test_qwen3coder_tool_parser.py 40.1 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
from vllm.entrypoints.openai.chat_completion.protocol import (
10
11
    ChatCompletionRequest,
    ChatCompletionToolsParam,
12
13
)
from vllm.entrypoints.openai.engine.protocol import (
14
15
16
17
    DeltaMessage,
    FunctionCall,
    ToolCall,
)
18
from vllm.tokenizers import TokenizerLike, get_tokenizer
19
from vllm.tokenizers.detokenizer_utils import detokenize_incrementally
20
21
22
23
from vllm.tool_parsers.qwen3coder_tool_parser import (
    Qwen3CoderToolParser,
)
from vllm.tool_parsers.qwen3xml_tool_parser import Qwen3XMLToolParser
24

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


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


@pytest.fixture
34
35
def qwen3_tool_parser(qwen3_tokenizer, sample_tools):
    return Qwen3CoderToolParser(qwen3_tokenizer, tools=sample_tools)
36
37


38
@pytest.fixture
39
40
def qwen3_xml_tool_parser(qwen3_tokenizer, sample_tools):
    return Qwen3XMLToolParser(qwen3_tokenizer, tools=sample_tools)
41
42


43
@pytest.fixture(params=["xml"])
44
def qwen3_tool_parser_parametrized(qwen3_tool_parser, qwen3_xml_tool_parser, request):
45
46
47
48
49
50
51
    """Parameterized fixture that provides both parser types for testing"""
    if request.param == "original":
        return qwen3_tool_parser
    else:
        return qwen3_xml_tool_parser


52
53
54
@pytest.fixture
def sample_tools():
    return [
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
        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"},
                    },
                },
            },
        ),
86
87
88
    ]


89
90
91
def assert_tool_calls(
    actual_tool_calls: list[ToolCall], expected_tool_calls: list[ToolCall]
):
92
93
    assert len(actual_tool_calls) == len(expected_tool_calls)

94
95
96
    for actual_tool_call, expected_tool_call in zip(
        actual_tool_calls, expected_tool_calls
    ):
97
98
        # Qwen3 parser doesn't generate IDs during extraction
        assert actual_tool_call.type == "function"
99
100
101
102
        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
        )
103
104
105


def stream_delta_message_generator(
106
    qwen3_tool_parser,
107
    qwen3_tokenizer: TokenizerLike,
108
    model_output: str,
109
    request: ChatCompletionRequest | None = None,
110
) -> Generator[DeltaMessage, None, None]:
111
    all_token_ids = qwen3_tokenizer.encode(model_output, add_special_tokens=False)
112
113
114
115
116
117
118
119

    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]
120
121
122
123
124
125
126
127
128
129
130
131
132
        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,
            )
        )
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148

        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
149
150
151
        previous_tokens = (
            previous_tokens + new_tokens if previous_tokens else new_tokens
        )
152
153
154
155
        prefix_offset = new_prefix_offset
        read_offset = new_read_offset


156
def test_extract_tool_calls_no_tools(qwen3_tool_parser_parametrized):
157
    model_output = "This is a test response without any tool calls"
158
    extracted_tool_calls = qwen3_tool_parser_parametrized.extract_tool_calls(
159
160
        model_output, request=None
    )  # type: ignore[arg-type]
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
    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=[
176
177
        (
            """<tool_call>
178
179
180
181
182
183
184
185
186
187
188
<function=get_current_weather>
<parameter=city>
Dallas
</parameter>
<parameter=state>
TX
</parameter>
<parameter=unit>
fahrenheit
</parameter>
</function>
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
</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>
204
205
206
207
208
209
210
211
212
213
214
<function=get_current_weather>
<parameter=city>
Dallas
</parameter>
<parameter=state>
TX
</parameter>
<parameter=unit>
fahrenheit
</parameter>
</function>
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
</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>
230
231
232
233
234
235
236
237
238
239
240
241
<function=calculate_area>
<parameter=shape>
rectangle
</parameter>
<parameter=dimensions>
{"width": 10, 
 "height": 20}
</parameter>
<parameter=precision>
2
</parameter>
</function>
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
</tool_call>""",
            [
                ToolCall(
                    function=FunctionCall(
                        name="calculate_area",
                        arguments=json.dumps(
                            {
                                "shape": "rectangle",
                                "dimensions": {"width": 10, "height": 20},
                                "precision": 2,
                            }
                        ),
                    )
                )
            ],
            None,
        ),
        (
            """<tool_call>
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
<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>
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
</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>
308
309
310
311
312
313
314
315
316
317
318
<function=calculate_area>
<parameter=shape>
circle
</parameter>
<parameter=dimensions>
{"radius": 15.5}
</parameter>
<parameter=precision>
3
</parameter>
</function>
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
</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.",
        ),
336
337
    ],
)
338
339
340
341
342
343
344
345
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)
346
    extracted_tool_calls = qwen3_tool_parser_parametrized.extract_tool_calls(
347
348
        model_output, request=request
    )
349
350
351
352
353
354
355
    assert extracted_tool_calls.tools_called

