diff --git a/.github/workflows/vllm_ascend_test_pd.yaml b/.github/workflows/vllm_ascend_test_pd.yaml
index 932b3e5..a824f60 100644
--- a/.github/workflows/vllm_ascend_test_pd.yaml
+++ b/.github/workflows/vllm_ascend_test_pd.yaml
@@ -41,7 +41,11 @@ jobs:
     if: ${{ contains(github.event.pull_request.labels.*.name, 'pd-test') && contains(github.event.pull_request.labels.*.name, 'ready-for-test') || github.event_name == 'schedule' }}
     strategy:
       matrix:
-        vllm_verison: [main, v0.9.1]
+        vllm_verison: [
+            # revert me when V1 disaggregation prefill is merged in main
+            # main, 
+            v0.9.1
+          ]
     name: vLLM Ascend prefilling decoding disaggregation test
     runs-on: linux-arm64-npu-static-8
 
diff --git a/tests/e2e/pd_disaggreate/setup_pd.sh b/tests/e2e/pd_disaggreate/setup_pd.sh
index 675bee4..c15f109 100644
--- a/tests/e2e/pd_disaggreate/setup_pd.sh
+++ b/tests/e2e/pd_disaggreate/setup_pd.sh
@@ -66,6 +66,7 @@ function run_prefill_instance() {
   --served-model-name Deepseek \
   --max-model-len 2000 \
   --trust-remote-code \
+  --enforce-eager \
   --kv-transfer-config "$KV_CONFIG"
 }
 
@@ -119,6 +120,7 @@ function run_decode_instance() {
     --max-num-batched-tokens 2000 \
     --trust-remote-code \
     --gpu-memory-utilization 0.9 \
+    --enforce-eager \
     --kv-transfer-config "$KV_CONFIG"
 }