequestRecord.ArchStep(r[0], r[1], r[2]) for r in cur.fetchall()]
arch_req_record = ArchitectureRequestRecord(ar_arch_name, ar_resp_len, ar_start, ar_end, (ar_end - ar_start), [tg_name], tg_name, f"(DB persisted) {ar_comment}", steps)
tg.add_record(arch_req_record)
test_groups.append(tg)
return test_groups
@classmethod
def load_all(cls, reload=False):
"""
Load all the available TestGroups, from both the json file and the DB
into the class variable - for efficiency do not reload unless requested
"""
if cls.all is None or reload:
ArchitectureRequestRecord.load_all(reload=reload)
records = {}
for arr in ArchitectureRequestRecord.all:
if arr.test_group is not None:
if arr.test_group not in records:
records[arr.test_group] = TestGroup(arr.test_group)
records[arr.test_group].add_record(arr)
db_records = cls.load_db_records()
for test_group in db_records:
records[test_group.test_group] = test_group
cls.all = records
@classmethod
def for_test_group_tag(cls, test_group_tag: str) -> TestGroup:
"""
Get a single TestGroup based on the test_group_tag which was assigned
when the test was run
"""
cls.load_all()
return cls.all[test_group_tag]
def move_test_records_to_db(hf_hub_token: str) -> None:
"""
This is an offline utility to move the transaction logs from
a flat file to a database. To keep things simpler transaction logs
are initially stored into a json file as a Hugging Face dataset, but
this can get cumbersome, so this utility will move the records
into an sqlite database. It can be run periodically just to move
them across.
"""
def download_latest_json_file(hf_hub_token: str) -> Repository:
"""
Wipe any local version of the json file and re-downlad from the HF Hub
"""
if os.path.exists(Architecture.trace_dir):
shutil.rmtree(Architecture.trace_dir)
return Repository(local_dir=Architecture.trace_dir, clone_from=Architecture.save_repo_url, token=hf_hub_token)
def create_local_db():
"""
Create the local database if it does not exist
"""
db_file = os.path.join(data_dir, 'sqlite', 'test_records.db')
con = sqlite3.connect(db_file)
sql = "CREATE TABLE test_groups (id INTEGER PRIMARY KEY AUTOINCREMENT, test_group TEXT NOT NULL, start INTEGER NOT NULL, end INTEGER NOT NULL);"
con.execute(sql)
sql = "CREATE TABLE arch_requests (id INTEGER PRIMARY KEY AUTOINCREMENT, arch_name TEXT NOT NULL, response_len INTEGER NOT NULL, start INTEGER NOT NULL, end INTEGER NOT NULL, comment TEXT NOT NULL, test_group_id INTEGER NOT NULL, FOREIGN KEY (test_group_id) REFERENCES test_groups (id))"
con.execute(sql)
sql = "CREATE TABLE arch_req_steps (id INTEGER PRIMARY KEY AUTOINCREMENT, name TEXT NOT NULL, start INTEGER NOT NULL, end INTEGER NOT NULL, arch_req_id INTEGER NOT NULL, FOREIGN KEY (arch_req_id) REFERENCES arch_requests(id))"
con.execute(sql)
def get_local_db() -> sqlite3.Connection:
"""
Get a connection to the local database and create it if it is not already there
"""
db_file = os.path.join(data_dir, 'sqlite', 'test_records.db')
if not os.path.exists(db_file):
create_local_db()
return sqlite3.connect(db_file)
def load_test_group_to_db(test_group: TestGroup, con: sqlite3.Connection) -> None:
"""
Load a single TestGroup object into the DB, decomposing to the TestGroup,
ArchitectureRequest within that and ArchitectureRequestSteps within those
"""
cur = con.cursor()
sql = f'SELECT count(*) from test_groups where test_group ="{test_group.test_group}"'
cur.execute(sql)
tg_not_in_db = cur.fetchall()[0][0] == 0
if tg_not_in_db:
sql = f'INSERT into test_groups (test_group, start, end) VALUES ("{test_group.test_group}", {test_group.start}, {test_group.end})'
tg_id = con.execute(sql).lastrowid
for arr in test_group.arch_request_records:
sql = f'INSERT INTO arch_requests (arch_name, response_len, start, end, comment, test_group_id) VALUES ("{arr.arch}", {arr.response_len}, {arr.start}, {arr.end}, "{arr.comment}", {tg_id})'
arr_id = con.execute(sql).lastrowid
for s in arr.steps:
sql= f'INSERT INTO arch_req_steps (name, start, end, arch_req_id) VALUES ("{s.name}", {s.start}, {s.end}, {arr_id})'
con.execute(sql)
con.commit()
else:
print(f"Warning TestGroup {tg_not_in_db} was not added to the DB as it already existed there")
def load_all_test_groups_to_db(con: sqlite3.Connection) -> None:
"""
Load a list of TestGroups to the DB, one at a time
"""
TestGroup.load_all()
for tg in TestGroup.all.values():
load_test_group_to_db(tg, con)
"""
Main control flow using utility nested functions above for better structure and readability
"""
download_latest_json_file(hf_hub_token)
conn = get_local_db()
load_all_test_groups_to_db(conn)
Architecture.wipe_trace(hf_hub_token)
print("REMINDER: need to commit the local sqlite file to make it available to the server")
if __name__ == "__main__":
# Expected to only be directly called for the json to db transfer - arg should be the HF token
hf_token = sys.argv[1]
move_test_records_to_db(hf_token)
|