🚧 WIP

src/guidellm/backend/vllm/__init__.py

@@ -16,10 +16,10 @@
 
 check_python_version(min_version="3.8", max_version="3.12")
 
-module_is_available(
-    module="vllm",
-    helper=("`vllm` package is not available. Try run: `pip install -e '.[vllm]'`"),
-)
+# module_is_available(
+#     module="vllm",
+#     helper=("`vllm` package is not available. Try run: `pip install -e '.[vllm]'`"),
+# )
 
 from .backend import VllmBackend  # noqa: E402
 

tests/dummy/__init__.py

@@ -5,4 +5,4 @@
 test.dummy.data.openai_completion_factory - openai.types.Completion test factory
 """
 
-from . import data, services  # noqa: F401
+from . import data, services, vllm  # noqa: F401

tests/dummy/data/__init__.py

@@ -1,3 +1,4 @@
+from . import vllm
 from .openai import openai_completion_factory, openai_model_factory
 
-__all__ = ["openai_completion_factory", "openai_model_factory"]
+__all__ = ["openai_completion_factory", "openai_model_factory", "vllm"]

tests/dummy/vllm.py

@@ -0,0 +1,65 @@
+"""
+This module includes data models factories for the `vllm` 3-rd party package
+"""
+
+import random
+from functools import partial
+from typing import List, Optional
+
+from pydantic import BaseModel, ConfigDict, Field
+
+from guidellm.utils import random_strings
+
+__all__ = ["TestLLM", "CompletionOutput"]
+
+
+class CompletionOutput(BaseModel):
+    """Test interface of `vllm.CompletionOutput`."""
+
+    text: str
+
+
+class SamplingParams(BaseModel):
+    """Test interface of `vllm.SamplingParams`."""
+
+    max_tokens: int
+
+
+class TestLLM(BaseModel):
+    """Test interface of `vllm.LLM`.
+
+    Args:
+        _outputs_number(int | None): the number of generated tokens per output.
+            Should be used only for testing purposes.
+            Default: randint(10..20)
+
+    """
+
+    model_config = ConfigDict(
+        extra="allow",
+        validate_assignment=True,
+        arbitrary_types_allowed=True,
+        from_attributes=True,
+    )
+
+    model: str
+    max_num_batched_tokens: int
+
+    _outputs_number: int = Field(default_factory=partial(random.randint, 10, 20))
+
+    def _generate_completion_outputs(self, max_tokens: int) -> List[CompletionOutput]:
+        self._outputs_number = random.randint(10, 20)
+
+        return [
+            CompletionOutput(text=text)
+            for text in random_strings(
+                min_chars=0, max_chars=max_tokens, n=self._outputs_number
+            )
+        ]
+
+    def generate(
+        self, inputs: List[str], sampling_params: SamplingParams
+    ) -> Optional[List[List[CompletionOutput]]]:
+        return [
+            self._generate_completion_outputs(max_tokens=sampling_params.max_tokens)
+        ]

tests/unit/backend/test_vllm.py

@@ -5,15 +5,14 @@
     the rimtime platform is not a Linux / WSL according to vllm documentation.
 """
 
-import importlib
-import sys
 from typing import Dict, List
 
 import pytest
 
 from guidellm.backend import Backend
 from guidellm.config import reload_settings
 from guidellm.core import TextGenerationRequest
+from tests import dummy
 
 # pytestmark = pytest.mark.skipif(
 #     sys.platform != "linux",
@@ -30,13 +29,9 @@ def backend_class():
 
 @pytest.fixture(autouse=True)
 def mock_vllm_llm(mocker):
-    module = importlib.import_module("vllm")
-    llm = getattr(module, "LLM")(
+    llm = dummy.vllm.TestLLM(
         model="facebook/opt-125m",
         max_num_batched_tokens=4096,
-        tensor_parallel_size=1,
-        gpu_memory_utilization=0.10,
-        enforce_eager=True,
     )
 
     return mocker.patch("vllm.LLM", return_value=llm)