test_qwen3coder_tool_parser.py 28.2 KB
Newer Older
1
2
3
4
5
6
7
8
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project

import json
from collections.abc import Generator

import pytest

9
10
11
12
13
14
15
from vllm.entrypoints.openai.protocol import (
    ChatCompletionRequest,
    ChatCompletionToolsParam,
    DeltaMessage,
    FunctionCall,
    ToolCall,
)
16
from vllm.tokenizers import TokenizerLike, get_tokenizer
17
from vllm.tokenizers.detokenizer_utils import detokenize_incrementally
18
19
20
21
from vllm.tool_parsers.qwen3coder_tool_parser import (
    Qwen3CoderToolParser,
)
from vllm.tool_parsers.qwen3xml_tool_parser import Qwen3XMLToolParser
22

23
MODEL = "Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8"
24
25
26
27
28
29
30
31
32
33
34
35


@pytest.fixture(scope="module")
def qwen3_tokenizer():
    return get_tokenizer(tokenizer_name=MODEL)


@pytest.fixture
def qwen3_tool_parser(qwen3_tokenizer):
    return Qwen3CoderToolParser(qwen3_tokenizer)


36
37
38
39
40
@pytest.fixture
def qwen3_xml_tool_parser(qwen3_tokenizer):
    return Qwen3XMLToolParser(qwen3_tokenizer)


41
@pytest.fixture(params=["xml"])
42
def qwen3_tool_parser_parametrized(qwen3_tool_parser, qwen3_xml_tool_parser, request):
43
44
45
46
47
48
49
    """Parameterized fixture that provides both parser types for testing"""
    if request.param == "original":
        return qwen3_tool_parser
    else:
        return qwen3_xml_tool_parser


50
51
52
@pytest.fixture
def sample_tools():
    return [
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
        ChatCompletionToolsParam(
            type="function",
            function={
                "name": "get_current_weather",
                "description": "Get the current weather",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "city": {"type": "string", "description": "The city name"},
                        "state": {"type": "string", "description": "The state code"},
                        "unit": {"type": "string", "enum": ["fahrenheit", "celsius"]},
                    },
                    "required": ["city", "state"],
                },
            },
        ),
        ChatCompletionToolsParam(
            type="function",
            function={
                "name": "calculate_area",
                "description": "Calculate area of a shape",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "shape": {"type": "string"},
                        "dimensions": {"type": "object"},
                        "precision": {"type": "integer"},
                    },
                },
            },
        ),
84
85
86
    ]


87
88
89
def assert_tool_calls(
    actual_tool_calls: list[ToolCall], expected_tool_calls: list[ToolCall]
):
90
91
    assert len(actual_tool_calls) == len(expected_tool_calls)

92
93
94
    for actual_tool_call, expected_tool_call in zip(
        actual_tool_calls, expected_tool_calls
    ):
95
96
        # Qwen3 parser doesn't generate IDs during extraction
        assert actual_tool_call.type == "function"
97
98
99
100
        assert actual_tool_call.function.name == expected_tool_call.function.name
        assert json.loads(actual_tool_call.function.arguments) == json.loads(
            expected_tool_call.function.arguments
        )
101
102
103


def stream_delta_message_generator(
104
    qwen3_tool_parser,
105
    qwen3_tokenizer: TokenizerLike,
106
    model_output: str,
107
    request: ChatCompletionRequest | None = None,
108
) -> Generator[DeltaMessage, None, None]:
109
    all_token_ids = qwen3_tokenizer.encode(model_output, add_special_tokens=False)
110
111
112
113
114
115
116
117

    previous_text = ""
    previous_tokens = None
    prefix_offset = 0
    read_offset = 0
    for i, delta_token in enumerate(all_token_ids):
        delta_token_ids = [delta_token]
        previous_token_ids = all_token_ids[:i]
118
119
120
121
122
123
124
125
126
127
128
129
130
        current_token_ids = all_token_ids[: i + 1]

