| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | # ruff: noqa: E501 |
| 4 | """Tests for the GLM-4.7 tool call parser.""" |
| 5 | |
| 6 | import json |
| 7 | from unittest.mock import Mock |
| 8 | |
| 9 | import pytest |
| 10 | from openai.types.responses import ResponseFunctionToolCall |
| 11 | |
| 12 | from vllm.entrypoints.generate.base.protocol import FunctionCall |
| 13 | from vllm.entrypoints.openai.chat_completion.protocol import ( |
| 14 | ChatCompletionRequest, |
| 15 | ChatCompletionToolsParam, |
| 16 | FunctionDefinition, |
| 17 | ) |
| 18 | from vllm.entrypoints.openai.responses.protocol import ResponsesRequest |
| 19 | from vllm.entrypoints.openai.responses.utils import build_response_output_items |
| 20 | from vllm.tokenizers import get_tokenizer |
| 21 | from vllm.tool_parsers.glm47_moe_tool_parser import Glm47MoeModelToolParser |
| 22 | |
| 23 | MODEL = "zai-org/GLM-4.7" |
| 24 | |
| 25 | |
| 26 | @pytest.fixture(scope="module") |
| 27 | def glm47_tokenizer(): |
| 28 | return get_tokenizer(tokenizer_name=MODEL) |
| 29 | |
| 30 | |
| 31 | @pytest.fixture |
| 32 | def sample_tools(): |
| 33 | return [ |
| 34 | ChatCompletionToolsParam( |
| 35 | function=FunctionDefinition(name="get_current_date", parameters={}), |
| 36 | ), |
| 37 | ChatCompletionToolsParam( |
| 38 | function=FunctionDefinition( |
| 39 | name="get_weather", |
| 40 | parameters={ |
| 41 | "type": "object", |
| 42 | "properties": { |
| 43 | "city": {"type": "string"}, |
| 44 | "date": {"type": "string"}, |
| 45 | }, |
| 46 | }, |
| 47 | ), |
| 48 | ), |
| 49 | ] |
| 50 | |
| 51 | |
| 52 | @pytest.fixture |
| 53 | def glm47_tool_parser(glm47_tokenizer, sample_tools): |
| 54 | return Glm47MoeModelToolParser(glm47_tokenizer, tools=sample_tools) |
| 55 | |
| 56 | |
| 57 | @pytest.fixture |
| 58 | def mock_request(sample_tools) -> ChatCompletionRequest: |
| 59 | request = Mock(spec=ChatCompletionRequest) |
| 60 | request.tools = sample_tools |
| 61 | request.tool_choice = "auto" |
| 62 | return request |
| 63 | |
| 64 | |
| 65 | @pytest.fixture |
| 66 | def namespace_tool_request() -> ResponsesRequest: |
| 67 | return ResponsesRequest.model_validate( |
| 68 | { |
| 69 | "input": "hi", |
| 70 | "tools": [ |
| 71 | { |
| 72 | "type": "namespace", |
| 73 | "name": "mcp__computer_use", |
| 74 | "description": "Computer use tools.", |
| 75 | "tools": [ |
| 76 | { |
| 77 | "type": "function", |
| 78 | "name": "get_app_state", |
| 79 | "description": "Get app state.", |
| 80 | "parameters": { |
| 81 | "type": "object", |
| 82 | "properties": { |
| 83 | "app": {"type": "string"}, |
| 84 | }, |
| 85 | }, |
| 86 | } |
| 87 | ], |
| 88 | } |
| 89 | ], |
| 90 | } |
| 91 | ) |
| 92 | |
| 93 | |
| 94 | class TestGlm47ExtractToolCalls: |
| 95 | def test_namespace_tool_call_round_trip_to_responses_output( |
| 96 | self, glm47_tokenizer, namespace_tool_request |
| 97 | ): |
| 98 | parser = Glm47MoeModelToolParser( |
| 99 | glm47_tokenizer, tools=namespace_tool_request.tools |
| 100 | ) |
| 101 | out = ( |
| 102 | "<tool_call>mcp__computer_use__get_app_state" |
| 103 | "<arg_key>app</arg_key>" |
| 104 | "<arg_value>Google Chrome</arg_value>" |
| 105 | "</tool_call>" |
| 106 | ) |
| 107 | |
| 108 | result = parser.extract_tool_calls(out, request=namespace_tool_request) |
| 109 | |
| 110 | assert result.tools_called |
| 111 | tool_call = result.tool_calls[0].function |