    assert_tool_calls(extracted_tool_calls.tool_calls, expected_tool_calls)

    assert extracted_tool_calls.content == expected_content


356
357
358
def test_extract_tool_calls_fallback_no_tags(
    qwen3_tool_parser_parametrized, sample_tools
):
359
    """Test fallback parsing when XML tags are missing"""
360
    model_output = """<function=get_current_weather>
361
362
363
364
365
366
<parameter=city>
Dallas
</parameter>
<parameter=state>
TX
</parameter>
367
</function>"""
368

369
    request = ChatCompletionRequest(model=MODEL, messages=[], tools=sample_tools)
370
    extracted_tool_calls = qwen3_tool_parser_parametrized.extract_tool_calls(
371
372
        model_output, request=request
    )
373
374
375

    assert extracted_tool_calls.tools_called
    assert len(extracted_tool_calls.tool_calls) == 1
376
    assert extracted_tool_calls.tool_calls[0].function.name == "get_current_weather"
377
378


379
def test_extract_tool_calls_type_conversion(qwen3_tokenizer):
380
381
    """Test parameter type conversion based on tool schema"""
    tools = [
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
        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"},
                    },
                },
            },
        )
398
399
    ]

400
    model_output = """<tool_call>
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
<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>
418
</tool_call>"""
419

420
    parser = Qwen3XMLToolParser(qwen3_tokenizer, tools=tools)
421
    request = ChatCompletionRequest(model=MODEL, messages=[], tools=tools)
422
    extracted_tool_calls = parser.extract_tool_calls(model_output, request=request)
423
424
425
426
427
428
429
430
431

    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"}


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
def test_extract_tool_calls_anyof_type_conversion(qwen3_tool_parser):
    """Test type conversion for anyOf/oneOf nullable schemas (Pydantic v2).

    Pydantic v2 emits anyOf for Optional[T] fields, e.g.:
        Optional[int] -> {"anyOf": [{"type": "integer"}, {"type": "null"}]}
    The parser must extract the non-null type and apply the correct
    conversion (int(), float(), etc.) instead of returning a raw string.
    """
    tools = [
        ChatCompletionToolsParam(
            type="function",
            function={
                "name": "test_anyof",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "anyof_int": {
                            "anyOf": [
                                {"type": "integer"},
                                {"type": "null"},
                            ],
                            "default": 5,
                        },
                        "anyof_str": {
                            "anyOf": [
                                {"type": "string"},
                                {"type": "null"},
                            ],
                        },
                        "anyof_array": {
                            "anyOf": [
                                {"type": "array", "items": {"type": "string"}},
                                {"type": "null"},
                            ],
                        },
                        "anyof_obj": {
                            "anyOf": [
                                {"type": "object"},
                                {"type": "null"},
                            ],
                        },
                        "type_as_array": {
                            "type": ["integer", "null"],
                        },
                        "multi_non_null": {
                            "anyOf": [
                                {"type": "string"},
                                {"type": "integer"},
                                {"type": "null"},
                            ],
                        },
                        "ref_param": {
                            "$ref": "#/$defs/ToolInput",
                        },
                    },
                },
            },
        )
    ]

