|
| 1 | +# Copyright (c) 2023 Oracle and/or its affiliates. |
| 2 | +# Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/ |
| 3 | + |
| 4 | +"""Unit tests for tool call optimization.""" |
| 5 | + |
| 6 | +from unittest.mock import MagicMock |
| 7 | + |
| 8 | +import pytest |
| 9 | +from langchain_core.messages import HumanMessage |
| 10 | + |
| 11 | +from langchain_oci.chat_models.oci_generative_ai import ChatOCIGenAI |
| 12 | + |
| 13 | + |
| 14 | +class MockResponseDict(dict): |
| 15 | + def __getattr__(self, val): # type: ignore[no-untyped-def] |
| 16 | + return self.get(val) |
| 17 | + |
| 18 | + |
| 19 | +@pytest.mark.requires("oci") |
| 20 | +def test_meta_tool_call_optimization() -> None: |
| 21 | + """Test that tool calls are formatted once and cached for Meta models.""" |
| 22 | + oci_gen_ai_client = MagicMock() |
| 23 | + |
| 24 | + # Mock response with tool call |
| 25 | + def mocked_response(*args): # type: ignore[no-untyped-def] |
| 26 | + return MockResponseDict( |
| 27 | + { |
| 28 | + "status": 200, |
| 29 | + "data": MockResponseDict( |
| 30 | + { |
| 31 | + "chat_response": MockResponseDict( |
| 32 | + { |
| 33 | + "api_format": "GENERIC", |
| 34 | + "choices": [ |
| 35 | + MockResponseDict( |
| 36 | + { |
| 37 | + "message": MockResponseDict( |
| 38 | + { |
| 39 | + "role": "ASSISTANT", |
| 40 | + "name": None, |
| 41 | + "content": [ |
| 42 | + MockResponseDict( |
| 43 | + { |
| 44 | + "text": "", |
| 45 | + "type": "TEXT", |
| 46 | + } |
| 47 | + ) |
| 48 | + ], |
| 49 | + "tool_calls": [ |
| 50 | + MockResponseDict( |
| 51 | + { |
| 52 | + "id": "test_id_123", |
| 53 | + "type": "FUNCTION", |
| 54 | + "function": MockResponseDict( |
| 55 | + { |
| 56 | + "name": "get_weather", |
| 57 | + "arguments": '{"location": "San Francisco"}', |
| 58 | + } |
| 59 | + ), |
| 60 | + } |
| 61 | + ) |
| 62 | + ], |
| 63 | + } |
| 64 | + ), |
| 65 | + "finish_reason": "TOOL_CALLS", |
| 66 | + "logprobs": None, |
| 67 | + "index": 0, |
| 68 | + } |
| 69 | + ) |
| 70 | + ], |
| 71 | + "time_created": "2024-01-01T00:00:00Z", |
| 72 | + "usage": MockResponseDict( |
| 73 | + { |
| 74 | + "input_tokens": 100, |
| 75 | + "output_tokens": 50, |
| 76 | + "total_tokens": 150, |
| 77 | + } |
| 78 | + ), |
| 79 | + } |
| 80 | + ), |
| 81 | + "model_id": "meta.llama-3.3-70b-instruct", |
| 82 | + "model_version": "1.0.0", |
| 83 | + } |
| 84 | + ), |
| 85 | + "request_id": "test_request_123", |
| 86 | + "headers": MockResponseDict( |
| 87 | + { |
| 88 | + "content-length": "500", |
| 89 | + } |
| 90 | + ), |
| 91 | + } |
| 92 | + ) |
| 93 | + |
| 94 | + oci_gen_ai_client.chat.side_effect = mocked_response |
| 95 | + |
| 96 | + # Create LLM with mocked client |
| 97 | + llm = ChatOCIGenAI(model_id="meta.llama-3.3-70b-instruct", client=oci_gen_ai_client) |
| 98 | + |
| 99 | + # Define a simple tool |
| 100 | + def get_weather(location: str) -> str: |
| 101 | + """Get weather for a location.""" |
| 102 | + return f"Weather in {location}" |
| 103 | + |
| 104 | + # Bind tools |
| 105 | + llm_with_tools = llm.bind_tools([get_weather]) |
| 106 | + |
| 107 | + # Invoke |
| 108 | + response = llm_with_tools.invoke([HumanMessage(content="What's the weather in SF?")]) |
| 109 | + |
| 110 | + # Verify tool_calls field is populated |
| 111 | + assert len(response.tool_calls) == 1, "Should have one tool call" |
| 112 | + tool_call = response.tool_calls[0] |
| 113 | + assert tool_call["name"] == "get_weather" |
| 114 | + assert tool_call["args"] == {"location": "San Francisco"} |
| 115 | + assert "id" in tool_call |
| 116 | + |