        (new_tokens, delta_text, new_prefix_offset, new_read_offset) = (
            detokenize_incrementally(
                tokenizer=qwen3_tokenizer,
                all_input_ids=current_token_ids,
                prev_tokens=previous_tokens,
                prefix_offset=prefix_offset,
                read_offset=read_offset,
                skip_special_tokens=False,
                spaces_between_special_tokens=True,
            )
        )
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146

        current_text = previous_text + delta_text

        delta_message = qwen3_tool_parser.extract_tool_calls_streaming(
            previous_text,
            current_text,
            delta_text,
            previous_token_ids,
            current_token_ids,
            delta_token_ids,
            request=request,
        )
        if delta_message:
            yield delta_message

        previous_text = current_text
147
148
149
        previous_tokens = (
            previous_tokens + new_tokens if previous_tokens else new_tokens
        )
150
151
152
153
        prefix_offset = new_prefix_offset
        read_offset = new_read_offset


154
def test_extract_tool_calls_no_tools(qwen3_tool_parser_parametrized):
155
    model_output = "This is a test response without any tool calls"
156
    extracted_tool_calls = qwen3_tool_parser_parametrized.extract_tool_calls(
157
158
        model_output, request=None
    )  # type: ignore[arg-type]
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
    assert not extracted_tool_calls.tools_called
    assert extracted_tool_calls.tool_calls == []
    assert extracted_tool_calls.content == model_output


@pytest.mark.parametrize(
    ids=[
        "single_tool",
        "single_tool_with_content",
        "single_tool_multiline_param",
        "parallel_tools",
        "tool_with_typed_params",
    ],
    argnames=["model_output", "expected_tool_calls", "expected_content"],
    argvalues=[
174
175
        (
            """<tool_call>
176
177
178
179
180
181
182
183
184
185
186
<function=get_current_weather>
<parameter=city>
Dallas
</parameter>
<parameter=state>
TX
</parameter>
<parameter=unit>
fahrenheit
</parameter>
</function>
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
</tool_call>""",
            [
                ToolCall(
                    function=FunctionCall(
                        name="get_current_weather",
                        arguments=json.dumps(
                            {"city": "Dallas", "state": "TX", "unit": "fahrenheit"}
                        ),
                    )
                )
            ],
            None,
        ),
        (
            """Sure! Let me check the weather for you.<tool_call>
202
203
204
205
206
207
208
209
210
211
212
<function=get_current_weather>
<parameter=city>
Dallas
</parameter>
<parameter=state>
TX
</parameter>
<parameter=unit>
fahrenheit
</parameter>
</function>
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
</tool_call>""",
            [
                ToolCall(
                    function=FunctionCall(
                        name="get_current_weather",
                        arguments=json.dumps(
                            {"city": "Dallas", "state": "TX", "unit": "fahrenheit"}
                        ),
                    )
                )
            ],
            "Sure! Let me check the weather for you.",
        ),
        (
            """<tool_call>
228
229
230
231
232
233
234
235
236
237
238
239
<function=calculate_area>
<parameter=shape>
rectangle
</parameter>
<parameter=dimensions>
{"width": 10, 
 "height": 20}
</parameter>
<parameter=precision>
2
</parameter>
</function>
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
</tool_call>""",
            [
                ToolCall(
                    function=FunctionCall(
                        name="calculate_area",
                        arguments=json.dumps(
                            {
                                "shape": "rectangle",
                                "dimensions": {"width": 10, "height": 20},
                                "precision": 2,
                            }
                        ),
                    )
                )
            ],
            None,
        ),
        (
            """<tool_call>
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
<function=get_current_weather>
<parameter=city>
Dallas
</parameter>
<parameter=state>
TX
</parameter>
<parameter=unit>
fahrenheit
</parameter>
</function>
</tool_call>
<tool_call>
<function=get_current_weather>
<parameter=city>
Orlando
</parameter>
<parameter=state>
FL
</parameter>
<parameter=unit>
fahrenheit
</parameter>
</function>
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
</tool_call>""",
            [
                ToolCall(
                    function=FunctionCall(
                        name="get_current_weather",
                        arguments=json.dumps(
                            {"city": "Dallas", "state": "TX", "unit": "fahrenheit"}
                        ),
                    )
                ),
                ToolCall(
                    function=FunctionCall(
                        name="get_current_weather",
                        arguments=json.dumps(
                            {"city": "Orlando", "state": "FL", "unit": "fahrenheit"}
                        ),
                    )
                ),
            ],
            None,
        ),
        (
            """Let me calculate that area for you.<tool_call>
306
307
308
309
310
311
312
313
314
315
316
<function=calculate_area>
<parameter=shape>
circle
</parameter>
<parameter=dimensions>
{"radius": 15.5}
</parameter>
<parameter=precision>
3
</parameter>
</function>
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
</tool_call>""",
            [
                ToolCall(
                    function=FunctionCall(
                        name="calculate_area",
                        arguments=json.dumps(
                            {
                                "shape": "circle",
                                "dimensions": {"radius": 15.5},
                                "precision": 3,
                            }
                        ),
                    )
                )
            ],
            "Let me calculate that area for you.",
        ),
334
335
    ],
)
336
337
338
339
340
341
342
343
def test_extract_tool_calls(
    qwen3_tool_parser_parametrized,
    sample_tools,
    model_output,
    expected_tool_calls,
    expected_content,
):
    request = ChatCompletionRequest(model=MODEL, messages=[], tools=sample_tools)
344
    extracted_tool_calls = qwen3_tool_parser_parametrized.extract_tool_calls(
345
346
        model_output, request=request
    )
347
348
349
350
351
352
353
    assert extracted_tool_calls.tools_called

