[Misc] Further clean up chunked prefill and prefix caching init (#29186)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

vllm/config/cache.py

@@ -73,8 +73,8 @@ class CacheConfig:
     sliding_window: int | None = None
     """Sliding window size for the KV cache. This is primarily set in
     `ModelConfig` and that value should be manually duplicated here."""
-    enable_prefix_caching: bool | None = None
-    """Whether to enable prefix caching. Enabled by default for V1."""
+    enable_prefix_caching: bool = True
+    """Whether to enable prefix caching."""
     prefix_caching_hash_algo: PrefixCachingHashAlgo = "sha256"
     """Set the hash algorithm for prefix caching:\n
     - "sha256" uses Pickle for object serialization before hashing.\n

vllm/engine/arg_utils.py

@@ -425,7 +425,7 @@ class EngineArgs:
         ParallelConfig.max_parallel_loading_workers
     )
     block_size: BlockSize | None = CacheConfig.block_size
-    enable_prefix_caching: bool | None = CacheConfig.enable_prefix_caching
+    enable_prefix_caching: bool | None = None
     prefix_caching_hash_algo: PrefixCachingHashAlgo = (
         CacheConfig.prefix_caching_hash_algo
     )
@@ -1975,10 +1975,11 @@ class EngineArgs:
         if self.prefill_context_parallel_size > 1:
             default_chunked_prefill = False
             default_prefix_caching = False
-            logger.warning(
+            logger.warning_once(
                 "--prefill-context-parallel-size > 1 is not compatible with "
                 "chunked prefill and prefix caching now. Chunked prefill "
-                "and prefix caching have been disabled by default."
+                "and prefix caching have been disabled by default.",
+                scope="local",
             )
 
         if self.enable_chunked_prefill is None:
@@ -1988,15 +1989,27 @@ class EngineArgs:
                 "%s chunked prefill by default",
                 "Enabling" if default_chunked_prefill else "Disabling",
             )
+        elif (
+            model_config.runner_type == "generate"
+            and not self.enable_chunked_prefill
+            and default_chunked_prefill
+        ):
+            logger.warning_once(
+                "This model does not officially support disabling chunked prefill. "
+                "Disabling this manually may cause the engine to crash "
+                "or produce incorrect outputs.",
+                scope="local",
+            )
         elif (
             model_config.runner_type == "pooling"
             and self.enable_chunked_prefill
             and not default_chunked_prefill
         ):
-            logger.warning(
+            logger.warning_once(
                 "This model does not officially support chunked prefill. "
                 "Enabling this manually may cause the engine to crash "
                 "or produce incorrect outputs.",
+                scope="local",
             )
 
         if self.enable_prefix_caching is None:
@@ -2011,10 +2024,11 @@ class EngineArgs:
             and self.enable_prefix_caching
             and not default_prefix_caching
         ):
-            logger.warning(
+            logger.warning_once(
                 "This model does not officially support prefix caching. "
                 "Enabling this manually may cause the engine to crash "
                 "or produce incorrect outputs.",
+                scope="local",
             )
 
         world_size = self.pipeline_parallel_size * self.tensor_parallel_size

vllm/v1/core/sched/scheduler.py

@@ -180,7 +180,7 @@ class Scheduler(SchedulerInterface):
         self.kv_cache_manager = KVCacheManager(
             kv_cache_config=kv_cache_config,
             max_model_len=self.max_model_len,
-            enable_caching=bool(self.cache_config.enable_prefix_caching),
+            enable_caching=self.cache_config.enable_prefix_caching,
             use_eagle=self.use_eagle,
             log_stats=self.log_stats,
             enable_kv_cache_events=self.enable_kv_cache_events,