File size: 30,260 Bytes
a402b9b | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 | """
Unit-tests for OpenAIServingChat — rewritten to use only the std-lib 'unittest'.
Run with either:
python tests/test_serving_chat_unit.py -v
or
python -m unittest discover -s tests -p "test_*unit.py" -v
"""
import json
import unittest
import uuid
from http import HTTPStatus
from typing import Optional
from unittest.mock import Mock, patch
from fastapi import Request
from sglang.srt.entrypoints.openai.protocol import (
ChatCompletionRequest,
MessageProcessingResult,
)
from sglang.srt.entrypoints.openai.serving_chat import OpenAIServingChat
from sglang.srt.managers.io_struct import GenerateReqInput
from sglang.srt.utils import get_or_create_event_loop
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
register_cuda_ci(est_time=10, suite="stage-b-test-small-1-gpu")
register_amd_ci(est_time=10, suite="stage-b-test-small-1-gpu-amd")
class _MockTokenizerManager:
"""Minimal mock that satisfies OpenAIServingChat."""
def __init__(self):
self.model_config = Mock(is_multimodal=False)
self.server_args = Mock(
enable_cache_report=False,
tool_call_parser="hermes",
reasoning_parser=None,
)
# Mock hf_config for _use_dpsk_v32_encoding check
mock_hf_config = Mock()
mock_hf_config.architectures = ["LlamaForCausalLM"]
self.model_config.hf_config = mock_hf_config
self.chat_template_name: Optional[str] = "llama-3"
# tokenizer stub
self.tokenizer = Mock()
self.tokenizer.encode.return_value = [1, 2, 3, 4, 5]
self.tokenizer.decode.return_value = "Test response"
self.tokenizer.chat_template = None
self.tokenizer.bos_token_id = 1
# async generator stub for generate_request
async def _mock_generate():
yield {
"text": "Test response",
"meta_info": {
"id": f"chatcmpl-{uuid.uuid4()}",
"prompt_tokens": 10,
"completion_tokens": 5,
"cached_tokens": 0,
"finish_reason": {"type": "stop", "matched": None},
"output_token_logprobs": [(0.1, 1, "Test"), (0.2, 2, "response")],
"output_top_logprobs": None,
},
"index": 0,
}
self.generate_request = Mock(return_value=_mock_generate())
self.create_abort_task = Mock()
class _MockTemplateManager:
"""Minimal mock for TemplateManager."""
def __init__(self):
self.chat_template_name: Optional[str] = "llama-3"
self.jinja_template_content_format: Optional[str] = None
self.completion_template_name: Optional[str] = None
class ServingChatTestCase(unittest.TestCase):
# ------------- common fixtures -------------
def setUp(self):
self.tm = _MockTokenizerManager()
self.template_manager = _MockTemplateManager()
self.chat = OpenAIServingChat(self.tm, self.template_manager)
# frequently reused requests
self.basic_req = ChatCompletionRequest(
model="x",
messages=[{"role": "user", "content": "Hi?"}],
temperature=0.7,
max_tokens=100,
stream=False,
)
self.stream_req = ChatCompletionRequest(
model="x",
messages=[{"role": "user", "content": "Hi?"}],
temperature=0.7,
max_tokens=100,
stream=True,
)
self.fastapi_request = Mock(spec=Request)
self.fastapi_request.headers = {}
# ------------- conversion tests -------------
def test_convert_to_internal_request_single(self):
with patch(
"sglang.srt.entrypoints.openai.serving_chat.generate_chat_conv"
) as conv_mock, patch.object(self.chat, "_process_messages") as proc_mock:
conv_ins = Mock()
conv_ins.get_prompt.return_value = "Test prompt"
conv_ins.image_data = conv_ins.audio_data = None
conv_ins.modalities = []
conv_ins.stop_str = ["</s>"]
conv_mock.return_value = conv_ins
proc_mock.return_value = MessageProcessingResult(
"Test prompt",
[1, 2, 3],
None,
None,
[],
["</s>"],
None,
)
adapted, processed = self.chat._convert_to_internal_request(self.basic_req)
self.assertIsInstance(adapted, GenerateReqInput)
self.assertFalse(adapted.stream)
self.assertEqual(processed, self.basic_req)
def test_jinja_uses_openai_tool_schema_first(self):
"""Ensure Jinja chat templates receive OpenAI-shaped tools by default."""