    assert_tool_calls(extracted_tool_calls.tool_calls, expected_tool_calls)

    assert extracted_tool_calls.content == expected_content


354
355
356
def test_extract_tool_calls_fallback_no_tags(
    qwen3_tool_parser_parametrized, sample_tools
):
357
    """Test fallback parsing when XML tags are missing"""
358
    model_output = """<function=get_current_weather>
359
360
361
362
363
364
<parameter=city>
Dallas
</parameter>
<parameter=state>
TX
</parameter>
365
</function>"""
366

367
    request = ChatCompletionRequest(model=MODEL, messages=[], tools=sample_tools)
368
    extracted_tool_calls = qwen3_tool_parser_parametrized.extract_tool_calls(
369
370
        model_output, request=request
    )
371
372
373

    assert extracted_tool_calls.tools_called
    assert len(extracted_tool_calls.tool_calls) == 1
374
    assert extracted_tool_calls.tool_calls[0].function.name == "get_current_weather"
375
376


377
def test_extract_tool_calls_type_conversion(qwen3_tool_parser_parametrized):
378
379
    """Test parameter type conversion based on tool schema"""
    tools = [
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
        ChatCompletionToolsParam(
            type="function",
            function={
                "name": "test_types",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "int_param": {"type": "integer"},
                        "float_param": {"type": "float"},
                        "bool_param": {"type": "boolean"},
                        "str_param": {"type": "string"},
                        "obj_param": {"type": "object"},
                    },
                },
            },
        )
396
397
    ]

398
    model_output = """<tool_call>
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
<function=test_types>
<parameter=int_param>
42
</parameter>
<parameter=float_param>
3.14
</parameter>
<parameter=bool_param>
true
</parameter>
<parameter=str_param>
hello world
</parameter>
<parameter=obj_param>
{"key": "value"}
</parameter>
</function>
416
</tool_call>"""
417
418

    request = ChatCompletionRequest(model=MODEL, messages=[], tools=tools)
419
    extracted_tool_calls = qwen3_tool_parser_parametrized.extract_tool_calls(
420
421
        model_output, request=request
    )
422
423
424
425
426
427
428
429
430
431
432
433
434
435

    args = json.loads(extracted_tool_calls.tool_calls[0].function.arguments)
    assert args["int_param"] == 42
    assert args["float_param"] == 3.14
    assert args["bool_param"] is True
    assert args["str_param"] == "hello world"
    assert args["obj_param"] == {"key": "value"}


