Disable chunked prefill and/or prefix caching when MLA is enabled (#1…

…2642) From @mgoin in #12638 I cannot push to that branch, therefore a new PR to unblock release. --------- Signed-off-by: mgoin <[email protected]> Signed-off-by: simon-mo <[email protected]> Co-authored-by: mgoin <[email protected]>
vllm-project · Feb 1, 2025 · 4f4d427 · 4f4d427
1 parent 1e36983
commit 4f4d427
Showing 1 changed file with 10 additions and 0 deletions.
diff --git a/vllm/config.py b/vllm/config.py
@@ -3252,6 +3252,16 @@ def __post_init__(self):
 
         current_platform.check_and_update_config(self)
 
+        # If MLA is enabled, force disable chunked prefill and prefix caching
+        if self.model_config and self.model_config.use_mla:
+            logger.info("MLA is enabled; forcing chunked prefill and prefix "
+                        "caching to be disabled.")
+            self.scheduler_config.enable_chunked_prefill = False
+            self.scheduler_config.chunked_prefill_enabled = False
+
+            if self.cache_config is not None:
+                self.cache_config.enable_prefix_caching = False
+
         if not self.instance_id:
             self.instance_id = random_uuid()[:5]