self.template_manager.chat_template_name = None
self.template_manager.jinja_template_content_format = "string"
req = ChatCompletionRequest(
model="x",
messages=[{"role": "user", "content": "What is 2+2?"}],
tools=[
{
"type": "function",
"function": {
"name": "add",
"description": "Add two numbers.",
"parameters": {
"type": "object",
"properties": {
"a": {"type": "integer"},
"b": {"type": "integer"},
},
"required": ["a", "b"],
},
},
}
],
)
self.chat._process_messages(req, is_multimodal=False)
expected_tools = [tool.model_dump() for tool in req.tools]
kwargs = self.tm.tokenizer.apply_chat_template.call_args.kwargs
self.assertEqual(kwargs["tools"], expected_tools)
def test_jinja_tool_schema_fallback_to_flat_function(self):
"""Fallback to function-only schema when template rejects OpenAI wrapper."""
self.template_manager.chat_template_name = None
self.template_manager.jinja_template_content_format = "string"
req = ChatCompletionRequest(
model="x",
messages=[{"role": "user", "content": "What is 2+2?"}],
tools=[
{
"type": "function",
"function": {
"name": "add",
"description": "Add two numbers.",
"parameters": {
"type": "object",
"properties": {
"a": {"type": "integer"},
"b": {"type": "integer"},
},
"required": ["a", "b"],
},
},
}
],
)
self.tm.tokenizer.apply_chat_template.side_effect = [
RuntimeError("template expects flat tools format"),
[1, 2, 3],
]
self.chat._process_messages(req, is_multimodal=False)
first_tools = self.tm.tokenizer.apply_chat_template.call_args_list[0].kwargs[
"tools"
]
second_tools = self.tm.tokenizer.apply_chat_template.call_args_list[1].kwargs[
"tools"
]
self.assertEqual(first_tools, [tool.model_dump() for tool in req.tools])
self.assertEqual(
second_tools, [tool.function.model_dump() for tool in req.tools]
)
def test_stop_str_isolation_between_requests(self):
"""Test that stop strings from one request don't affect subsequent requests.
This tests the fix for the bug where conv.stop_str was being mutated globally,
causing stop strings from one request to persist in subsequent requests.
"""
# Mock conversation template with initial stop_str
initial_stop_str = ["\n"]
with patch(
"sglang.srt.entrypoints.openai.serving_chat.generate_chat_conv"
) as conv_mock:
# Create a mock conversation object that will be returned by generate_chat_conv
conv_ins = Mock()
conv_ins.get_prompt.return_value = "Test prompt"
conv_ins.image_data = None
conv_ins.audio_data = None
conv_ins.modalities = []
conv_ins.stop_str = (
initial_stop_str.copy()
) # Template's default stop strings
conv_mock.return_value = conv_ins
# First request with additional stop string
req1 = ChatCompletionRequest(
model="x",
messages=[{"role": "user", "content": "First request"}],
stop=["CUSTOM_STOP"],
)
# Call the actual _apply_conversation_template method (not mocked)
result1 = self.chat._apply_conversation_template(req1, is_multimodal=False)
# Verify first request has both stop strings
expected_stop1 = initial_stop_str + ["CUSTOM_STOP"]
self.assertEqual(result1.stop, expected_stop1)
# Verify the original template's stop_str wasn't mutated after first request
self.assertEqual(conv_ins.stop_str, initial_stop_str)
# Second request without additional stop string
req2 = ChatCompletionRequest(
model="x",
messages=[{"role": "user", "content": "Second request"}],
# No custom stop strings
)
result2 = self.chat._apply_conversation_template(req2, is_multimodal=False)
# Verify second request only has original stop strings (no CUSTOM_STOP from req1)
self.assertEqual(result2.stop, initial_stop_str)
self.assertNotIn("CUSTOM_STOP", result2.stop)
self.assertEqual(conv_ins.stop_str, initial_stop_str)
def test_unstreamed_tool_args_completion(self):
"""Test that remaining tool call arguments are sent when generation finishes."""
