Fix schedule error in splitwise deployment (#5149)

fastdeploy/engine/sched/resource_manager_v1.py

@@ -514,6 +514,8 @@ class ResourceManagerV1(ResourceManager):
             error_reqs: list[tuple[str, str]] = []
             token_budget = self.config.scheduler_config.max_num_batched_tokens
 
+            self.check_and_free_block_tables()
+
             # First, schedule the RUNNING requests.
             req_index = 0
             num_decoding_req_nums = 0

fastdeploy/scheduler/local_scheduler.py

@@ -285,7 +285,7 @@ class LocalScheduler:
                         if short_partial_requests + long_partial_requests > self.max_num_partial_prefills:
                             break
                     else:
-                        if current_prefill_tokens > max_num_batched_tokens:
+                        if current_prefill_tokens > max_num_batched_tokens and len(requests) > 0:
                             break
                 requests.append(request.raw)