[Bugfix] Guard for negative counter metrics to prevent crash (#10430)

Signed-off-by: Travis Johnson <tsjohnso@us.ibm.com>
2024-11-18 21:57:10 -07:00
parent 74f8c2cf5f
commit 272e31c0bd
2 changed files with 6 additions and 1 deletions
--- a/vllm/engine/llm_engine.py
+++ b/vllm/engine/llm_engine.py
@@ -1716,7 +1716,7 @@ class LLMEngine:
            # not counted (to avoid double counting)
            actual_num_batched_tokens = scheduler_outputs.num_batched_tokens  # type: ignore

-            num_generation_tokens_from_prefill_groups = 0.
+            num_generation_tokens_from_prefill_groups = 0
            # NOTE: if scheduler_outputs.num_prefill_groups > 0 and
            # the len of scheduler_outputs.scheduled_seq_groups is !=
            # scheduler_outputs.num_prefill_groups, this means that