| 112 | assert tool_call == FunctionCall( |
| 113 | name="mcp__computer_use__get_app_state", |
| 114 | arguments='{"app": "Google Chrome"}', |
| 115 | ) |
| 116 | |
| 117 | output_items = build_response_output_items( |
| 118 | reasoning=None, |
| 119 | content=None, |
| 120 | tool_calls=[tool_call], |
| 121 | tools=namespace_tool_request.tools, |
| 122 | ) |
| 123 | output_tool_call = output_items[0] |
| 124 | assert isinstance(output_tool_call, ResponseFunctionToolCall) |
| 125 | assert output_tool_call.name == "get_app_state" |
| 126 | assert output_tool_call.namespace == "mcp__computer_use" |
| 127 | |
| 128 | def test_no_tool_call(self, glm47_tool_parser, mock_request): |
| 129 | out = "This is a plain response." |
| 130 | r = glm47_tool_parser.extract_tool_calls(out, request=mock_request) |
| 131 | assert not r.tools_called |
| 132 | assert r.content == out |
| 133 | |
| 134 | def test_zero_arg_inline(self, glm47_tool_parser, mock_request): |
| 135 | out = "<tool_call>get_current_date</tool_call>" |
| 136 | r = glm47_tool_parser.extract_tool_calls(out, request=mock_request) |
| 137 | assert r.tools_called |
| 138 | assert r.tool_calls[0].function.name == "get_current_date" |
| 139 | assert json.loads(r.tool_calls[0].function.arguments) == {} |
| 140 | assert r.content is None |
| 141 | |
| 142 | def test_zero_arg_newline(self, glm47_tool_parser, mock_request): |
| 143 | out = "<tool_call>get_current_date\n</tool_call>" |
| 144 | r = glm47_tool_parser.extract_tool_calls(out, request=mock_request) |
| 145 | assert r.tools_called |
| 146 | assert r.tool_calls[0].function.name == "get_current_date" |
| 147 | |
| 148 | def test_args_same_line(self, glm47_tool_parser, mock_request): |
| 149 | out = "<tool_call>get_weather<arg_key>city</arg_key><arg_value>Beijing</arg_value></tool_call>" |
| 150 | r = glm47_tool_parser.extract_tool_calls(out, request=mock_request) |
| 151 | assert r.tools_called |
| 152 | assert json.loads(r.tool_calls[0].function.arguments) == {"city": "Beijing"} |
| 153 | |
| 154 | def test_args_with_newlines(self, glm47_tool_parser, mock_request): |
| 155 | out = "<tool_call>get_weather\n<arg_key>city</arg_key>\n<arg_value>Beijing</arg_value>\n</tool_call>" |
| 156 | r = glm47_tool_parser.extract_tool_calls(out, request=mock_request) |
| 157 | assert r.tools_called |
| 158 | assert json.loads(r.tool_calls[0].function.arguments) == {"city": "Beijing"} |
| 159 | |
| 160 | def test_whitespace_preserved_in_arg_values(self, glm47_tool_parser, mock_request): |
| 161 | out = "<tool_call>get_weather<arg_key>city</arg_key><arg_value> Beijing </arg_value></tool_call>" |
| 162 | r = glm47_tool_parser.extract_tool_calls(out, request=mock_request) |
| 163 | assert r.tools_called |
| 164 | assert json.loads(r.tool_calls[0].function.arguments) == {"city": " Beijing "} |
| 165 | |
| 166 | def test_content_before(self, glm47_tool_parser, mock_request): |
| 167 | out = "Checking.<tool_call>get_current_date</tool_call>" |
| 168 | r = glm47_tool_parser.extract_tool_calls(out, request=mock_request) |
| 169 | assert r.tools_called |
| 170 | assert r.content == "Checking." |
| 171 | |
| 172 | def test_multiple(self, glm47_tool_parser, mock_request): |
| 173 | out = ( |
| 174 | "<tool_call>get_weather<arg_key>city</arg_key><arg_value>Beijing</arg_value></tool_call>" |
