File size: 17,235 Bytes
246d201 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 |
import copy
from unittest.mock import MagicMock, patch
import pytest
from litellm.exceptions import (
APIConnectionError,
InternalServerError,
RateLimitError,
ServiceUnavailableError,
)
from openhands.core.config import LLMConfig
from openhands.core.exceptions import OperationCancelled
from openhands.core.message import Message, TextContent
from openhands.llm.llm import LLM
from openhands.llm.metrics import Metrics
@pytest.fixture(autouse=True)
def mock_logger(monkeypatch):
# suppress logging of completion data to file
mock_logger = MagicMock()
monkeypatch.setattr('openhands.llm.debug_mixin.llm_prompt_logger', mock_logger)
monkeypatch.setattr('openhands.llm.debug_mixin.llm_response_logger', mock_logger)
monkeypatch.setattr('openhands.llm.llm.logger', mock_logger)
return mock_logger
@pytest.fixture
def default_config():
return LLMConfig(
model='gpt-4o',
api_key='test_key',
num_retries=2,
retry_min_wait=1,
retry_max_wait=2,
)
def test_llm_init_with_default_config(default_config):
llm = LLM(default_config)
assert llm.config.model == 'gpt-4o'
assert llm.config.api_key.get_secret_value() == 'test_key'
assert isinstance(llm.metrics, Metrics)
assert llm.metrics.model_name == 'gpt-4o'
@patch('openhands.llm.llm.litellm.get_model_info')
def test_llm_init_with_model_info(mock_get_model_info, default_config):
mock_get_model_info.return_value = {
'max_input_tokens': 8000,
'max_output_tokens': 2000,
}
llm = LLM(default_config)
llm.init_model_info()
assert llm.config.max_input_tokens == 8000
assert llm.config.max_output_tokens == 2000
@patch('openhands.llm.llm.litellm.get_model_info')
def test_llm_init_without_model_info(mock_get_model_info, default_config):
mock_get_model_info.side_effect = Exception('Model info not available')
llm = LLM(default_config)
llm.init_model_info()
assert llm.config.max_input_tokens == 4096
assert llm.config.max_output_tokens == 4096
def test_llm_init_with_custom_config():
custom_config = LLMConfig(
model='custom-model',
api_key='custom_key',
max_input_tokens=5000,
max_output_tokens=1500,
temperature=0.8,
top_p=0.9,
)
llm = LLM(custom_config)
assert llm.config.model == 'custom-model'
assert llm.config.api_key.get_secret_value() == 'custom_key'
assert llm.config.max_input_tokens == 5000
assert llm.config.max_output_tokens == 1500
assert llm.config.temperature == 0.8
assert llm.config.top_p == 0.9
def test_llm_init_with_metrics():
config = LLMConfig(model='gpt-4o', api_key='test_key')
metrics = Metrics()
llm = LLM(config, metrics=metrics)
assert llm.metrics is metrics
assert (
llm.metrics.model_name == 'default'
) # because we didn't specify model_name in Metrics init
@patch('openhands.llm.llm.litellm_completion')
@patch('time.time')
def test_response_latency_tracking(mock_time, mock_litellm_completion):
# Mock time.time() to return controlled values
mock_time.side_effect = [1000.0, 1002.5] # Start time, end time (2.5s difference)
# Mock the completion response with a specific ID
mock_response = {
'id': 'test-response-123',
'choices': [{'message': {'content': 'Test response'}}],
}
mock_litellm_completion.return_value = mock_response
# Create LLM instance and make a completion call
config = LLMConfig(model='gpt-4o', api_key='test_key')
llm = LLM(config)
response = llm.completion(messages=[{'role': 'user', 'content': 'Hello!'}])
# Verify the response latency was tracked correctly
assert len(llm.metrics.response_latencies) == 1
latency_record = llm.metrics.response_latencies[0]
assert latency_record.model == 'gpt-4o'
assert (
latency_record.latency == 2.5
) # Should be the difference between our mocked times
assert latency_record.response_id == 'test-response-123'
# Verify the completion response was returned correctly
assert response['id'] == 'test-response-123'
assert response['choices'][0]['message']['content'] == 'Test response'