    model_output = """<tool_call>
<function=test_anyof>
<parameter=anyof_int>
5
</parameter>
<parameter=anyof_str>
hello
</parameter>
<parameter=anyof_array>
["a", "b", "c"]
</parameter>
<parameter=anyof_obj>
{"key": "value"}
</parameter>
<parameter=type_as_array>
42
</parameter>
<parameter=multi_non_null>
some text
</parameter>
<parameter=ref_param>
{"city": "Paris"}
</parameter>
</function>
</tool_call>"""

    request = ChatCompletionRequest(model=MODEL, messages=[], tools=tools)
    extracted = qwen3_tool_parser.extract_tool_calls(model_output, request=request)

    args = json.loads(extracted.tool_calls[0].function.arguments)
    assert args["anyof_int"] == 5
    assert isinstance(args["anyof_int"], int)
    assert args["anyof_str"] == "hello"
    assert isinstance(args["anyof_str"], str)
    assert args["anyof_array"] == ["a", "b", "c"]
    assert isinstance(args["anyof_array"], list)
    assert args["anyof_obj"] == {"key": "value"}
    assert isinstance(args["anyof_obj"], dict)
    assert args["type_as_array"] == 42
    assert isinstance(args["type_as_array"], int)
    # Multi non-null: anyOf[string, integer, null] → first non-null is string
    assert args["multi_non_null"] == "some text"
    assert isinstance(args["multi_non_null"], str)
    # $ref: treated as object, parsed via json.loads
    assert args["ref_param"] == {"city": "Paris"}
    assert isinstance(args["ref_param"], dict)


def test_extract_tool_calls_anyof_type_conversion_streaming(
    qwen3_tool_parser, qwen3_tokenizer
):
    """Test streaming e2e for anyOf/oneOf nullable schemas (Pydantic v2).

    Verifies that the full streaming pipeline — tokenize, incrementally
    decode, extract_tool_calls_streaming — correctly resolves types from
    anyOf schemas and produces valid JSON with properly typed values.
    """
    tools = [
        ChatCompletionToolsParam(
            type="function",
            function={
                "name": "search_web",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "query": {
                            "anyOf": [
                                {"type": "string"},
                                {"type": "null"},
                            ],
                        },
                        "count": {
                            "anyOf": [
                                {"type": "integer"},
                                {"type": "null"},
                            ],
                            "default": 5,
                        },
                        "verbose": {
                            "anyOf": [
                                {"type": "boolean"},
                                {"type": "null"},
                            ],
                        },
                        "filters": {
                            "$ref": "#/$defs/SearchFilters",
                        },
                    },
                },
            },
        )
    ]

    model_output = """<tool_call>
<function=search_web>
<parameter=query>
vllm tool parser
</parameter>
<parameter=count>
10
</parameter>
<parameter=verbose>
true
</parameter>
<parameter=filters>
{"lang": "en", "year": 2025}
</parameter>
</function>
</tool_call>"""

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

    tool_states = {}
    for delta_message in stream_delta_message_generator(
        qwen3_tool_parser, qwen3_tokenizer, model_output, request
    ):
        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] = {"name": None, "arguments": ""}
                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

    assert len(tool_states) == 1
    assert tool_states[0]["name"] == "search_web"
    assert tool_states[0]["arguments"] is not None
    args = json.loads(tool_states[0]["arguments"])
    assert args["query"] == "vllm tool parser"
    assert isinstance(args["query"], str)
    assert args["count"] == 10
    assert isinstance(args["count"], int)
    assert args["verbose"] is True
    assert isinstance(args["verbose"], bool)
    # $ref: treated as object, parsed via json.loads
    assert args["filters"] == {"lang": "en", "year": 2025}
    assert isinstance(args["filters"], dict)


