bug fix for text request (vllm-project#52)

Co-authored-by: dengyunyang <[email protected]>

Author: Bounty-hunter

vllm/v1/core/sched/scheduler.py

@@ -317,7 +317,8 @@ def schedule(self) -> SchedulerOutput:
                 for i in encoder_inputs_to_schedule:
                     self.encoder_cache_manager.allocate(request, i)
                 encoder_budget = new_encoder_budget
-            if (self.ec_connector is not None and external_load_encoder_input):
+            if (self.ec_connector is not None and request.has_encoder_inputs
+                    and external_load_encoder_input):
                 for i in external_load_encoder_input:
                     self.encoder_cache_manager.allocate(request, i)
                     self.ec_connector.update_state_after_alloc(request, i)
@@ -503,6 +504,7 @@ def schedule(self) -> SchedulerOutput:
                     encoder_budget = new_encoder_budget
                 # Allocate for external load encoder cache
                 if (self.ec_connector is not None
+                        and request.has_encoder_inputs
                         and external_load_encoder_input):
                     for i in external_load_encoder_input:
                         self.encoder_cache_manager.allocate(request, i)