diff --git a/vllm/v1/core/scheduler.py b/vllm/v1/core/scheduler.py index d14819eb720fb..8ded5e5787133 100644 --- a/vllm/v1/core/scheduler.py +++ b/vllm/v1/core/scheduler.py @@ -212,7 +212,6 @@ class Scheduler: num_computed_tokens -= self.block_size num_new_tokens = self.block_size computed_blocks.pop() - num_new_tokens = min(num_new_tokens, token_budget) assert num_new_tokens > 0