diff --git a/scripts/run-qwen3.5-35B-A3B-npu.sh b/scripts/run-qwen3.5-35B-A3B-npu.sh
index 4eb56515..f652c5cf 100644
--- a/scripts/run-qwen3.5-35B-A3B-npu.sh
+++ b/scripts/run-qwen3.5-35B-A3B-npu.sh
@@ -134,10 +134,9 @@ OPTIMIZER_ARGS=(
 
 VLLM_ARGS=(
    --rollout-num-gpus-per-engine 2
-   --vllm-gpu-memory-utilization 0.7
+   --vllm-gpu-memory-utilization 0.85
    --vllm-enable-sleep-mode
    --vllm-weight-sync-mode native
-   --vllm-enforce-eager
 )
 
 MISC_ARGS=(
diff --git a/tests/test_glm4.7_30B_A3B_npu.py b/tests/test_glm4.7_30B_A3B_npu.py
index e694052a..c46d87d1 100644
--- a/tests/test_glm4.7_30B_A3B_npu.py
+++ b/tests/test_glm4.7_30B_A3B_npu.py
@@ -44,7 +44,7 @@ def execute():
         "--num-rollout 2 "
         "--rollout-batch-size 4 "
         "--n-samples-per-prompt 4 "
-        "--rollout-max-response-len 2048 "
+        "--rollout-max-response-len 1024 "
         "--rollout-temperature 1 "
         "--global-batch-size 16 "
         "--balance-data "