@pytest.mark.parametrize(
    ids=[
        "no_tools",
        "single_tool",
        "single_tool_with_content",
436
        "single_tool_multiline_param",
437
        "parallel_tools",
438
        "tool_with_typed_params",  # Added this test case
439
440
441
442
    ],
    argnames=["model_output", "expected_tool_calls", "expected_content"],
    argvalues=[
        ("This is a test without tools", [], "This is a test without tools"),
443
444
        (
            """<tool_call>
445
446
447
448
449
450
451
452
453
454
455
<function=get_current_weather>
<parameter=city>
Dallas
</parameter>
<parameter=state>
TX
</parameter>
<parameter=unit>
fahrenheit
</parameter>
</function>
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
</tool_call>""",
            [
                ToolCall(
                    function=FunctionCall(
                        name="get_current_weather",
                        arguments=json.dumps(
                            {"city": "Dallas", "state": "TX", "unit": "fahrenheit"}
                        ),
                    )
                )
            ],
            None,
        ),
        (
            """Sure! Let me check the weather for you.<tool_call>
471
472
473
474
475
476
477
478
479
480
481
<function=get_current_weather>
<parameter=city>
Dallas
</parameter>
<parameter=state>
TX
</parameter>
<parameter=unit>
fahrenheit
</parameter>
</function>
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
</tool_call>""",
            [
                ToolCall(
                    function=FunctionCall(
                        name="get_current_weather",
                        arguments=json.dumps(
                            {"city": "Dallas", "state": "TX", "unit": "fahrenheit"}
                        ),
                    )
                )
            ],
            "Sure! Let me check the weather for you.",
        ),
        (
            """<tool_call>
497
498
499
500
501
502
503
504
505
506
507
508
<function=calculate_area>
<parameter=shape>
rectangle
</parameter>
<parameter=dimensions>
{"width": 10, 
 "height": 20}
</parameter>
<parameter=precision>
2
</parameter>
</function>
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
</tool_call>""",
            [
                ToolCall(
                    function=FunctionCall(
                        name="calculate_area",
                        arguments=json.dumps(
                            {
                                "shape": "rectangle",
                                "dimensions": {"width": 10, "height": 20},
                                "precision": 2,
                            }
                        ),
                    )
                )
            ],
            None,
        ),
        (
            """<tool_call>
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
<function=get_current_weather>
<parameter=city>
Dallas
</parameter>
<parameter=state>
TX
</parameter>
<parameter=unit>
fahrenheit
</parameter>
</function>
</tool_call>
<tool_call>
<function=get_current_weather>
<parameter=city>
Orlando
</parameter>
<parameter=state>
FL
</parameter>
<parameter=unit>
celsius
</parameter>
</function>
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
</tool_call>""",
            [
                ToolCall(
                    function=FunctionCall(
                        name="get_current_weather",
                        arguments=json.dumps(
                            {"city": "Dallas", "state": "TX", "unit": "fahrenheit"}
                        ),
                    )
                ),
                ToolCall(
                    function=FunctionCall(
                        name="get_current_weather",
                        arguments=json.dumps(
                            {"city": "Orlando", "state": "FL", "unit": "celsius"}
                        ),
                    )
                ),
            ],
            None,
        ),
573
        # Added tool_with_typed_params test case
574
575
        (
            """Let me calculate that area for you.<tool_call>
576
577
578
579
580
581
582
583
584
585
586
<function=calculate_area>
<parameter=shape>
circle
</parameter>
<parameter=dimensions>
{"radius": 15.5}
</parameter>
<parameter=precision>
3
</parameter>
</function>
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
</tool_call>""",
            [
                ToolCall(
                    function=FunctionCall(
                        name="calculate_area",
                        arguments=json.dumps(
                            {
                                "shape": "circle",
                                "dimensions": {"radius": 15.5},
                                "precision": 3,
                            }
                        ),
                    )
                )
            ],
            "Let me calculate that area for you.",
        ),
604
605
    ],
)
606
607
608
609
610
611
612
613
def test_extract_tool_calls_streaming(
    qwen3_tool_parser_parametrized,
    qwen3_tokenizer,
    sample_tools,
    model_output,
    expected_tool_calls,
    expected_content,
):
614
    """Test incremental streaming behavior including typed parameters"""
615
    request = ChatCompletionRequest(model=MODEL, messages=[], tools=sample_tools)
616