634
635
636
637
638
@pytest.mark.parametrize(
    ids=[
        "no_tools",
        "single_tool",
        "single_tool_with_content",
639
        "single_tool_multiline_param",
640
        "parallel_tools",
641
        "tool_with_typed_params",  # Added this test case
642
643
644
645
    ],
    argnames=["model_output", "expected_tool_calls", "expected_content"],
    argvalues=[
        ("This is a test without tools", [], "This is a test without tools"),
646
647
        (
            """<tool_call>
648
649
650
651
652
653
654
655
656
657
658
<function=get_current_weather>
<parameter=city>
Dallas
</parameter>
<parameter=state>
TX
</parameter>
<parameter=unit>
fahrenheit
</parameter>
</function>
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
</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>
674
675
676
677
678
679
680
681
682
683
684
<function=get_current_weather>
<parameter=city>
Dallas
</parameter>
<parameter=state>
TX
</parameter>
<parameter=unit>
fahrenheit
</parameter>
</function>
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
</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>
700
701
702
703
704
705
706
707
708
709
710
711
<function=calculate_area>
<parameter=shape>
rectangle
</parameter>
<parameter=dimensions>
{"width": 10, 
 "height": 20}
</parameter>
<parameter=precision>
2
</parameter>
</function>
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
</tool_call>""",
            [
                ToolCall(
                    function=FunctionCall(
                        name="calculate_area",
                        arguments=json.dumps(
                            {
                                "shape": "rectangle",
                                "dimensions": {"width": 10, "height": 20},
                                "precision": 2,
                            }
                        ),
                    )
                )
            ],
            None,
        ),
        (
            """<tool_call>
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
<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>
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
</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,
        ),
776
        # Added tool_with_typed_params test case
777
778
        (
            """Let me calculate that area for you.<tool_call>
779
780
781
782
783
784
785
786
787
788
789
<function=calculate_area>
<parameter=shape>
circle
</parameter>
<parameter=dimensions>
{"radius": 15.5}
</parameter>
<parameter=precision>
3
</parameter>
</function>
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
</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.",
        ),
807
808
    ],
)
809
810
811
812
813
814
815
816
def test_extract_tool_calls_streaming(
    qwen3_tool_parser_parametrized,
    qwen3_tokenizer,
    sample_tools,
    model_output,
    expected_tool_calls,
    expected_content,
):
817
    """Test incremental streaming behavior including typed parameters"""
818
    request = ChatCompletionRequest(model=MODEL, messages=[], tools=sample_tools)
819

820
    other_content = ""
821
822
823
    tool_states = {}  # Track state per tool index

    for delta_message in stream_delta_message_generator(
824
825
        qwen3_tool_parser_parametrized, qwen3_tokenizer, model_output, request
    ):
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
        # 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": "",
842
                        "type": None,
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
                    }

                # 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
861
                        tool_states[idx]["arguments"] += tool_call.function.arguments
862
863

    # Verify final content
864
    assert other_content == (expected_content or "")  # Handle None case
865
866
867

    # Verify we got all expected tool calls
    assert len(tool_states) == len(expected_tool_calls)
868
869
870
    assert len(qwen3_tool_parser_parametrized.prev_tool_call_arr) == len(
        expected_tool_calls
    )
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886

    # 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


887
def test_extract_tool_calls_missing_closing_parameter_tag(
888
889
    qwen3_tool_parser_parametrized, sample_tools
):
890
891
    """Test handling of missing closing </parameter> tag"""
    # Using get_current_weather from sample_tools but with malformed XML
892
    model_output = """Let me check the weather for you:
893
894
895
896
897
898
899
900
901
902
903
<tool_call>
<function=get_current_weather>
<parameter=city>
Dallas
<parameter=state>
TX
</parameter>
<parameter=unit>
fahrenheit
</parameter>
</function>
904
</tool_call>"""
905

906
    request = ChatCompletionRequest(model=MODEL, messages=[], tools=sample_tools)
907
    extracted_tool_calls = qwen3_tool_parser_parametrized.extract_tool_calls(
908
909
        model_output, request=request
    )
910
911
912
913
914
915

    # 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
916
    assert extracted_tool_calls.tool_calls[0].function.name == "get_current_weather"
917
918
919
920
921
922
923
924
925
926
927
928
929

