[Misc] Move LRUCache into its own file (#26342)

Signed-off-by: DarkLight1337 <[email protected]>

Author: DarkLight1337

tests/utils_/test_cache.py

@@ -0,0 +1,125 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+from vllm.utils.cache import CacheInfo, LRUCache
+
+
+class TestLRUCache(LRUCache):
+    def _on_remove(self, key, value):
+        if not hasattr(self, "_remove_counter"):
+            self._remove_counter = 0
+        self._remove_counter += 1
+
+
+def test_lru_cache():
+    cache = TestLRUCache(3)
+    assert cache.stat() == CacheInfo(hits=0, total=0)
+    assert cache.stat(delta=True) == CacheInfo(hits=0, total=0)
+
+    cache.put(1, 1)
+    assert len(cache) == 1
+
+    cache.put(1, 1)
+    assert len(cache) == 1
+
+    cache.put(2, 2)
+    assert len(cache) == 2
+
+    cache.put(3, 3)
+    assert len(cache) == 3
+    assert set(cache.cache) == {1, 2, 3}
+
+    cache.put(4, 4)
+    assert len(cache) == 3
+    assert set(cache.cache) == {2, 3, 4}
+    assert cache._remove_counter == 1
+
+    assert cache.get(2) == 2
+    assert cache.stat() == CacheInfo(hits=1, total=1)
+    assert cache.stat(delta=True) == CacheInfo(hits=1, total=1)
+
+    assert cache[2] == 2
+    assert cache.stat() == CacheInfo(hits=2, total=2)
+    assert cache.stat(delta=True) == CacheInfo(hits=1, total=1)
+
+    cache.put(5, 5)
+    assert set(cache.cache) == {2, 4, 5}
+    assert cache._remove_counter == 2
+
+    assert cache.pop(5) == 5
+    assert len(cache) == 2
+    assert set(cache.cache) == {2, 4}
+    assert cache._remove_counter == 3
+
+    assert cache.get(-1) is None
+    assert cache.stat() == CacheInfo(hits=2, total=3)
+    assert cache.stat(delta=True) == CacheInfo(hits=0, total=1)
+
+    cache.pop(10)
+    assert len(cache) == 2
+    assert set(cache.cache) == {2, 4}
+    assert cache._remove_counter == 3
+
+    cache.get(10)
+    assert len(cache) == 2
+    assert set(cache.cache) == {2, 4}
+    assert cache._remove_counter == 3
+
+    cache.put(6, 6)
+    assert len(cache) == 3
+    assert set(cache.cache) == {2, 4, 6}
+    assert 2 in cache
+    assert 4 in cache
+    assert 6 in cache
+
+    cache.remove_oldest()
+    assert len(cache) == 2
+    assert set(cache.cache) == {2, 6}
+    assert cache._remove_counter == 4
+
+    cache.clear()
+    assert len(cache) == 0
+    assert cache._remove_counter == 6
+    assert cache.stat() == CacheInfo(hits=0, total=0)
+    assert cache.stat(delta=True) == CacheInfo(hits=0, total=0)
+
+    cache._remove_counter = 0
+
+    cache[1] = 1
+    assert len(cache) == 1
+
+    cache[1] = 1
+    assert len(cache) == 1
+
+    cache[2] = 2
+    assert len(cache) == 2
+
+    cache[3] = 3
+    assert len(cache) == 3
+    assert set(cache.cache) == {1, 2, 3}
+
+    cache[4] = 4
+    assert len(cache) == 3
+    assert set(cache.cache) == {2, 3, 4}
+    assert cache._remove_counter == 1
+    assert cache[2] == 2
+
+    cache[5] = 5
+    assert set(cache.cache) == {2, 4, 5}
+    assert cache._remove_counter == 2
+
+    del cache[5]
+    assert len(cache) == 2
+    assert set(cache.cache) == {2, 4}
+    assert cache._remove_counter == 3
+
+    cache.pop(10)
+    assert len(cache) == 2
+    assert set(cache.cache) == {2, 4}
+    assert cache._remove_counter == 3
+
+    cache[6] = 6
+    assert len(cache) == 3
+    assert set(cache.cache) == {2, 4, 6}
+    assert 2 in cache
+    assert 4 in cache
+    assert 6 in cache

tests/utils_/test_utils.py

@@ -23,11 +23,8 @@
 from vllm.config import ParallelConfig, VllmConfig, set_current_vllm_config
 from vllm.transformers_utils.detokenizer_utils import convert_ids_list_to_tokens
 
-# isort: off
 from vllm.utils import (
-    CacheInfo,
     FlexibleArgumentParser,
-    LRUCache,
     MemorySnapshot,
     PlaceholderModule,
     bind_kv_cache,
@@ -50,7 +47,6 @@
     unique_filepath,
 )
 
-# isort: on
 from ..utils import create_new_process_for_each_test, error_on_warning
 
 
@@ -557,128 +553,6 @@ def test_bind_kv_cache_pp():
         assert ctx["layers.0.self_attn"].kv_cache[1] is kv_cache[1][0]
 