# To make sure the metrics fail gracefully, set the start/end time to go backwards.
mock_time.side_effect = [1000.0, 999.0]
llm.completion(messages=[{'role': 'user', 'content': 'Hello!'}])
# There should now be 2 latencies, the last of which has the value clipped to 0
assert len(llm.metrics.response_latencies) == 2
latency_record = llm.metrics.response_latencies[-1]
assert latency_record.latency == 0.0 # Should be lifted to 0 instead of being -1!
def test_llm_reset():
llm = LLM(LLMConfig(model='gpt-4o-mini', api_key='test_key'))
initial_metrics = copy.deepcopy(llm.metrics)
initial_metrics.add_cost(1.0)
initial_metrics.add_response_latency(0.5, 'test-id')
llm.reset()
assert llm.metrics.accumulated_cost != initial_metrics.accumulated_cost
assert llm.metrics.costs != initial_metrics.costs
assert llm.metrics.response_latencies != initial_metrics.response_latencies
assert isinstance(llm.metrics, Metrics)
@patch('openhands.llm.llm.litellm.get_model_info')
def test_llm_init_with_openrouter_model(mock_get_model_info, default_config):
default_config.model = 'openrouter:gpt-4o-mini'
mock_get_model_info.return_value = {
'max_input_tokens': 7000,
'max_output_tokens': 1500,
}
llm = LLM(default_config)
llm.init_model_info()
assert llm.config.max_input_tokens == 7000
assert llm.config.max_output_tokens == 1500
mock_get_model_info.assert_called_once_with('openrouter:gpt-4o-mini')
# Tests involving completion and retries
@patch('openhands.llm.llm.litellm_completion')
def test_completion_with_mocked_logger(
mock_litellm_completion, default_config, mock_logger
):
mock_litellm_completion.return_value = {
'choices': [{'message': {'content': 'Test response'}}]
}
llm = LLM(config=default_config)
response = llm.completion(
messages=[{'role': 'user', 'content': 'Hello!'}],
stream=False,
)
assert response['choices'][0]['message']['content'] == 'Test response'
assert mock_litellm_completion.call_count == 1
mock_logger.debug.assert_called()
@pytest.mark.parametrize(
'exception_class,extra_args,expected_retries',
[
(
APIConnectionError,
{'llm_provider': 'test_provider', 'model': 'test_model'},
2,
),
(
InternalServerError,
{'llm_provider': 'test_provider', 'model': 'test_model'},
2,
),
(
ServiceUnavailableError,
{'llm_provider': 'test_provider', 'model': 'test_model'},
2,
),
(RateLimitError, {'llm_provider': 'test_provider', 'model': 'test_model'}, 2),
],
)
@patch('openhands.llm.llm.litellm_completion')
def test_completion_retries(
mock_litellm_completion,
default_config,
exception_class,
extra_args,
expected_retries,
):
mock_litellm_completion.side_effect = [
exception_class('Test error message', **extra_args),
{'choices': [{'message': {'content': 'Retry successful'}}]},
]
llm = LLM(config=default_config)
response = llm.completion(
messages=[{'role': 'user', 'content': 'Hello!'}],
stream=False,
)
assert response['choices'][0]['message']['content'] == 'Retry successful'
assert mock_litellm_completion.call_count == expected_retries
@patch('openhands.llm.llm.litellm_completion')
def test_completion_rate_limit_wait_time(mock_litellm_completion, default_config):
with patch('time.sleep') as mock_sleep:
mock_litellm_completion.side_effect = [
RateLimitError(
'Rate limit exceeded', llm_provider='test_provider', model='test_model'
),
{'choices': [{'message': {'content': 'Retry successful'}}]},
]
llm = LLM(config=default_config)
response = llm.completion(
messages=[{'role': 'user', 'content': 'Hello!'}],
stream=False,
)
assert response['choices'][0]['message']['content'] == 'Retry successful'
assert mock_litellm_completion.call_count == 2
mock_sleep.assert_called_once()
wait_time = mock_sleep.call_args[0][0]
assert (
default_config.retry_min_wait <= wait_time <= default_config.retry_max_wait
