[V1][Minor] Minor optimizations for update_from_output (#12454)

Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>

vllm/v1/core/scheduler.py

@@ -411,6 +411,10 @@ class Scheduler:
         num_scheduled_tokens = scheduler_output.num_scheduled_tokens
         new_running: List[Request] = []
         outputs: List[EngineCoreOutput] = []
+
+        # NOTE(woosuk): As len(self.running) can be up to 1K or more, the below
+        # loop can be a performance bottleneck. We should do our best to avoid
+        # expensive operations inside the loop.
         for request in self.running:
             req_id = request.request_id
             request.num_computed_tokens += num_scheduled_tokens[req_id]
@@ -421,6 +425,8 @@ class Scheduler:
 
             cached_encoder_input_ids = (
                 self.encoder_cache_manager.get_cached_input_ids(request))
+            # OPTIMIZATION: Avoid list(set) if the set is empty.
+            if cached_encoder_input_ids:
                 for input_id in list(cached_encoder_input_ids):
                     start_pos = request.mm_positions[input_id]["offset"]
                     num_tokens = request.mm_positions[input_id]["length"]