diff --git a/scripts/coverage_run.sh b/scripts/coverage_run.sh
index b5c6578a5..3952f43b0 100644
--- a/scripts/coverage_run.sh
+++ b/scripts/coverage_run.sh
@@ -1,104 +1,85 @@
 #!/bin/bash
 DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
-echo "$DIR"
-
 run_path="$DIR/../tests/"
-cd ${run_path}
-ls
-
-exclude=("ci_use" "ce")
-for d in */ ; do
-  dir_name="${d%/}"
-  if [[ -d "$dir_name" ]]; then
-    skip=false
-    for ex in "${exclude[@]}"; do
-      if [[ "$dir_name" == "$ex" ]]; then
-        skip=true
-        break
-      fi
-    done
-    if ! $skip; then
-      dirs+=("$dir_name")
-    fi
-  fi
-done
+export PYTEST_INI="$DIR/../tests/pytest.ini"
+cd "$run_path" || exit 1
 
 failed_tests_file="failed_tests.log"
 > "$failed_tests_file"
-disabled_tests=(
-  layers/test_append_attention.py
-  layers/test_attention.py
-  operators/test_rejection_top_p_sampling.py
-  operators/test_perchannel_gemm.py
-  operators/test_scaled_gemm_f8_i4_f16.py
-  operators/test_topp_sampling.py
-  operators/test_stop_generation.py
-  operators/test_air_topp_sampling.py
-  operators/test_fused_moe.py
-  operators/test_stop_generation_multi_ends.py
-  graph_optimization/test_cuda_graph.py
+
+##################################
+# 执行特殊单测case(不符合unittest/pytest格式)
+##################################
+special_tests=(
+    "graph_optimization/test_cuda_graph_dynamic_subgraph.py"
+    "graph_optimization/test_cuda_graph_spec_decode.py"
+    "layers/test_quant_layer.py"
+    "operators/test_token_penalty.py"
+    "operators/test_split_fuse.py"
+    "operators/test_flash_mask_attn.py"
+    "operators/test_w4afp8_gemm.py"
+    "model_loader/test_load_ernie_vl.py"
+    "operators/test_tree_mask.py"
 )
-is_disabled() {
-  local test_file_rel="$1"
-  for disabled in "${disabled_tests[@]}"; do
-    if [[ "$test_file_rel" == "$disabled" ]]; then
-      return 0
+
+failed_special=0
+success_special=0
+
+for test_file in "${special_tests[@]}"; do
+    if [ -f "$test_file" ]; then
+        echo "Running special test: $test_file"
+        python -m coverage run --parallel-mode "$test_file"
+        status=$?
+        if [ "$status" -ne 0 ]; then
+            echo "$test_file" >> "$failed_tests_file"
+            failed_special=$((failed_special+1))
+        else
+            success_special=$((success_special+1))
+        fi
+    else
+        echo "Warning: $test_file not found"
+        failed_special=$((failed_special+1))
     fi
-  done
-  return 1
-}
-
-total=0
-fail=0
-success=0
-
-for dir in "${dirs[@]}"; do
-  if [ -d "$dir" ]; then
-    echo "Running tests in directory: $dir"
-    while IFS= read -r -d '' test_file; do
-      total=$((total + 1))
-      echo "Running $test_file"
-
-      if is_disabled "$test_file"; then
-        echo "Skipping disabled test: $test_file"
-        continue
-      fi
-      # TODO: Add a framework to manage unit test execution time
-      timeout 600 python -m coverage run "$test_file"
-      if [ $? -ne 0 ]; then
-        echo "$test_file" >> "$failed_tests_file"
-        fail=$((fail + 1))
-
-        PORTS=($FLASK_PORT $FD_API_PORT $FD_ENGINE_QUEUE_PORT $FD_METRICS_PORT)
-        echo "==== PORT CLEAN AFTER UT FAILED ===="
-
-        for port in "${PORTS[@]}"; do
-            PIDS=$(lsof -t -i :$port)
-            if [ -n "$PIDS" ]; then
-                echo "Port $port is occupied by PID(s): $PIDS"
-                echo "$PIDS" | xargs -r kill -9
-                echo "Port $port cleared"
-            else
-                echo "Port $port is free"
-            fi
-        done
-      else
-        success=$((success + 1))
-      fi
-    done < <(find "$dir" -type f -name "test_*.py" -print0)
-  else
-    echo "Directory $dir not found, skipping."
-  fi
 done
 
-echo "===================================="
-echo "Total test files run: $total"
-echo "Successful tests: $success"
-echo "Failed tests: $fail"
-echo "Failed test cases are listed in $failed_tests_file"
+##################################
+# 执行 pytest，每个文件单独跑
+##################################
+# 收集 pytest 文件
+TEST_FILES=$(python -m pytest --collect-only -q -c pytest.ini --disable-warnings | grep -Eo '^.*test_.*\.py' | sort | uniq)
 