    # 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(
930
931
    qwen3_tool_parser_parametrized, qwen3_tokenizer, sample_tools
):
932
933
    """Test streaming with missing closing </parameter> tag"""
    # Using get_current_weather from sample_tools but with malformed XML
934
    model_output = """Let me check the weather for you:
935
936
937
938
939
940
941
942
943
944
945
<tool_call>
<function=get_current_weather>
<parameter=city>
Dallas
<parameter=state>
TX
</parameter>
<parameter=unit>
fahrenheit
</parameter>
</function>
946
</tool_call>"""
947

948
    request = ChatCompletionRequest(model=MODEL, messages=[], tools=sample_tools)
949

950
    other_content = ""
951
952
953
    tool_states = {}

    for delta_message in stream_delta_message_generator(
954
955
        qwen3_tool_parser_parametrized, qwen3_tokenizer, model_output, request
    ):
956
957
958
959
960
961
962
963
964
965
966
967
        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": "",
968
                        "type": None,
969
970
971
972
973
974
975
976
977
978
979
980
981
982
                    }

                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:
983
                        tool_states[idx]["arguments"] += tool_call.function.arguments
984
985
986
987
988

    # 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
989
990
    assert len(qwen3_tool_parser_parametrized.prev_tool_call_arr) == 1

991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
    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"


1004
def test_extract_tool_calls_streaming_incremental(
1005
1006
    qwen3_tool_parser_parametrized, qwen3_tokenizer, sample_tools
):
1007
    """Test that streaming is truly incremental"""
1008
    model_output = """I'll check the weather.<tool_call>
1009
1010
1011
1012
1013
1014
1015
1016
<function=get_current_weather>
<parameter=city>
Dallas
</parameter>
<parameter=state>
TX
</parameter>
</function>
1017
</tool_call>"""
1018

1019
    request = ChatCompletionRequest(model=MODEL, messages=[], tools=sample_tools)
1020
1021
1022

    chunks = []
    for delta_message in stream_delta_message_generator(
1023
1024
        qwen3_tool_parser_parametrized, qwen3_tokenizer, model_output, request
    ):
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
        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
1039
            assert chunk.tool_calls[0].function.name == "get_current_weather"
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
            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"
1060
1061
1062


def test_extract_tool_calls_complex_type_with_single_quote(
1063
    qwen3_tokenizer,
1064
):
1065
1066
    """Test parameter type conversion based on tool schema"""
    tools = [
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
        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"},
                    },
                },
            },
        )
1083
1084
    ]

1085
    model_output = """<tool_call>
1086
1087
1088
1089
1090
<function=test_types>
<parameter=obj_param>
{'key': 'value'}
</parameter>
</function>
1091
</tool_call>"""
1092

1093
    parser = Qwen3XMLToolParser(qwen3_tokenizer, tools=tools)
1094
    request = ChatCompletionRequest(model=MODEL, messages=[], tools=tools)
1095
    extracted_tool_calls = parser.extract_tool_calls(model_output, request=request)
1096
1097
1098

    args = json.loads(extracted_tool_calls.tool_calls[0].function.arguments)
    assert args["obj_param"] == {"key": "value"}
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178


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"
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332


def test_malformed_xml_no_gt_delimiter(qwen3_tool_parser, sample_tools):
    """Regression: malformed XML without '>' must not crash (PR #36774)."""
    model_output = (
        "<tool_call>\n"
        "<function=get_current_weather\n"
        "<parameter=city>Dallas</parameter>\n"
        "</function>\n"
        "</tool_call>"
    )

    request = ChatCompletionRequest(model=MODEL, messages=[], tools=sample_tools)
    result = qwen3_tool_parser.extract_tool_calls(model_output, request=request)
    assert result is not None
    assert isinstance(result.tool_calls, list)
    assert all(tc is not None for tc in result.tool_calls)