# Mock FunctionCallParser with detector that has partial tool call data
mock_parser = Mock()
mock_detector = Mock()
# Simulate a tool call that was partially streamed
mock_detector.prev_tool_call_arr = [
{
"name": "get_weather",
"arguments": {"location": "San Francisco", "unit": "celsius"},
}
]
mock_detector.streamed_args_for_tool = [
'{"location": "San Francisco"' # Partial arguments streamed so far
]
mock_parser.detector = mock_detector
content = {
"meta_info": {
"id": "chatcmpl-test123",
}
}
request = ChatCompletionRequest(
model="test",
messages=[{"role": "user", "content": "What's the weather?"}],
tools=[{"type": "function", "function": {"name": "get_weather"}}],
)
# Test the completion method
result = self.chat._check_for_unstreamed_tool_args(
parser=mock_parser,
content=content,
request=request,
index=0,
)
# Should return a chunk with remaining arguments
self.assertIsNotNone(result, "Should return chunk with remaining arguments")
# Parse the result to verify content
self.assertTrue(result.startswith("data: "))
chunk = json.loads(result[6:])
tool_calls = chunk["choices"][0]["delta"]["tool_calls"]
self.assertEqual(len(tool_calls), 1)
arguments = tool_calls[0]["function"]["arguments"]
self.assertIn(', "unit": "celsius"}', arguments)
self.assertIn(
'"finish_reason":null',
result,
"Should not include finish_reason in completion chunk",
)
def test_unstreamed_tool_args_no_completion_needed(self):
"""Test that no completion chunk is sent when all arguments were already streamed."""
# Mock FunctionCallParser with detector that has complete tool call data
mock_parser = Mock()
mock_detector = Mock()
# Simulate a tool call that was completely streamed
mock_detector.prev_tool_call_arr = [
{"name": "get_weather", "arguments": {"location": "San Francisco"}}
]
mock_detector.streamed_args_for_tool = [
'{"location": "San Francisco"}' # All arguments already streamed
]
mock_parser.detector = mock_detector
content = {
"meta_info": {
"id": "chatcmpl-test123",
}
}
request = ChatCompletionRequest(
model="test",
messages=[{"role": "user", "content": "What's the weather?"}],
tools=[{"type": "function", "function": {"name": "get_weather"}}],
)
# Test the completion method
result = self.chat._check_for_unstreamed_tool_args(
parser=mock_parser,
content=content,
request=request,
index=0,
)
# Should return None since no completion is needed
self.assertIsNone(result, "Should return None when no completion is needed")
def test_unstreamed_tool_args_no_parser_data(self):
"""Test that no completion chunk is sent when parser has no tool call data."""
# Mock FunctionCallParser with empty detector
mock_parser = Mock()
mock_detector = Mock()
mock_detector.prev_tool_call_arr = []
mock_detector.streamed_args_for_tool = []
mock_parser.detector = mock_detector
content = {
"meta_info": {
"id": "chatcmpl-test123",
}
}
request = ChatCompletionRequest(
model="test",
messages=[{"role": "user", "content": "What's the weather?"}],
tools=[{"type": "function", "function": {"name": "get_weather"}}],
)
# Test the completion method
result = self.chat._check_for_unstreamed_tool_args(
parser=mock_parser,
content=content,
request=request,
index=0,
)
# Should return None since there's no parser data
self.assertIsNone(
result, "Should return None when parser has no tool call data"
)
# ------------- kimi_k2 tool_call_id formatting -------------
def test_kimi_k2_non_streaming_tool_call_id_format(self):
"""Ensure non-streaming tool_call.id matches functions.{name}:{index} for kimi_k2 parser."""