617
    other_content = ""
618
619
620
    tool_states = {}  # Track state per tool index

    for delta_message in stream_delta_message_generator(
621
622
        qwen3_tool_parser_parametrized, qwen3_tokenizer, model_output, request
    ):
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
        # role should never be streamed from tool parser
        assert not delta_message.role

        if delta_message.content:
            other_content += delta_message.content

        if delta_message.tool_calls:
            for tool_call in delta_message.tool_calls:
                idx = tool_call.index

                # Initialize state for new tool
                if idx not in tool_states:
                    tool_states[idx] = {
                        "id": None,
                        "name": None,
                        "arguments": "",
639
                        "type": None,
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
                    }

                # First chunk should have id, name, and type
                if tool_call.id:
                    tool_states[idx]["id"] = tool_call.id

                if tool_call.type:
                    assert tool_call.type == "function"
                    tool_states[idx]["type"] = tool_call.type

                if tool_call.function:
                    if tool_call.function.name:
                        # Should only be set once
                        assert tool_states[idx]["name"] is None
                        tool_states[idx]["name"] = tool_call.function.name

                    if tool_call.function.arguments is not None:
                        # Accumulate arguments incrementally
658
                        tool_states[idx]["arguments"] += tool_call.function.arguments
659
660

    # Verify final content
661
    assert other_content == (expected_content or "")  # Handle None case
662
663
664

    # Verify we got all expected tool calls
    assert len(tool_states) == len(expected_tool_calls)
665
666
667
    assert len(qwen3_tool_parser_parametrized.prev_tool_call_arr) == len(
        expected_tool_calls
    )
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683

    # Verify each tool call
    for idx, expected_tool in enumerate(expected_tool_calls):
        state = tool_states[idx]
        assert state["id"] is not None
        assert state["type"] == "function"
        assert state["name"] == expected_tool.function.name

        # Parse accumulated arguments
        arguments_str = state["arguments"]
        assert arguments_str is not None
        actual_args = json.loads(arguments_str)
        expected_args = json.loads(expected_tool.function.arguments)
        assert actual_args == expected_args


684
def test_extract_tool_calls_missing_closing_parameter_tag(
685
686
    qwen3_tool_parser_parametrized, sample_tools
):
687
688
    """Test handling of missing closing </parameter> tag"""
    # Using get_current_weather from sample_tools but with malformed XML
689
    model_output = """Let me check the weather for you:
690
691
692
693
694
695
696
697
698
699
700
<tool_call>
<function=get_current_weather>
<parameter=city>
Dallas
<parameter=state>
TX
</parameter>
<parameter=unit>
fahrenheit
</parameter>
</function>
701
</tool_call>"""
702

703
    request = ChatCompletionRequest(model=MODEL, messages=[], tools=sample_tools)
704
    extracted_tool_calls = qwen3_tool_parser_parametrized.extract_tool_calls(
705
706
        model_output, request=request
    )
707
708
709
710
711
712

    # The parser should handle the malformed XML gracefully
    assert extracted_tool_calls.tools_called
    assert len(extracted_tool_calls.tool_calls) == 1

    # Verify the function name is correct
713
    assert extracted_tool_calls.tool_calls[0].function.name == "get_current_weather"
714
715
716
717
718
719
720
721
722
723
724
725
726

    # Verify the arguments are parsed despite the missing closing tag
    args = json.loads(extracted_tool_calls.tool_calls[0].function.arguments)
    assert "city" in args
    assert args["city"] == "Dallas"
    assert args["state"] == "TX"
    assert args["unit"] == "fahrenheit"

