[Test] Consolidate tool parser unit tests to tests/tool_parsers (#37834)

Signed-off-by: Ben Browning <bbrownin@redhat.com>

tests/entrypoints/openai/tool_parsers/test_hermes_tool_parser.py

@@ -9,8 +9,6 @@ import pytest_asyncio
 from huggingface_hub import snapshot_download
 from typing_extensions import TypedDict
 
-from vllm.entrypoints.openai.chat_completion.protocol import ChatCompletionRequest
-from vllm.tokenizers import TokenizerLike
 from vllm.tool_parsers.abstract_tool_parser import ToolParser
 from vllm.tool_parsers.granite4_tool_parser import Granite4ToolParser
 from vllm.tool_parsers.hermes_tool_parser import Hermes2ProToolParser
@@ -325,202 +323,3 @@ async def test_streaming_product_tool_call(
     print("\n[Streaming Product Test Passed]")
     print(f"Reconstructed Tool Call: {reconstructed_tool_call['name']}")
     print(f"Reconstructed Arguments: {arguments}")
-
-
-@pytest.fixture
-def qwen_tokenizer() -> TokenizerLike:
-    from vllm.tokenizers import get_tokenizer
-
-    return get_tokenizer("Qwen/Qwen3-32B")
-
-
-@pytest.fixture(params=CONFIGS.keys())
-def hermes_parser(request, qwen_tokenizer: TokenizerLike) -> ToolParser:
-    config = CONFIGS[request.param]
-    return config["tool_parser"](qwen_tokenizer)
-
-
-@pytest.fixture
-def any_chat_request() -> ChatCompletionRequest:
-    return ChatCompletionRequest(
-        seed=42,
-        model="Qwen/Qwen3-32B",
-        messages=[],
-    )
-
-
-def test_hermes_parser_streaming_just_forward_text(
-    qwen_tokenizer: TokenizerLike,
-    hermes_parser: ToolParser,
-    any_chat_request: ChatCompletionRequest,
-) -> None:
-    text = """This is some prior text that has nothing to do with tool calling."""
-    tokens = qwen_tokenizer.encode(text)
-    previous_text = ""
-    delta_messages = []
-    for token in tokens:
-        delta_text = qwen_tokenizer.decode([token])
-        current_text = previous_text + delta_text
-        delta = hermes_parser.extract_tool_calls_streaming(
-            previous_text=previous_text,
-            current_text=current_text,
-            delta_text=delta_text,
-            previous_token_ids=[],
-            current_token_ids=[],
-            delta_token_ids=[],
-            request=any_chat_request,
-        )
-        previous_text = current_text
-        delta_messages.append(delta)
-
-    for delta in delta_messages:
-        assert delta is not None
-        assert not delta.tool_calls
-
-    print(delta_messages)
-    assert "".join([delta.content for delta in delta_messages]) == text
-
-
-def test_hermes_parser_streaming_failure_case_bug_19056(
-    qwen_tokenizer: TokenizerLike,
-    hermes_parser: ToolParser,
-    any_chat_request: ChatCompletionRequest,
-) -> None:
-    text = """<tool_call>
-{"name": "final_answer", "arguments": {"trigger": true}}
-</tool_call>"""
-    tokens = qwen_tokenizer.encode(text)
-    previous_text = ""
-    delta_messages = []
-    for token in tokens:
-        text = qwen_tokenizer.decode([token])
-        current_text = previous_text + text
-        delta = hermes_parser.extract_tool_calls_streaming(
-            previous_text=previous_text,
-            current_text=current_text,
-            delta_text=text,
-            previous_token_ids=[],
-            current_token_ids=[],
-            delta_token_ids=[],
-            request=any_chat_request,
-        )
-        previous_text = current_text
-        if delta is not None:
-            delta_messages.append(delta)
-
-    assert delta_messages[0].tool_calls[0].function.name == "final_answer"
-    tool_call_args = "".join(
-        delta.tool_calls[0].function.arguments or "" for delta in delta_messages
