[multi-step] Raise error if not using async engine (#7703)

91f4522c · William Lin · GitHub · 1b32e026 · 91f4522c
Unverified Commit 91f4522c authored Aug 21, 2024 by William Lin Committed by GitHub Aug 21, 2024
Show whitespace changes
Inline Side-by-side

Showing with 5 additions and 0 deletions

vllm/engine/llm_engine.py vllm/engine/llm_engine.py +5 -0

No files found.
--- a/vllm/engine/llm_engine.py
+++ b/vllm/engine/llm_engine.py
@@ -1302,6 +1302,11 @@ class LLMEngine:
            raise NotImplementedError(
                "Pipeline parallelism is only supported through AsyncLLMEngine "
                "as performance will be severely degraded otherwise.")
+        if self.scheduler_config.num_scheduler_steps > 1:
+            raise NotImplementedError(
+                "Multiple scheduler steps (multi-step) are only supported "
+                "through AsyncLLMEngine. ")
        seq_group_metadata_list, scheduler_outputs = self.scheduler[
            0].schedule()