diff --git a/.github/workflows/_npu-pr-test-stage.yml b/.github/workflows/_npu-pr-test-stage.yml
index 2171550d4..42cd4b0e4 100644
--- a/.github/workflows/_npu-pr-test-stage.yml
+++ b/.github/workflows/_npu-pr-test-stage.yml
@@ -255,6 +255,9 @@ jobs:
             install_pkg "${pkg}"
           done
 
+          # Install sgl-eval (CLI used by run_eval.py's `sgl-eval run ...`) if missing.
+          command -v sgl-eval >/dev/null 2>&1 || install_pkg "sgl-eval==0.1.0"
+
           cd test
           # Append `--nightly` and `--continue-on-error` when the job belongs to the nightly pipeline.
           # --nightly: Collect test cases registered for nightly runs.
diff --git a/test/registered/npu/basic_function/dp_attn/test_npu_dp_attention.py b/test/registered/npu/basic_function/dp_attn/test_npu_dp_attention.py
index a830ea2b6..dc3489181 100644
--- a/test/registered/npu/basic_function/dp_attn/test_npu_dp_attention.py
+++ b/test/registered/npu/basic_function/dp_attn/test_npu_dp_attention.py
@@ -24,7 +24,7 @@ from sglang.test.test_utils import (
 )
 
 register_npu_ci(est_time=400, suite="base-b-test-4-npu-a3")
-register_npu_ci(est_time=400, suite="nightly-4-npu-a3", nightly=True)
+register_npu_ci(est_time=2400, suite="nightly-4-npu-a3", nightly=True)
 
 
 class TestDPAttentionDP2TP2(
@@ -74,6 +74,8 @@ class TestDPAttentionMixedChunk(
     CustomTestCase,
     NPUGSM8KMixin,
 ):
+    # Use full GSM8K dataset to avoid sampling variance.
+    gsm8k_num_examples = 1319
     gsm8k_accuracy_thres = 0.34
 
     @classmethod