-    )
-    assert tool_call_args == '{"trigger": true}'
-
-
-def test_hermes_parser_streaming(
-    qwen_tokenizer: TokenizerLike,
-    hermes_parser: ToolParser,
-    any_chat_request: ChatCompletionRequest,
-) -> None:
-    text = '<tool_call>\
-{"name": "get_current_temperature",\
-"arguments": {"location":\
-"San Francisco, California, United States", "unit": "celsius"}}\
-</tool_call>'
-
-    tokens = qwen_tokenizer.encode(text)
-    previous_text = ""
-    delta_messages = []
-    for token in tokens:
-        text = qwen_tokenizer.decode([token])
-        current_text = previous_text + text
-        delta = hermes_parser.extract_tool_calls_streaming(
-            previous_text=previous_text,
-            current_text=current_text,
-            delta_text=text,
-            previous_token_ids=[],
-            current_token_ids=[],
-            delta_token_ids=[],
-            request=any_chat_request,
-        )
-        previous_text = current_text
-        if delta is not None:
-            delta_messages.append(delta)
-    print(delta_messages)
-    assert delta_messages[0].tool_calls[0].function.name == "get_current_temperature"
-    # load to normalize whitespace
-    tool_call_args = json.loads(
-        "".join(
-            delta.tool_calls[0].function.arguments or "" for delta in delta_messages
-        )
-    )
-    assert tool_call_args == {
-        "location": "San Francisco, California, United States",
-        "unit": "celsius",
-    }
-
-
-def test_hermes_parser_non_streaming_no_tool_call(
-    hermes_parser: ToolParser,
-    any_chat_request: ChatCompletionRequest,
-) -> None:
-    text = """This is not a tool call."""
-    tool_call = hermes_parser.extract_tool_calls(
-        model_output=text,
-        request=any_chat_request,
-    )
-
-    assert tool_call is not None
-    assert not tool_call.tools_called
-
-
-def test_hermes_parser_non_streaming_tool_call_between_tags(
-    hermes_parser: ToolParser,
-    any_chat_request: ChatCompletionRequest,
-) -> None:
-    text = """<tool_call>
-{"name": "final_answer", "arguments": {"trigger": true}}
-</tool_call>"""
-    tool_call = hermes_parser.extract_tool_calls(
-        model_output=text,
-        request=any_chat_request,
-    )
-
-    assert tool_call is not None
-    assert tool_call.tools_called
-    assert tool_call.tool_calls[0].function.name == "final_answer"
-    assert tool_call.tool_calls[0].function.arguments == '{"trigger": true}'
-
-
-def test_hermes_parser_non_streaming_tool_call_until_eos(
-    hermes_parser: ToolParser,
-    any_chat_request: ChatCompletionRequest,
-) -> None:
-    if isinstance(hermes_parser, Granite4ToolParser):
-        pytest.skip(reason="The Granite4 tool parser enforces a complete response")
-
-    text = """<tool_call>
-{"name": "final_answer", "arguments": {"trigger": true}}"""
-    tool_call = hermes_parser.extract_tool_calls(
-        model_output=text,
-        request=any_chat_request,
-    )
-
-    assert tool_call is not None
-    assert tool_call.tools_called
-    assert tool_call.tool_calls[0].function.name == "final_answer"
-    assert tool_call.tool_calls[0].function.arguments == '{"trigger": true}'
-
-
-def test_hermes_parser_non_streaming_tool_call_invalid_json(
-    hermes_parser: ToolParser,
-    any_chat_request: ChatCompletionRequest,
-) -> None:
-    # Missing closing brace to trigger exception
-    text = """<tool_call>
-{"name": "final_answer", "arguments": {"trigger": true}"""
-    tool_call = hermes_parser.extract_tool_calls(
-        model_output=text,
-        request=any_chat_request,
-    )
-
-    assert tool_call is not None
-    assert not tool_call.tools_called