), f'Expected wait time between {default_config.retry_min_wait} and {default_config.retry_max_wait} seconds, but got {wait_time}'
@patch('openhands.llm.llm.litellm_completion')
def test_completion_exhausts_retries(mock_litellm_completion, default_config):
mock_litellm_completion.side_effect = APIConnectionError(
'Persistent error', llm_provider='test_provider', model='test_model'
)
llm = LLM(config=default_config)
with pytest.raises(APIConnectionError):
llm.completion(
messages=[{'role': 'user', 'content': 'Hello!'}],
stream=False,
)
assert mock_litellm_completion.call_count == llm.config.num_retries
@patch('openhands.llm.llm.litellm_completion')
def test_completion_operation_cancelled(mock_litellm_completion, default_config):
mock_litellm_completion.side_effect = OperationCancelled('Operation cancelled')
llm = LLM(config=default_config)
with pytest.raises(OperationCancelled):
llm.completion(
messages=[{'role': 'user', 'content': 'Hello!'}],
stream=False,
)
assert mock_litellm_completion.call_count == 1
@patch('openhands.llm.llm.litellm_completion')
def test_completion_keyboard_interrupt(mock_litellm_completion, default_config):
def side_effect(*args, **kwargs):
raise KeyboardInterrupt('Simulated KeyboardInterrupt')
mock_litellm_completion.side_effect = side_effect
llm = LLM(config=default_config)
with pytest.raises(OperationCancelled):
try:
llm.completion(
messages=[{'role': 'user', 'content': 'Hello!'}],
stream=False,
)
except KeyboardInterrupt:
raise OperationCancelled('Operation cancelled due to KeyboardInterrupt')
assert mock_litellm_completion.call_count == 1
@patch('openhands.llm.llm.litellm_completion')
def test_completion_keyboard_interrupt_handler(mock_litellm_completion, default_config):
global _should_exit
def side_effect(*args, **kwargs):
global _should_exit
_should_exit = True
return {'choices': [{'message': {'content': 'Simulated interrupt response'}}]}
mock_litellm_completion.side_effect = side_effect
llm = LLM(config=default_config)
result = llm.completion(
messages=[{'role': 'user', 'content': 'Hello!'}],
stream=False,
)
assert mock_litellm_completion.call_count == 1
assert result['choices'][0]['message']['content'] == 'Simulated interrupt response'
assert _should_exit
_should_exit = False
@patch('openhands.llm.llm.litellm_completion')
def test_completion_with_litellm_mock(mock_litellm_completion, default_config):
mock_response = {
'choices': [{'message': {'content': 'This is a mocked response.'}}]
}
mock_litellm_completion.return_value = mock_response
test_llm = LLM(config=default_config)
response = test_llm.completion(
messages=[{'role': 'user', 'content': 'Hello!'}],
stream=False,
drop_params=True,
)
# Assertions
assert response['choices'][0]['message']['content'] == 'This is a mocked response.'
mock_litellm_completion.assert_called_once()
# Check if the correct arguments were passed to litellm_completion
call_args = mock_litellm_completion.call_args[1] # Get keyword arguments
assert call_args['model'] == default_config.model
assert call_args['messages'] == [{'role': 'user', 'content': 'Hello!'}]
assert not call_args['stream']
@patch('openhands.llm.llm.litellm_completion')
def test_completion_with_two_positional_args(mock_litellm_completion, default_config):
mock_response = {
'choices': [{'message': {'content': 'Response to positional args.'}}]
}
mock_litellm_completion.return_value = mock_response
test_llm = LLM(config=default_config)
response = test_llm.completion(
'some-model-to-be-ignored',
[{'role': 'user', 'content': 'Hello from positional args!'}],
stream=False,
)
# Assertions
assert (
response['choices'][0]['message']['content'] == 'Response to positional args.'