-if [ "$fail" -ne 0 ]; then
-  echo "Failed test cases:"
-  cat "$failed_tests_file"
-  exit 8
+
+failed_pytest=0
+success_pytest=0
+
+for file in $TEST_FILES; do
+    echo "Running pytest file: $file"
+    python -m coverage run --parallel-mode -m pytest "$file"
+    status=$?
+    if [ "$status" -ne 0 ]; then
+        echo "$file" >> "$failed_tests_file"
+        failed_pytest=$((failed_pytest+1))
+    else
+        success_pytest=$((success_pytest+1))
+    fi
+done
+
+##################################
+# 汇总结果
+##################################
+echo "===================================="
+echo "Pytest total: $((failed_pytest + success_pytest))"
+echo "Pytest successful: $success_pytest"
+echo "Pytest failed: $failed_pytest"
+
+echo "Special tests total: ${#special_tests[@]}"
+echo "Special tests successful: $success_special"
+echo "Special tests failed: $failed_special"
+
+if [ "$failed_pytest" -ne 0 ] || [ "$failed_special" -ne 0 ]; then
+    echo "Failed test cases are listed in $failed_tests_file"
+    cat "$failed_tests_file"
+    exit 8
 fi
+
+echo "All tests passed!"
diff --git a/test/e2e/test_EB_Lite_serving.py b/tests/e2e/test_EB_Lite_serving.py
similarity index 98%
rename from test/e2e/test_EB_Lite_serving.py
rename to tests/e2e/test_EB_Lite_serving.py
index 70364f460..62f40b571 100644
--- a/test/e2e/test_EB_Lite_serving.py
+++ b/tests/e2e/test_EB_Lite_serving.py
@@ -14,6 +14,7 @@
 
 import os
 import re
+import shutil
 import signal
 import socket
 import subprocess
@@ -52,8 +53,14 @@ def kill_process_on_port(port: int):
     """
     try:
         output = subprocess.check_output(f"lsof -i:{port} -t", shell=True).decode().strip()
+        current_pid = os.getpid()
+        parent_pid = os.getppid()
         for pid in output.splitlines():
-            os.kill(int(pid), signal.SIGKILL)
+            pid = int(pid)
+            if pid in (current_pid, parent_pid):
+                print(f"Skip killing current process (pid={pid}) on port {port}")
+                continue
+            os.kill(pid, signal.SIGKILL)
             print(f"Killed process on port {port}, pid={pid}")
     except subprocess.CalledProcessError:
         pass
@@ -65,6 +72,7 @@ def clean_ports():
     """
     for port in PORTS_TO_CLEAN:
         kill_process_on_port(port)
+    time.sleep(2)
 
 
 @pytest.fixture(scope="session", autouse=True)
@@ -78,7 +86,9 @@ def setup_and_run_server():
     """
     print("Pre-test port cleanup...")
     clean_ports()
-
+    print("log dir clean ")
+    if os.path.exists("log") and os.path.isdir("log"):
+        shutil.rmtree("log")
     base_path = os.getenv("MODEL_PATH")
     if base_path:
         model_path = os.path.join(base_path, "ernie-4_5-21b-a3b-bf16-paddle")
diff --git a/test/e2e/test_EB_VL_Lite_serving.py b/tests/e2e/test_EB_VL_Lite_serving.py
similarity index 97%
rename from test/e2e/test_EB_VL_Lite_serving.py
rename to tests/e2e/test_EB_VL_Lite_serving.py
index df579c34d..86d18a6e1 100644
--- a/test/e2e/test_EB_VL_Lite_serving.py
+++ b/tests/e2e/test_EB_VL_Lite_serving.py
@@ -15,6 +15,7 @@
 import json
 import os
 import re
+import shutil
 import signal
 import socket
 import subprocess
@@ -53,8 +54,14 @@ def kill_process_on_port(port: int):
     """
     try:
         output = subprocess.check_output(f"lsof -i:{port} -t", shell=True).decode().strip()
+        current_pid = os.getpid()
+        parent_pid = os.getppid()
         for pid in output.splitlines():
-            os.kill(int(pid), signal.SIGKILL)
+            pid = int(pid)
+            if pid in (current_pid, parent_pid):
+                print(f"Skip killing current process (pid={pid}) on port {port}")
+                continue
+            os.kill(pid, signal.SIGKILL)
             print(f"Killed process on port {port}, pid={pid}")
     except subprocess.CalledProcessError:
         pass
@@ -66,6 +73,7 @@ def clean_ports():
     """
     for port in PORTS_TO_CLEAN:
         kill_process_on_port(port)
+    time.sleep(2)
 
 
 @pytest.fixture(scope="session", autouse=True)
@@ -79,6 +87,9 @@ def setup_and_run_server():
     """
     print("Pre-test port cleanup...")
     clean_ports()
+    print("log dir clean ")
+    if os.path.exists("log") and os.path.isdir("log"):
+        shutil.rmtree("log")
 
     base_path = os.getenv("MODEL_PATH")
     if base_path:
@@ -150,6 +161,7 @@ def setup_and_run_server():
     try:
         os.killpg(process.pid, signal.SIGTERM)
         print(f"API server (pid={process.pid}) terminated")
+        clean_ports()
     except Exception as e:
         print(f"Failed to terminate API server: {e}")
 
