fix

python/sglang/srt/managers/controller/tp_worker.py

@@ -611,6 +611,10 @@ def cache_filled_batch(self, batch: Batch):
             )
             req.prefix_indices, req.last_node = new_prefix_indices, new_last_node
 
+            if req is self.current_inflight_req:
+                # inflight request would get a new req idx
+                self.req_to_token_pool.free(int(req_pool_indices_cpu[i]))
+
     def forward_decode_batch(self, batch: Batch):
         # Check if decode out of memory
         if not batch.check_decode_mem():

python/sglang/srt/memory_pool.py

@@ -45,7 +45,7 @@ def alloc(self, need_size: int):
 
         return select_index
 
-    def free(self, free_index: int):
+    def free(self, free_index):
         self.mem_state[free_index] = True
         if isinstance(free_index, (int,)):
             self.can_use_mem_size += 1