    # Check that content before the tool call is preserved
    assert "Let me check the weather for you:" in extracted_tool_calls.content


def test_extract_tool_calls_streaming_missing_closing_tag(
727
728
    qwen3_tool_parser_parametrized, qwen3_tokenizer, sample_tools
):
729
730
    """Test streaming with missing closing </parameter> tag"""
    # Using get_current_weather from sample_tools but with malformed XML
731
    model_output = """Let me check the weather for you:
732
733
734
735
736
737
738
739
740
741
742
<tool_call>
<function=get_current_weather>
<parameter=city>
Dallas
<parameter=state>
TX
</parameter>
<parameter=unit>
fahrenheit
</parameter>
</function>
743
</tool_call>"""
744

745
    request = ChatCompletionRequest(model=MODEL, messages=[], tools=sample_tools)
746

747
    other_content = ""
748
749
750
    tool_states = {}

    for delta_message in stream_delta_message_generator(
751
752
        qwen3_tool_parser_parametrized, qwen3_tokenizer, model_output, request
    ):
753
754
755
756
757
758
759
760
761
762
763
764
        if delta_message.content:
            other_content += delta_message.content

        if delta_message.tool_calls:
            for tool_call in delta_message.tool_calls:
                idx = tool_call.index

                if idx not in tool_states:
                    tool_states[idx] = {
                        "id": None,
                        "name": None,
                        "arguments": "",
765
                        "type": None,
766
767
768
769
770
771
772
773
774
775
776
777
778
779
                    }

                if tool_call.id:
                    tool_states[idx]["id"] = tool_call.id

                if tool_call.type:
                    assert tool_call.type == "function"
                    tool_states[idx]["type"] = tool_call.type

                if tool_call.function:
                    if tool_call.function.name:
                        tool_states[idx]["name"] = tool_call.function.name

                    if tool_call.function.arguments is not None:
780
                        tool_states[idx]["arguments"] += tool_call.function.arguments
781
782
783
784
785

    # Verify content was streamed
    assert "Let me check the weather for you:" in other_content
    # Verify we got the tool call
    assert len(tool_states) == 1
786
787
    assert len(qwen3_tool_parser_parametrized.prev_tool_call_arr) == 1

788
789
790
791
792
793
794
795
796
797
798
799
800
    state = tool_states[0]
    assert state["id"] is not None
    assert state["type"] == "function"
    assert state["name"] == "get_current_weather"

    # Verify arguments were parsed correctly despite missing closing tag
    assert state["arguments"] is not None
    args = json.loads(state["arguments"])
    assert args["city"] == "Dallas"
    assert args["state"] == "TX"
    assert args["unit"] == "fahrenheit"


801
def test_extract_tool_calls_streaming_incremental(
802
803
    qwen3_tool_parser_parametrized, qwen3_tokenizer, sample_tools
):
804
    """Test that streaming is truly incremental"""
805
    model_output = """I'll check the weather.<tool_call>
806
807
808
809
810
811
812
813
<function=get_current_weather>
<parameter=city>
Dallas
</parameter>
<parameter=state>
TX
</parameter>
</function>
814
</tool_call>"""
815

816
    request = ChatCompletionRequest(model=MODEL, messages=[], tools=sample_tools)
817
818
819

    chunks = []
    for delta_message in stream_delta_message_generator(
820
821
        qwen3_tool_parser_parametrized, qwen3_tokenizer, model_output, request
    ):
822
823
824
825
826
827
828
829
830
831
832
833
834
835
        chunks.append(delta_message)

    # Should have multiple chunks
    assert len(chunks) > 3

    # First chunk(s) should be content
    assert chunks[0].content is not None
    assert chunks[0].tool_calls is None or chunks[0].tool_calls == []

    # Should have a chunk with tool header (id, name, type)
    header_found = False
    for chunk in chunks:
        if chunk.tool_calls and chunk.tool_calls[0].id:
            header_found = True
836
            assert chunk.tool_calls[0].function.name == "get_current_weather"
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
            assert chunk.tool_calls[0].type == "function"
            # Empty initially
            assert chunk.tool_calls[0].function.arguments == ""
            break
    assert header_found