| 175 | "<tool_call>get_weather<arg_key>city</arg_key><arg_value>Shanghai</arg_value></tool_call>" |
| 176 | ) |
| 177 | r = glm47_tool_parser.extract_tool_calls(out, request=mock_request) |
| 178 | assert len(r.tool_calls) == 2 |
| 179 | |
| 180 | def test_empty_content_none(self, glm47_tool_parser, mock_request): |
| 181 | out = "<tool_call>get_current_date</tool_call>" |
| 182 | r = glm47_tool_parser.extract_tool_calls(out, request=mock_request) |
| 183 | assert r.content is None |
| 184 | |
| 185 | def test_whitespace_content_none(self, glm47_tool_parser, mock_request): |
| 186 | out = " \n <tool_call>get_current_date</tool_call>" |
| 187 | r = glm47_tool_parser.extract_tool_calls(out, request=mock_request) |
| 188 | assert r.content is None |
| 189 | |
| 190 | |
| 191 | def _reset(parser): |
| 192 | parser.current_tool_name_sent = False |
| 193 | parser.prev_tool_call_arr = [] |
| 194 | parser.current_tool_id = -1 |
| 195 | parser.streamed_args_for_tool = [] |
| 196 | parser._tool_call_ids = [] |
| 197 | parser._sent_content_idx = 0 |
| 198 | |
| 199 | |
| 200 | class TestGlm47Streaming: |
| 201 | def test_no_args(self, glm47_tool_parser, mock_request): |
| 202 | _reset(glm47_tool_parser) |
| 203 | chunks = ["<tool_call>", "get_current_date", "</tool_call>"] |
| 204 | current_text = "" |
| 205 | deltas = [] |
| 206 | for chunk in chunks: |
| 207 | current_text += chunk |
| 208 | delta = glm47_tool_parser.extract_tool_calls_streaming( |
| 209 | previous_text="", |
| 210 | current_text=current_text, |
| 211 | delta_text=chunk, |
| 212 | previous_token_ids=[], |
| 213 | current_token_ids=[], |
| 214 | delta_token_ids=[], |
| 215 | request=mock_request, |
| 216 | ) |
| 217 | if delta: |
| 218 | deltas.append(delta) |
| 219 | tool_calls = [ |
| 220 | tool_call for delta in deltas for tool_call in (delta.tool_calls or []) |
| 221 | ] |
| 222 | names = [ |
| 223 | tool_call.function.name |
| 224 | for tool_call in tool_calls |
| 225 | if tool_call.function and tool_call.function.name |
| 226 | ] |
| 227 | arguments = [ |
| 228 | tool_call.function.arguments |
| 229 | for tool_call in tool_calls |
| 230 | if tool_call.function and tool_call.function.arguments |
| 231 | ] |
| 232 | assert names == ["get_current_date"] |
| 233 | assert "".join(arguments) == "{}" |
| 234 | |
| 235 | def test_with_args(self, glm47_tool_parser, mock_request): |
| 236 | _reset(glm47_tool_parser) |
| 237 | chunks = [ |
| 238 | "<tool_call>", |
| 239 | "get_weather\n", |
| 240 | "<arg_key>city</arg_key>", |
| 241 | "<arg_value>", |
| 242 | "Beijing", |
| 243 | "</arg_value>", |
| 244 | "</tool_call>", |
| 245 | ] |
| 246 | current_text = "" |
| 247 | deltas = [] |
| 248 | for chunk in chunks: |
| 249 | current_text += chunk |
| 250 | delta = glm47_tool_parser.extract_tool_calls_streaming( |
| 251 | previous_text="", |
| 252 | current_text=current_text, |
| 253 | delta_text=chunk, |
| 254 | previous_token_ids=[], |
| 255 | current_token_ids=[], |
| 256 | delta_token_ids=[], |
| 257 | request=mock_request, |
| 258 | ) |
| 259 | if delta: |
| 260 | deltas.append(delta) |
| 261 | arguments = [ |
| 262 | tool_call.function.arguments |
| 263 | for delta in deltas |
| 264 | for tool_call in (delta.tool_calls or []) |
| 265 | if tool_call.function and tool_call.function.arguments |
| 266 | ] |
| 267 | args = json.loads("".join(arguments)) |
| 268 | assert args["city"] == "Beijing" |