| 117 | + # Verify additional_kwargs contains formatted tool calls |
| 118 | + assert "tool_calls" in response.additional_kwargs, "Should have tool_calls in additional_kwargs" |
| 119 | + additional_tool_calls = response.additional_kwargs["tool_calls"] |
| 120 | + assert len(additional_tool_calls) == 1 |
| 121 | + assert additional_tool_calls[0]["type"] == "function" |
| 122 | + assert additional_tool_calls[0]["function"]["name"] == "get_weather" |
| 123 | + assert "location" in str(additional_tool_calls[0]["function"]["arguments"]) |
| 124 | + |
| 125 | + |
| 126 | +@pytest.mark.requires("oci") |
| 127 | +def test_cohere_tool_call_optimization() -> None: |
| 128 | + """Test that tool calls are formatted once and cached for Cohere models.""" |
| 129 | + oci_gen_ai_client = MagicMock() |
| 130 | + |
| 131 | + # Mock response with tool call |
| 132 | + def mocked_response(*args): # type: ignore[no-untyped-def] |
| 133 | + return MockResponseDict( |
| 134 | + { |
| 135 | + "status": 200, |
| 136 | + "data": MockResponseDict( |
| 137 | + { |
| 138 | + "chat_response": MockResponseDict( |
| 139 | + { |
| 140 | + "text": "", |
| 141 | + "finish_reason": "TOOL_CALL", |
| 142 | + "tool_calls": [ |
| 143 | + MockResponseDict( |
| 144 | + { |
| 145 | + "name": "get_weather", |
| 146 | + "parameters": {"location": "London"}, |
| 147 | + } |
| 148 | + ) |
| 149 | + ], |
| 150 | + "usage": MockResponseDict( |
| 151 | + { |
| 152 | + "total_tokens": 100, |
| 153 | + } |
| 154 | + ), |
| 155 | + } |
| 156 | + ), |
| 157 | + "model_id": "cohere.command-r-plus", |
| 158 | + "model_version": "1.0.0", |
| 159 | + } |
| 160 | + ), |
| 161 | + "request_id": "test_request_456", |
| 162 | + "headers": MockResponseDict( |
| 163 | + { |
| 164 | + "content-length": "300", |
| 165 | + } |
| 166 | + ), |
| 167 | + } |
| 168 | + ) |
| 169 | + |
| 170 | + oci_gen_ai_client.chat.side_effect = mocked_response |
| 171 | + |
| 172 | + # Create LLM with mocked client |
| 173 | + llm = ChatOCIGenAI(model_id="cohere.command-r-plus", client=oci_gen_ai_client) |
| 174 | + |
| 175 | + # Define a simple tool |
| 176 | + def get_weather(location: str) -> str: |
| 177 | + """Get weather for a location.""" |
| 178 | + return f"Weather in {location}" |
| 179 | + |
| 180 | + # Bind tools |
| 181 | + llm_with_tools = llm.bind_tools([get_weather]) |
| 182 | + |
| 183 | + # Invoke |
| 184 | + response = llm_with_tools.invoke([HumanMessage(content="What's the weather in London?")]) |
| 185 | + |
| 186 | + # Verify tool_calls field is populated |
| 187 | + assert len(response.tool_calls) == 1, "Should have one tool call" |
| 188 | + tool_call = response.tool_calls[0] |
| 189 | + assert tool_call["name"] == "get_weather" |
| 190 | + assert tool_call["args"] == {"location": "London"} |
| 191 | + assert "id" in tool_call |
| 192 | + assert isinstance(tool_call["id"], str) |
| 193 | + assert len(tool_call["id"]) > 0, "Tool call ID should not be empty" |
| 194 | + |
| 195 | + # Verify additional_kwargs contains formatted tool calls |
| 196 | + assert "tool_calls" in response.additional_kwargs, "Should have tool_calls in additional_kwargs" |
| 197 | + additional_tool_calls = response.additional_kwargs["tool_calls"] |
| 198 | + assert len(additional_tool_calls) == 1 |
| 199 | + assert additional_tool_calls[0]["type"] == "function" |
| 200 | + assert additional_tool_calls[0]["function"]["name"] == "get_weather" |
| 201 | + |
| 202 | + |
| 203 | +@pytest.mark.requires("oci") |
| 204 | +def test_multiple_tool_calls_optimization() -> None: |
| 205 | + """Test optimization with multiple tool calls.""" |