# Force kimi_k2 parser
self.chat.tool_call_parser = "kimi_k2"
# Mock FunctionCallParser.parse_non_stream to return one tool call
with patch(
"sglang.srt.entrypoints.openai.serving_chat.FunctionCallParser"
) as ParserMock:
parser_instance = ParserMock.return_value
# Build a mock ToolCallItem-like object
call_info = Mock()
call_info.name = "get_weather"
call_info.parameters = '{"city":"Paris"}'
call_info.tool_index = 0
parser_instance.has_tool_call.return_value = True
parser_instance.parse_non_stream.return_value = ("", [call_info])
finish_reason = {"type": "stop", "matched": None}
tools = [
{"type": "function", "function": {"name": "get_weather"}},
]
tool_calls, remaining_text, finish_reason = self.chat._process_tool_calls(
text="<|tool_calls_section_begin|>...",
tools=tools,
finish_reason=finish_reason,
)
self.assertIsNotNone(tool_calls)
self.assertEqual(len(tool_calls), 1)
self.assertEqual(tool_calls[0].id, "functions.get_weather:0")
self.assertEqual(tool_calls[0].function.name, "get_weather")
def test_kimi_k2_streaming_tool_call_id_format(self):
"""Ensure streaming first chunk tool_call.id matches functions.{name}:{index} for kimi_k2 parser."""
# Force kimi_k2 parser
self.chat.tool_call_parser = "kimi_k2"
# Prepare request with tools
req = ChatCompletionRequest(
model="x",
messages=[{"role": "user", "content": "Hi?"}],
tools=[{"type": "function", "function": {"name": "get_weather"}}],
stream=True,
)
# Patch FunctionCallParser used inside _process_tool_call_stream
with patch(
"sglang.srt.entrypoints.openai.serving_chat.FunctionCallParser"
) as ParserMock:
parser_instance = ParserMock.return_value
# First call returns one ToolCallItem-like chunk (with name)
first_chunk_call = Mock()
first_chunk_call.tool_index = 0
first_chunk_call.name = "get_weather"
first_chunk_call.parameters = ""
parser_instance.parse_stream_chunk.side_effect = [
("", [first_chunk_call]),
("", []),
]
async def collect_first_tool_chunk():
gen = self.chat._process_tool_call_stream(
index=0,
delta="irrelevant",
parser_dict={},
content={"meta_info": {"id": "chatcmpl-test"}},
request=req,
has_tool_calls={},
)
# Get first yielded SSE line
line = None
async for emitted in gen:
line = emitted
break
return line
loop = get_or_create_event_loop()
line = loop.run_until_complete(collect_first_tool_chunk())
self.assertIsNotNone(line)
self.assertTrue(line.startswith("data: "))
payload = json.loads(line[len("data: ") :])
tool_calls = payload["choices"][0]["delta"]["tool_calls"]
self.assertEqual(tool_calls[0]["id"], "functions.get_weather:0")
def test_kimi_k2_non_streaming_tool_call_id_with_history(self):
"""Ensure non-streaming tool_call.id increase with tool calls history for kimi_k2 parser."""
