[NPU] Increase the retry count to 3 for the GSM8K. (#34103)

Co-authored-by: Even Zhou <even.y.zhou@outlook.com>
Co-authored-by: sglang-npu-bot <sglangnpu@163.com>

.github/workflows/_npu-single-node-test-stage.yml

@@ -260,7 +260,7 @@ jobs:
           echo "Running test: ${tc_name}"
           test_exit_code=0
           cd test
-          ${PYTHON_FOR_SGLANG} -u run_suite.py --hw npu --suite ${test_suite} --timeout-per-file 3600 2>&1 | tee /tmp/test_output.log || test_exit_code=$?
+          ${PYTHON_FOR_SGLANG} -u run_suite.py --hw npu --suite ${test_suite} --timeout-per-file 7200 2>&1 | tee /tmp/test_output.log || test_exit_code=$?
 
           echo "Finished test: ${tc_name}"
 

python/sglang/test/ascend/e2e/test_npu_accuracy_utils.py

@@ -61,6 +61,7 @@ DATASET_FLUCTUATION = {
     "aime25": 2,
     "aime26": 2,
     "gpqa_diamond": 5,
+    "gsm8k": 3,
 }
 
 MAX_RETRY_COUNT = 3
@@ -291,7 +292,6 @@ class TestNpuAccuracyTestCaseBase(CustomTestCase):
     other_args = None
     server_timeout = DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH
     envs = None
-    max_attempts = 2
     n_runs = 3
     accuracy = 0.1
     test_type = "accuracy"
@@ -549,7 +549,6 @@ class TestNpuAccuracyMultiNodePdMixTestCaseBase(CustomTestCase):
     other_args = None
     server_timeout = DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH
     envs = None
-    max_attempts = 2
     accuracy = 0.1
 
     @classmethod
@@ -651,7 +650,6 @@ class TestNpuAccuracyMultiNodePdSepTestCaseBase(CustomTestCase):
     eval_type = "openai_api"
     other_args = None
     server_timeout = DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH
-    max_attempts = 2
     accuracy = 0.1
 
     @classmethod