    # Should have chunks with incremental arguments
    arg_chunks = []
    for chunk in chunks:
        if chunk.tool_calls and chunk.tool_calls[0].function.arguments:
            arg_chunks.append(chunk.tool_calls[0].function.arguments)

    # Arguments should be streamed incrementally
    assert len(arg_chunks) > 1

    # Concatenated arguments should form valid JSON
    full_args = "".join(arg_chunks)
    parsed_args = json.loads(full_args)
    assert parsed_args["city"] == "Dallas"
    assert parsed_args["state"] == "TX"
857
858
859


def test_extract_tool_calls_complex_type_with_single_quote(
860
861
    qwen3_tool_parser_parametrized,
):
862
863
    """Test parameter type conversion based on tool schema"""
    tools = [
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
        ChatCompletionToolsParam(
            type="function",
            function={
                "name": "test_types",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "int_param": {"type": "integer"},
                        "float_param": {"type": "float"},
                        "bool_param": {"type": "boolean"},
                        "str_param": {"type": "string"},
                        "obj_param": {"type": "object"},
                    },
                },
            },
        )
880
881
    ]

882
    model_output = """<tool_call>
883
884
885
886
887
<function=test_types>
<parameter=obj_param>
{'key': 'value'}
</parameter>
</function>
888
</tool_call>"""
889
890
891

    request = ChatCompletionRequest(model=MODEL, messages=[], tools=tools)
    extracted_tool_calls = qwen3_tool_parser_parametrized.extract_tool_calls(
892
893
        model_output, request=request
    )
894
895
896

    args = json.loads(extracted_tool_calls.tool_calls[0].function.arguments)
    assert args["obj_param"] == {"key": "value"}
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976


def test_extract_tool_calls_streaming_missing_opening_tag(
    qwen3_tool_parser_parametrized, qwen3_tokenizer, sample_tools
):
    """Test streaming with missing opening <tool_call> tag

    This tests that the streaming parser correctly handles
    tool calls that start directly with <function=...>
    """
    model_output = """I'll check the weather for you.

<function=get_current_weather>
<parameter=city>
Dallas
</parameter>
<parameter=state>
TX
</parameter>
<parameter=unit>
fahrenheit
</parameter>
</function>
</tool_call>"""

    request = ChatCompletionRequest(model=MODEL, messages=[], tools=sample_tools)

    other_content = ""
    tool_states = {}

    for delta_message in stream_delta_message_generator(
        qwen3_tool_parser_parametrized, qwen3_tokenizer, model_output, request
    ):
        if delta_message.content:
            other_content += delta_message.content

        if delta_message.tool_calls:
            for tool_call in delta_message.tool_calls:
                idx = tool_call.index

                if idx not in tool_states:
                    tool_states[idx] = {
                        "id": None,
                        "name": None,
                        "arguments": "",
                        "type": None,
                    }

                if tool_call.id:
                    tool_states[idx]["id"] = tool_call.id

                if tool_call.type:
                    assert tool_call.type == "function"
                    tool_states[idx]["type"] = tool_call.type

                if tool_call.function:
                    if tool_call.function.name:
                        tool_states[idx]["name"] = tool_call.function.name

                    if tool_call.function.arguments is not None:
                        tool_states[idx]["arguments"] += tool_call.function.arguments

    # Verify content was streamed
    assert "I'll check the weather for you." in other_content

    # Verify we got the tool call
    assert len(tool_states) == 1
    assert len(qwen3_tool_parser_parametrized.prev_tool_call_arr) == 1

    state = tool_states[0]
    assert state["id"] is not None
    assert state["type"] == "function"
    assert state["name"] == "get_current_weather"

    # Verify arguments were parsed correctly despite missing opening tag
    assert state["arguments"] is not None
    args = json.loads(state["arguments"])
    assert args["city"] == "Dallas"
    assert args["state"] == "TX"
    assert args["unit"] == "fahrenheit"