# Force kimi_k2 parser
self.chat.tool_call_parser = "kimi_k2"
# Prepare request with tool calls history
req = ChatCompletionRequest(
model="x",
messages=[
{"role": "user", "content": "What's the weather today in paris?"},
{
"role": "assistant",
"content": "Let me do some search first.",
"tool_calls": [
{
"id": "functions.get_weather:0",
"type": "function",
"function": {
"name": "get_weather",
"arguments": '{"city": "Paris"}',
},
}
],
},
{
"role": "tool",
"content": "It's rainy in paris now.",
"tool_call_id": "functions.get_weather:0",
},
{
"role": "assistant",
"content": "It's rainy now.",
},
{
"role": "user",
"content": "What about LA and Tokyo?",
},
],
tools=[{"type": "function", "function": {"name": "get_weather"}}],
stream=False,
)
# Mock FunctionCallParser.parse_non_stream to return one tool call
with patch(
"sglang.srt.entrypoints.openai.serving_chat.FunctionCallParser"
) as ParserMock:
parser_instance = ParserMock.return_value
# Build a mock ToolCallItem-like object
call_info = Mock()
call_info.name = "get_weather"
call_info.parameters = '{"city":"Loa Angeles"}'
# Kimi-K2 series models might generate fixed number tool_indx,
# ignoring the tool calls history and mess up all the following tool calls
call_info.tool_index = 0
call_info2 = Mock()
call_info2.name = "get_weather"
call_info2.parameters = '{"city":"Tokyo"}'
call_info2.tool_index = 1
parser_instance.has_tool_call.return_value = True
parser_instance.parse_non_stream.return_value = (
"",
[call_info, call_info2],
)
finish_reason = {"type": "stop", "matched": None}
tools = [
{"type": "function", "function": {"name": "get_weather"}},
]
history_tool_calls_cnt = self.chat._get_history_tool_calls_cnt(req)
tool_calls, remaining_text, _ = self.chat._process_tool_calls(
text="<|tool_calls_section_begin|>...",
tools=tools,
finish_reason=finish_reason,
history_tool_calls_cnt=history_tool_calls_cnt,
)
self.assertEqual(history_tool_calls_cnt, 1)
self.assertIsNotNone(tool_calls)
self.assertEqual(len(tool_calls), 2)
self.assertEqual(tool_calls[0].id, "functions.get_weather:1")
self.assertEqual(tool_calls[0].function.name, "get_weather")
self.assertEqual(tool_calls[1].id, "functions.get_weather:2")
self.assertEqual(tool_calls[1].function.name, "get_weather")
def test_kimi_k2_streaming_tool_call_id_with_history(self):
"""Ensure streaming first chunk tool_call.id increase with tool calls history for kimi_k2 parser."""
# Force kimi_k2 parser
self.chat.tool_call_parser = "kimi_k2"
# Prepare request with tool calls history
req = ChatCompletionRequest(
model="x",
messages=[
{"role": "user", "content": "What's the weather today in paris?"},
{
"role": "assistant",
"content": "Let me do some search first.",
"tool_calls": [
{
"id": "functions.get_weather:0",
"type": "function",
"function": {
"name": "get_weather",
"arguments": '{"city": "Paris"}',
},
}
],
},
{
"role": "tool",
"content": "It's rainy in paris now.",
"tool_call_id": "functions.get_weather:0",
},
{
"role": "assistant",
"content": "It's rainy now.",
},
{
"role": "user",
"content": "What about LA?",
},
],
tools=[{"type": "function", "function": {"name": "get_weather"}}],
stream=True,
)
# Patch FunctionCallParser used inside _process_tool_call_stream
with patch(
"sglang.srt.entrypoints.openai.serving_chat.FunctionCallParser"
) as ParserMock:
parser_instance = ParserMock.return_value
# First call returns one ToolCallItem-like chunk (with name)
first_chunk_call = Mock()
# Kimi-K2 series models might generate fixed number tool_indx,
# ignoring the tool calls history and mess up all the following tool calls
first_chunk_call.tool_index = 0
first_chunk_call.name = "get_weather"
first_chunk_call.parameters = ""
parser_instance.parse_stream_chunk.side_effect = [
("", [first_chunk_call]),
("", []),
]
async def collect_first_tool_chunk():
gen = self.chat._process_tool_call_stream(
index=0,
delta="irrelevant",
parser_dict={},
content={"meta_info": {"id": "chatcmpl-test"}},
request=req,
has_tool_calls={},
)
# Get first yielded SSE line
line = None
async for emitted in gen:
line = emitted
break
return line
loop = get_or_create_event_loop()
line = loop.run_until_complete(collect_first_tool_chunk())
self.assertIsNotNone(line)
self.assertTrue(line.startswith("data: "))
payload = json.loads(line[len("data: ") :])
tool_calls = payload["choices"][0]["delta"]["tool_calls"]
self.assertEqual(tool_calls[0]["id"], "functions.get_weather:1")
def test_dpsk_v32_encoding_path(self):
"""Test DeepSeek V3.2 encoding path detection and application."""
