diff --git a/python/sglang/srt/batch_overlap/two_batch_overlap.py b/python/sglang/srt/batch_overlap/two_batch_overlap.py
index e2840fee0..d167e65c7 100644
--- a/python/sglang/srt/batch_overlap/two_batch_overlap.py
+++ b/python/sglang/srt/batch_overlap/two_batch_overlap.py
@@ -653,6 +653,7 @@ class TboForwardBatchPreparer:
             "extend_seq_lens_cpu",
             "extend_logprob_start_lens_cpu",
             "lora_ids",
+            "rids",
         ]:
             old_value = getattr(batch, key)
             if old_value is None:
diff --git a/python/sglang/srt/server_args.py b/python/sglang/srt/server_args.py
index 065ce26ba..b3bc04f42 100644
--- a/python/sglang/srt/server_args.py
+++ b/python/sglang/srt/server_args.py
@@ -1048,7 +1048,7 @@ class ServerArgs:
                     reserved_mem += 6 * 1024
                 elif self.speculative_algorithm != "NGRAM":
                     # eagle draft models and cuda graphs
-                    reserved_mem += 2 * 1024
+                    reserved_mem += 4 * 1024
 
             self.mem_fraction_static = (
                 round((gpu_mem - reserved_mem) / gpu_mem, 3)