)
mock_litellm_completion.assert_called_once()
# Check if the correct arguments were passed to litellm_completion
call_args, call_kwargs = mock_litellm_completion.call_args
assert (
call_kwargs['model'] == default_config.model
) # Should use the model from config, not the first arg
assert call_kwargs['messages'] == [
{'role': 'user', 'content': 'Hello from positional args!'}
]
assert not call_kwargs['stream']
# Ensure the first positional argument (model) was ignored
assert (
len(call_args) == 0
) # No positional args should be passed to litellm_completion here
@patch('openhands.llm.llm.litellm_completion')
def test_llm_cloudflare_blockage(mock_litellm_completion, default_config):
from litellm.exceptions import APIError
from openhands.core.exceptions import CloudFlareBlockageError
llm = LLM(default_config)
mock_litellm_completion.side_effect = APIError(
message='Attention Required! | Cloudflare',
llm_provider='test_provider',
model='test_model',
status_code=403,
)
with pytest.raises(CloudFlareBlockageError, match='Request blocked by CloudFlare'):
llm.completion(messages=[{'role': 'user', 'content': 'Hello'}])
# Ensure the completion was called
mock_litellm_completion.assert_called_once()
@patch('openhands.llm.llm.litellm.token_counter')
def test_get_token_count_with_dict_messages(mock_token_counter, default_config):
mock_token_counter.return_value = 42
llm = LLM(default_config)
messages = [{'role': 'user', 'content': 'Hello!'}]
token_count = llm.get_token_count(messages)
assert token_count == 42
mock_token_counter.assert_called_once_with(
model=default_config.model, messages=messages, custom_tokenizer=None
)
@patch('openhands.llm.llm.litellm.token_counter')
def test_get_token_count_with_message_objects(
mock_token_counter, default_config, mock_logger
):
llm = LLM(default_config)
# Create a Message object and its equivalent dict
message_obj = Message(role='user', content=[TextContent(text='Hello!')])
message_dict = {'role': 'user', 'content': 'Hello!'}
# Mock token counter to return different values for each call
mock_token_counter.side_effect = [42, 42] # Same value for both cases
# Get token counts for both formats
token_count_obj = llm.get_token_count([message_obj])
token_count_dict = llm.get_token_count([message_dict])
# Verify both formats get the same token count
assert token_count_obj == token_count_dict
assert mock_token_counter.call_count == 2
@patch('openhands.llm.llm.litellm.token_counter')
@patch('openhands.llm.llm.create_pretrained_tokenizer')
def test_get_token_count_with_custom_tokenizer(
mock_create_tokenizer, mock_token_counter, default_config
):
mock_tokenizer = MagicMock()
mock_create_tokenizer.return_value = mock_tokenizer
mock_token_counter.return_value = 42
config = copy.deepcopy(default_config)
config.custom_tokenizer = 'custom/tokenizer'
llm = LLM(config)
messages = [{'role': 'user', 'content': 'Hello!'}]
token_count = llm.get_token_count(messages)
assert token_count == 42
mock_create_tokenizer.assert_called_once_with('custom/tokenizer')
mock_token_counter.assert_called_once_with(
model=config.model, messages=messages, custom_tokenizer=mock_tokenizer
)
@patch('openhands.llm.llm.litellm.token_counter')
def test_get_token_count_error_handling(
mock_token_counter, default_config, mock_logger
):
mock_token_counter.side_effect = Exception('Token counting failed')
llm = LLM(default_config)
messages = [{'role': 'user', 'content': 'Hello!'}]
token_count = llm.get_token_count(messages)
assert token_count == 0
mock_token_counter.assert_called_once()
mock_logger.error.assert_called_once_with(
'Error getting token count for\n model gpt-4o\nToken counting failed'
)
|