@@ -524,7 +536,8 @@ def test_chat_with_thinking(openai_client, capsys):
         stream=True,
         max_tokens=10,
     )
-    completion_tokens = reasoning_tokens = 1
+    completion_tokens = 1
+    reasoning_tokens = 0
     total_tokens = 0
     for chunk_id, chunk in enumerate(response):
         if chunk_id == 0:  # the first chunk is an extra chunk
diff --git a/test/e2e/test_Qwen2-7B-Instruct_serving.py b/tests/e2e/test_Qwen2-7B-Instruct_serving.py
similarity index 98%
rename from test/e2e/test_Qwen2-7B-Instruct_serving.py
rename to tests/e2e/test_Qwen2-7B-Instruct_serving.py
index 4b03a9835..eddde9fca 100644
--- a/test/e2e/test_Qwen2-7B-Instruct_serving.py
+++ b/tests/e2e/test_Qwen2-7B-Instruct_serving.py
@@ -16,6 +16,7 @@ import concurrent.futures
 import json
 import os
 import re
+import shutil
 import signal
 import socket
 import subprocess
@@ -55,8 +56,14 @@ def kill_process_on_port(port: int):
     """
     try:
         output = subprocess.check_output(f"lsof -i:{port} -t", shell=True).decode().strip()
+        current_pid = os.getpid()
+        parent_pid = os.getppid()
         for pid in output.splitlines():
-            os.kill(int(pid), signal.SIGKILL)
+            pid = int(pid)
+            if pid in (current_pid, parent_pid):
+                print(f"Skip killing current process (pid={pid}) on port {port}")
+                continue
+            os.kill(pid, signal.SIGKILL)
             print(f"Killed process on port {port}, pid={pid}")
     except subprocess.CalledProcessError:
         pass
@@ -68,6 +75,7 @@ def clean_ports():
     """
     for port in PORTS_TO_CLEAN:
         kill_process_on_port(port)
+    time.sleep(2)
 
 
 @pytest.fixture(scope="session", autouse=True)
@@ -82,6 +90,10 @@ def setup_and_run_server():
     print("Pre-test port cleanup...")
     clean_ports()
 
+    print("log dir clean ")
+    if os.path.exists("log") and os.path.isdir("log"):
+        shutil.rmtree("log")
+
     base_path = os.getenv("MODEL_PATH")
     if base_path:
         model_path = os.path.join(base_path, "Qwen2-7B-Instruct")
@@ -139,6 +151,7 @@ def setup_and_run_server():
     print("\n===== Post-test server cleanup... =====")
     try:
         os.killpg(process.pid, signal.SIGTERM)
+        clean_ports()
         print(f"API server (pid={process.pid}) terminated")
     except Exception as e:
         print(f"Failed to terminate API server: {e}")
diff --git a/test/logger/test_formatters.py b/tests/logger/test_formatters.py
similarity index 100%
rename from test/logger/test_formatters.py
rename to tests/logger/test_formatters.py
diff --git a/test/logger/test_handlers.py b/tests/logger/test_handlers.py
similarity index 100%
rename from test/logger/test_handlers.py
rename to tests/logger/test_handlers.py
diff --git a/test/logger/test_logger.py b/tests/logger/test_logger.py
similarity index 100%
rename from test/logger/test_logger.py
rename to tests/logger/test_logger.py
diff --git a/test/logger/test_setup_logging.py b/tests/logger/test_setup_logging.py
similarity index 100%
rename from test/logger/test_setup_logging.py
rename to tests/logger/test_setup_logging.py
diff --git a/tests/pytest.ini b/tests/pytest.ini
new file mode 100644
index 000000000..49cec43ea
--- /dev/null
+++ b/tests/pytest.ini
@@ -0,0 +1,27 @@
+[pytest]
+# 跳过目录
+addopts =
+    --ignore=ci_use
+    --ignore=ce
+    --ignore=layers/test_append_attention.py
+    --ignore=layers/test_attention.py
+    --ignore=operators/test_rejection_top_p_sampling.py
+    --ignore=operators/test_perchannel_gemm.py
+    --ignore=operators/test_scaled_gemm_f8_i4_f16.py
+    --ignore=operators/test_topp_sampling.py
+    --ignore=operators/test_stop_generation.py
+    --ignore=operators/test_air_topp_sampling.py
+    --ignore=operators/test_fused_moe.py
+    --ignore=operators/test_stop_generation_multi_ends.py
+    --ignore=graph_optimization/test_cuda_graph.py
+    --ignore=graph_optimization/test_cuda_graph_dynamic_subgraph.py
+    --ignore=graph_optimization/test_cuda_graph_spec_decode
+    --ignore=layers/test_quant_layer.py
+    --ignore=operators/test_token_penalty.py
+    --ignore=operators/test_split_fuse.py
+    --ignore=operators/test_flash_mask_attn.py
+    --ignore=operators/test_w4afp8_gemm.py
+    --ignore=operators/test_tree_mask.py
+
+# 输出更详细的结果
+console_output_style = progress