[minor fix] llama4 hybrid memory (#7950)

Author: Ying1123

python/sglang/srt/mem_cache/memory_pool.py

@@ -520,8 +520,13 @@ def __init__(
             self.layers_mapping[global_layer_id] = (swa_layer_id, True)
         self.full_to_swa_index_mapping: Optional[torch.Tensor] = None
 
+        k_size, v_size = self.get_kv_size_bytes()
+        self.mem_usage = (k_size + v_size) / GB
+
     def get_kv_size_bytes(self):
-        raise NotImplementedError
+        k_size, v_size = self.full_kv_pool.get_kv_size_bytes()
+        k_size_swa, v_size_swa = self.swa_kv_pool.get_kv_size_bytes()
+        return k_size + k_size_swa, v_size + v_size_swa
 
     def get_contiguous_buf_infos(self):
         full_kv_data_ptrs, full_kv_data_lens, full_kv_item_lens = (