[CPU] Add device-memory-utilization CLI alias (#56547)

Signed-off-by: louie-tsai <louie.tsai@intel.com>
Signed-off-by: Louie Tsai <louie.tsai@intel.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>

tests/engine/test_arg_utils.py

@@ -9,7 +9,13 @@ from typing import Annotated, Literal
 import pytest
 from pydantic import Field
 
-from vllm.config import AttentionConfig, CompilationConfig, ModelConfig, config
+from vllm.config import (
+    AttentionConfig,
+    CacheConfig,
+    CompilationConfig,
+    ModelConfig,
+    config,
+)
 from vllm.engine.arg_utils import (
     EngineArgs,
     _expand_json_human_readable_numbers,
@@ -86,6 +92,26 @@ def test_watermark_config_cli():
     assert config.deduplicate_contexts_max_history == 32
 
 
+@pytest.mark.parametrize(
+    "option",
+    ["--gpu-memory-utilization", "--device-memory-utilization"],
+)
+def test_memory_utilization_cli_aliases(option):
+    parser = EngineArgs.add_cli_args(FlexibleArgumentParser())
+    args = EngineArgs.from_cli_args(parser.parse_args([option, "0.8"]))
+
+    assert args.gpu_memory_utilization == 0.8
+
+
+def test_device_memory_utilization_property():
+    config = CacheConfig(gpu_memory_utilization=0.8)
+
+    assert config.device_memory_utilization == 0.8
+
+    config.device_memory_utilization = 0.7
+    assert config.gpu_memory_utilization == 0.7
+
+
 @pytest.mark.parametrize(
     "options",
     [

vllm/config/cache.py

@@ -107,7 +107,19 @@ class CacheConfig:
     per-instance limit, and only applies to the current vLLM instance. It does
     not matter if you have another vLLM instance running on the same GPU. For
     example, if you have two vLLM instances running on the same GPU, you can
-    set the GPU memory utilization to 0.5 for each instance."""
+    set the GPU memory utilization to 0.5 for each instance. On non-GPU
+    installs, this value controls the corresponding device memory utilization.
+    """
+
+    @property
+    def device_memory_utilization(self) -> float:
+        """Device-neutral alias for ``gpu_memory_utilization``."""
+        return self.gpu_memory_utilization
+
+    @device_memory_utilization.setter
+    def device_memory_utilization(self, value: float) -> None:
+        self.gpu_memory_utilization = value
+
     cache_dtype: CacheDType = "auto"
     """Data type for kv cache storage. If "auto", will use model data type.
     CUDA 11.8+ supports fp8 (=fp8_e4m3) and fp8_e5m2. ROCm (AMD GPU) supports

vllm/engine/arg_utils.py

@@ -1298,7 +1298,10 @@ class EngineArgs:
         )
         cache_group.add_argument("--block-size", **cache_kwargs["block_size"])
         cache_group.add_argument(
-            "--gpu-memory-utilization", **cache_kwargs["gpu_memory_utilization"]
+            "--gpu-memory-utilization",
+            "--device-memory-utilization",
+            dest="gpu_memory_utilization",
+            **cache_kwargs["gpu_memory_utilization"],
         )
         cache_group.add_argument(
             "--kv-cache-memory-bytes", **cache_kwargs["kv_cache_memory_bytes"]