from sglang.srt.managers.template_manager import TemplateManager
from sglang.srt.server_args import PortArgs, ServerArgs
server_args = ServerArgs(model_path="deepseek-ai/DeepSeek-V3.2")
port_args = PortArgs.init_new(server_args)
# Use mocks for TokenizerManager components to avoid full initialization
with patch(
"sglang.srt.managers.tokenizer_manager.TokenizerManager"
) as MockTokenizerManager:
tokenizer_manager = MockTokenizerManager(server_args, port_args)
tokenizer_manager.server_args = server_args
tokenizer_manager.model_config = Mock()
tokenizer_manager.model_config.get_default_sampling_params.return_value = (
None
)
# Mock hf_config
mock_hf_config = Mock()
mock_hf_config.architectures = ["DeepseekV32ForCausalLM"]
tokenizer_manager.model_config.hf_config = mock_hf_config
# Case 1: No chat template in tokenizer -> should use dpsk encoding
tokenizer_manager.tokenizer = Mock()
tokenizer_manager.tokenizer.chat_template = None
serving_chat = OpenAIServingChat(tokenizer_manager, TemplateManager())
self.assertTrue(serving_chat.use_dpsk_v32_encoding)
# Case 2: Chat template exists -> should NOT use dpsk encoding
tokenizer_manager.tokenizer.chat_template = "some template"
serving_chat = OpenAIServingChat(tokenizer_manager, TemplateManager())
self.assertFalse(serving_chat.use_dpsk_v32_encoding)
# Case 3: Not DeepSeek V3.2 architecture -> should NOT use dpsk encoding
tokenizer_manager.tokenizer.chat_template = None
mock_hf_config.architectures = ["LlamaForCausalLM"]
serving_chat = OpenAIServingChat(tokenizer_manager, TemplateManager())
self.assertFalse(serving_chat.use_dpsk_v32_encoding)
def test_streaming_abort_yields_error(self):
"""Test that an abort finish reason during streaming correctly yields an error and stops."""
err_msg = "Aborted by scheduler"
err_code = HTTPStatus.INTERNAL_SERVER_ERROR
async def _mock_generate_abort():
yield {
"text": "Partial ",
"meta_info": {
"id": "chatcmpl-test",
"prompt_tokens": 10,
"completion_tokens": 2,
"cached_tokens": 0,
"finish_reason": {
"type": "abort",
"status_code": err_code,
"message": err_msg,
},
"output_token_logprobs": None,
"output_top_logprobs": None,
},
"index": 0,
}
self.tm.generate_request.return_value = _mock_generate_abort()
req = ChatCompletionRequest(
model="x",
messages=[{"role": "user", "content": "Hi?"}],
temperature=0.7,
max_tokens=100,
stream=True,
)
with patch(
"sglang.srt.entrypoints.openai.serving_chat.generate_chat_conv"
) as conv_mock:
# Create a mock conversation object
conv_ins = Mock()
conv_ins.get_prompt.return_value = "Test prompt"
conv_mock.return_value = conv_ins
adapted_request, _ = self.chat._convert_to_internal_request(
req, self.fastapi_request
)
async def run_stream():
chunks = []
try:
async for chunk in self.chat._generate_chat_stream(
adapted_request, req, self.fastapi_request
):
chunks.append(chunk)
except Exception as e:
print(f"Error during stream iteration: {e}")
return chunks
loop = get_or_create_event_loop()
chunks = loop.run_until_complete(run_stream())
error_chunk_data = None
for c in chunks:
if "error" in c:
error_chunk_data = json.loads(c[len("data: ") :])
break
self.assertIsNotNone(error_chunk_data, "Error chunk not found in stream")
self.assertEqual(error_chunk_data["error"]["message"], err_msg)
self.assertEqual(error_chunk_data["error"]["code"], err_code.value)
# Ensure the stream stops after the abort error
# The last chunk should be "data: [DONE]\n\n"
self.assertEqual(chunks[-1], "data: [DONE]\n\n")
# Check that there is an error chunk and a DONE chunk
self.assertEqual(len(chunks), 2)
self.assertIn("error", chunks[0])
if __name__ == "__main__":
unittest.main(verbosity=2)
|