def test_none_tool_calls_filtered(qwen3_tool_parser, sample_tools):
    """Regression: None tool calls filtered from output (PR #36774)."""
    model_output = (
        "<tool_call>\n"
        "<function=bad_func_no_gt\n"
        "</function>\n"
        "</tool_call>\n"
        "<tool_call>\n"
        "<function=get_current_weather>\n"
        "<parameter=city>Dallas</parameter>\n"
        "<parameter=state>TX</parameter>\n"
        "</function>\n"
        "</tool_call>"
    )

    request = ChatCompletionRequest(model=MODEL, messages=[], tools=sample_tools)
    result = qwen3_tool_parser.extract_tool_calls(model_output, request=request)
    assert all(tc is not None for tc in result.tool_calls)
    assert result.tools_called
    assert len(result.tool_calls) == 1
    assert result.tool_calls[0].function.name == "get_current_weather"
    args = json.loads(result.tool_calls[0].function.arguments)
    assert args["city"] == "Dallas"
    assert args["state"] == "TX"


def test_anyof_parameter_not_double_encoded(qwen3_tokenizer):
    """Regression: anyOf parameters must not be double-encoded (PR #36032)."""
    tools = [
        ChatCompletionToolsParam(
            type="function",
            function={
                "name": "update_record",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "data": {
                            "anyOf": [{"type": "object"}, {"type": "null"}],
                        },
                    },
                },
            },
        )
    ]

    parser = Qwen3CoderToolParser(qwen3_tokenizer, tools=tools)

    model_output = (
        "<tool_call>\n"
        "<function=update_record>\n"
        '<parameter=data>{"key": "value", "count": 42}</parameter>\n'
        "</function>\n"
        "</tool_call>"
    )

    request = ChatCompletionRequest(model=MODEL, messages=[], tools=tools)
    result = parser.extract_tool_calls(model_output, request=request)

    assert result.tools_called
    assert len(result.tool_calls) == 1
    args = json.loads(result.tool_calls[0].function.arguments)
    assert isinstance(args["data"], dict)
    assert args["data"] == {"key": "value", "count": 42}


def test_streaming_multi_param_single_chunk(
    qwen3_tool_parser, qwen3_tokenizer, sample_tools
):
    """Regression: speculative decode delivering multiple params at once (PR #35615)."""
    request = ChatCompletionRequest(model=MODEL, messages=[], tools=sample_tools)

    deltas = [
        "<tool_call>",
        "\n<function=get_current_weather>",
        "\n",  # triggers json_started -> sends "{"
        # This single delta delivers all three parameters at once
        "<parameter=city>\nDallas\n</parameter>"
        "\n<parameter=state>\nTX\n</parameter>"
        "\n<parameter=unit>\nfahrenheit\n</parameter>",
        "\n</function>",
        "\n</tool_call>",
    ]

    from tests.tool_parsers.utils import (
        run_tool_extraction_streaming,
    )

    reconstructor = run_tool_extraction_streaming(
        qwen3_tool_parser,
        deltas,
        request,
        assert_one_tool_per_delta=False,
    )

    assert len(reconstructor.tool_calls) == 1
    args = json.loads(reconstructor.tool_calls[0].function.arguments)
    assert args["city"] == "Dallas"
    assert args["state"] == "TX"
    assert args["unit"] == "fahrenheit"


def test_no_double_serialization_string_args(qwen3_tool_parser):
    """Regression: string arguments must not be double-serialized (PR #35615)."""
    tools = [
        ChatCompletionToolsParam(
            type="function",
            function={
                "name": "greet",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "message": {"type": "string"},
                    },
                },
            },
        )
    ]

    model_output = (
        "<tool_call>\n"
        "<function=greet>\n"
        "<parameter=message>hello world</parameter>\n"
        "</function>\n"
        "</tool_call>"
    )

    request = ChatCompletionRequest(model=MODEL, messages=[], tools=tools)
    result = qwen3_tool_parser.extract_tool_calls(model_output, request=request)

    assert result.tools_called
    assert len(result.tool_calls) == 1
    raw_arguments = result.tool_calls[0].function.arguments
    args = json.loads(raw_arguments)
    assert args["message"] == "hello world"
    assert '\\"hello world\\"' not in raw_arguments