[V1] Support long_prefill_token_threshold in v1 scheduler (#15419)

Signed-off-by: Lu Fang <lufang@fb.com>
2025-03-25 14:22:26 -07:00
parent 6aa196c8dc
commit 082ab86f5f
4 changed files with 113 additions and 4 deletions
--- a/vllm/engine/arg_utils.py
+++ b/vllm/engine/arg_utils.py
@@ -1625,9 +1625,7 @@ class EngineArgs:
        if (self.max_num_partial_prefills
                != EngineArgs.max_num_partial_prefills
                or self.max_long_partial_prefills
-                != EngineArgs.max_long_partial_prefills
-                or self.long_prefill_token_threshold
-                != EngineArgs.long_prefill_token_threshold):
+                != EngineArgs.max_long_partial_prefills):
            _raise_or_fallback(feature_name="Concurrent Partial Prefill",
                               recommend_to_remove=False)
            return False