 
-class TestLRUCache(LRUCache):
-    def _on_remove(self, key, value):
-        if not hasattr(self, "_remove_counter"):
-            self._remove_counter = 0
-        self._remove_counter += 1
-
-
-def test_lru_cache():
-    cache = TestLRUCache(3)
-    assert cache.stat() == CacheInfo(hits=0, total=0)
-    assert cache.stat(delta=True) == CacheInfo(hits=0, total=0)
-
-    cache.put(1, 1)
-    assert len(cache) == 1
-
-    cache.put(1, 1)
-    assert len(cache) == 1
-
-    cache.put(2, 2)
-    assert len(cache) == 2
-
-    cache.put(3, 3)
-    assert len(cache) == 3
-    assert set(cache.cache) == {1, 2, 3}
-
-    cache.put(4, 4)
-    assert len(cache) == 3
-    assert set(cache.cache) == {2, 3, 4}
-    assert cache._remove_counter == 1
-
-    assert cache.get(2) == 2
-    assert cache.stat() == CacheInfo(hits=1, total=1)
-    assert cache.stat(delta=True) == CacheInfo(hits=1, total=1)
-
-    assert cache[2] == 2
-    assert cache.stat() == CacheInfo(hits=2, total=2)
-    assert cache.stat(delta=True) == CacheInfo(hits=1, total=1)
-
-    cache.put(5, 5)
-    assert set(cache.cache) == {2, 4, 5}
-    assert cache._remove_counter == 2
-
-    assert cache.pop(5) == 5
-    assert len(cache) == 2
-    assert set(cache.cache) == {2, 4}
-    assert cache._remove_counter == 3
-
-    assert cache.get(-1) is None
-    assert cache.stat() == CacheInfo(hits=2, total=3)
-    assert cache.stat(delta=True) == CacheInfo(hits=0, total=1)
-
-    cache.pop(10)
-    assert len(cache) == 2
-    assert set(cache.cache) == {2, 4}
-    assert cache._remove_counter == 3
-
-    cache.get(10)
-    assert len(cache) == 2
-    assert set(cache.cache) == {2, 4}
-    assert cache._remove_counter == 3
-
-    cache.put(6, 6)
-    assert len(cache) == 3
-    assert set(cache.cache) == {2, 4, 6}
-    assert 2 in cache
-    assert 4 in cache
-    assert 6 in cache
-
-    cache.remove_oldest()
-    assert len(cache) == 2
-    assert set(cache.cache) == {2, 6}
-    assert cache._remove_counter == 4
-
-    cache.clear()
-    assert len(cache) == 0
-    assert cache._remove_counter == 6
-    assert cache.stat() == CacheInfo(hits=0, total=0)
-    assert cache.stat(delta=True) == CacheInfo(hits=0, total=0)
-
-    cache._remove_counter = 0
-
-    cache[1] = 1
-    assert len(cache) == 1
-
-    cache[1] = 1
-    assert len(cache) == 1
-
-    cache[2] = 2
-    assert len(cache) == 2
-
-    cache[3] = 3
-    assert len(cache) == 3
-    assert set(cache.cache) == {1, 2, 3}
-
-    cache[4] = 4
-    assert len(cache) == 3
-    assert set(cache.cache) == {2, 3, 4}
-    assert cache._remove_counter == 1
-    assert cache[2] == 2
-
-    cache[5] = 5
-    assert set(cache.cache) == {2, 4, 5}
-    assert cache._remove_counter == 2
-
-    del cache[5]
-    assert len(cache) == 2
-    assert set(cache.cache) == {2, 4}
-    assert cache._remove_counter == 3
-
-    cache.pop(10)
-    assert len(cache) == 2
-    assert set(cache.cache) == {2, 4}
-    assert cache._remove_counter == 3
-
-    cache[6] = 6
-    assert len(cache) == 3
-    assert set(cache.cache) == {2, 4, 6}
-    assert 2 in cache
-    assert 4 in cache
-    assert 6 in cache
-
-
 @pytest.mark.parametrize(
     ("src_dtype", "tgt_dtype", "expected_result"),
     [

vllm/lora/models.py

@@ -32,7 +32,8 @@
 from vllm.model_executor.models.module_mapping import MultiModelKeys
 from vllm.model_executor.models.utils import PPMissingLayer, WeightsMapper
 from vllm.model_executor.utils import get_packed_modules_mapping
-from vllm.utils import LRUCache, is_pin_memory_available
+from vllm.utils import is_pin_memory_available
+from vllm.utils.cache import LRUCache
 
 logger = init_logger(__name__)
 

vllm/multimodal/cache.py

@@ -17,7 +17,8 @@
 )
 from vllm.envs import VLLM_OBJECT_STORAGE_SHM_BUFFER_NAME
 from vllm.logger import init_logger
-from vllm.utils import GiB_bytes, LRUCache, MiB_bytes
+from vllm.utils import GiB_bytes, MiB_bytes
+from vllm.utils.cache import LRUCache
 from vllm.utils.jsontree import json_count_leaves, json_map_leaves, json_reduce_leaves
 
 from .inputs import (