| 206 | + oci_gen_ai_client = MagicMock() |
| 207 | + |
| 208 | + # Mock response with multiple tool calls |
| 209 | + def mocked_response(*args): # type: ignore[no-untyped-def] |
| 210 | + return MockResponseDict( |
| 211 | + { |
| 212 | + "status": 200, |
| 213 | + "data": MockResponseDict( |
| 214 | + { |
| 215 | + "chat_response": MockResponseDict( |
| 216 | + { |
| 217 | + "api_format": "GENERIC", |
| 218 | + "choices": [ |
| 219 | + MockResponseDict( |
| 220 | + { |
| 221 | + "message": MockResponseDict( |
| 222 | + { |
| 223 | + "role": "ASSISTANT", |
| 224 | + "content": [ |
| 225 | + MockResponseDict( |
| 226 | + { |
| 227 | + "text": "", |
| 228 | + "type": "TEXT", |
| 229 | + } |
| 230 | + ) |
| 231 | + ], |
| 232 | + "tool_calls": [ |
| 233 | + MockResponseDict( |
| 234 | + { |
| 235 | + "id": "call_1", |
| 236 | + "type": "FUNCTION", |
| 237 | + "function": MockResponseDict( |
| 238 | + { |
| 239 | + "name": "get_weather", |
| 240 | + "arguments": '{"location": "Tokyo"}', |
| 241 | + } |
| 242 | + ), |
| 243 | + } |
| 244 | + ), |
| 245 | + MockResponseDict( |
| 246 | + { |
| 247 | + "id": "call_2", |
| 248 | + "type": "FUNCTION", |
| 249 | + "function": MockResponseDict( |
| 250 | + { |
| 251 | + "name": "get_population", |
| 252 | + "arguments": '{"city": "Tokyo"}', |
| 253 | + } |
| 254 | + ), |
| 255 | + } |
| 256 | + ), |
| 257 | + ], |
| 258 | + } |
| 259 | + ), |
| 260 | + "finish_reason": "TOOL_CALLS", |
| 261 | + "index": 0, |
| 262 | + } |
| 263 | + ) |
| 264 | + ], |
| 265 | + "usage": MockResponseDict( |
| 266 | + { |
| 267 | + "total_tokens": 200, |
| 268 | + } |
| 269 | + ), |
| 270 | + } |
| 271 | + ), |
| 272 | + "model_id": "meta.llama-3.3-70b-instruct", |
| 273 | + "model_version": "1.0.0", |
| 274 | + } |
| 275 | + ), |
| 276 | + "request_id": "test_request_789", |
| 277 | + } |
| 278 | + ) |
| 279 | + |
| 280 | + oci_gen_ai_client.chat.side_effect = mocked_response |
| 281 | + |
| 282 | + # Create LLM with mocked client |
| 283 | + llm = ChatOCIGenAI(model_id="meta.llama-3.3-70b-instruct", client=oci_gen_ai_client) |
| 284 | + |
| 285 | + # Define tools |
| 286 | + def get_weather(location: str) -> str: |
| 287 | + """Get weather.""" |
| 288 | + return f"Weather in {location}" |
| 289 | + |
| 290 | + def get_population(city: str) -> int: |
| 291 | + """Get population.""" |
| 292 | + return 1000000 |
| 293 | + |
| 294 | + # Bind tools |
| 295 | + llm_with_tools = llm.bind_tools([get_weather, get_population]) |
| 296 | + |
| 297 | + # Invoke |
| 298 | + response = llm_with_tools.invoke([HumanMessage(content="Weather and population of Tokyo?")]) |
| 299 | + |
| 300 | + # Verify tool_calls field has both calls |
| 301 | + assert len(response.tool_calls) == 2, "Should have two tool calls" |
| 302 | + |
| 303 | + # Check first tool call |
| 304 | + assert response.tool_calls[0]["name"] == "get_weather" |
| 305 | + assert response.tool_calls[0]["args"] == {"location": "Tokyo"} |
| 306 | + assert "id" in response.tool_calls[0] |
| 307 | + |
| 308 | + # Check second tool call |
| 309 | + assert response.tool_calls[1]["name"] == "get_population" |
| 310 | + assert response.tool_calls[1]["args"] == {"city": "Tokyo"} |
| 311 | + assert "id" in response.tool_calls[1] |
| 312 | + |
| 313 | + # Verify IDs are unique |
| 314 | + assert response.tool_calls[0]["id"] != response.tool_calls[1]["id"] |
| 315 | + |
| 316 | + # Verify additional_kwargs has both formatted calls |
| 317 | + assert "tool_calls" in response.additional_kwargs |
| 318 | + assert len(response.additional_kwargs["tool_calls"]) == 2 |
0 commit comments