diff --git a/scripts/performance/benchmark.sh b/scripts/performance/benchmark.sh
new file mode 100644
index 0000000..2948f10
--- /dev/null
+++ b/scripts/performance/benchmark.sh
@@ -0,0 +1,55 @@
+#!/bin/bash
+set -euo pipefail
+
+# step 1: please pip install pandas openpyxl
+# step 3: change the variables as below
+# step 4: bash benchmark.sh
+
+##################### need changed #######################
+MODEL_PATH="/data/models/deepseek-ai/DeepSeek-V3.2-Exp/"
+ISL_LIST=(1023 2047 4093 6143)
+CONC_LIST=(8 4 1)
+OSL=1024
+PORT=8888
+##########################################################
+
+LOG_FILE="result.txt"
+# clean old file
+if [ -f "$LOG_FILE" ]; then
+    rm "$LOG_FILE"
+fi
+
+# health check
+curl -sf "http://localhost:$PORT/health" > /dev/null || {
+    echo "ERROR: Server not running on port $PORT at vllm backend"
+    exit 1
+}
+
+for ISL in "${ISL_LIST[@]}"; do
+
+    for CONC in "${CONC_LIST[@]}"; do
+    echo "=========================================" | tee -a "$LOG_FILE"
+    echo "Start Test ISL=$ISL, OSL=$OSL, CONC=$CONC" | tee -a "$LOG_FILE"
+    echo "=========================================" | tee -a "$LOG_FILE"
+        
+        python -m atom.benchmarks.benchmark_serving \
+            --model="$MODEL_PATH" \
+            --backend=vllm \
+            --base-url="http://localhost:$PORT" \
+            --dataset-name=random \
+            --random-input-len="$ISL" \
+            --random-output-len="$OSL" \
+            --random-range-ratio 0.8 \
+            --num-prompts=$(( CONC * 10)) \
+            --max-concurrency="$CONC" \
+            --request-rate=inf \
+            --ignore-eos \
+            --percentile-metrics="ttft,tpot,itl,e2el"  2>&1 | tee -a "$LOG_FILE"
+        
+        echo -e "\n" | tee -a "$LOG_FILE"
+    done
+done
+
+if [ -f "$LOG_FILE" ]; then
+    python save_csv.py $LOG_FILE
+fi
\ No newline at end of file
diff --git a/scripts/performance/save_csv.py b/scripts/performance/save_csv.py
new file mode 100644
index 0000000..198e5f8
--- /dev/null
+++ b/scripts/performance/save_csv.py
@@ -0,0 +1,248 @@
+import os
+import sys
+import re
+import argparse
+from openpyxl import Workbook
+from openpyxl.styles import PatternFill, Font, Alignment, Border, Side
+from openpyxl.utils import get_column_letter
+
+def parse_log_file(log_file_path):
+    """Parse log file and extract key metrics"""
+    results = []
+    
+    with open(log_file_path, 'r', encoding='utf-8') as f:
+        content = f.read()
+    
+    test_blocks = re.findall(
+        r'Start Test ISL=(\d+), OSL=(\d+), CONC=(\d+)(.*?)==================================================',
+        content,
+        re.DOTALL
+    )
+    
+    for isl, osl, conc, block in test_blocks:
+        mean_ttft = re.search(r'Mean TTFT \(ms\):\s+([\d.]+)', block)
+        mean_tpot = re.search(r'Mean TPOT \(ms\):\s+([\d.]+)', block)
+        total_throughput = re.search(r'Total Token throughput \(tok/s\):\s+([\d.]+)', block)
+        
+        if mean_ttft and mean_tpot and total_throughput:
+            results.append({
+                'ISL': int(isl),
+                'OSL': int(osl),
+                'CONC': int(conc),
+                'Mean TTFT (ms)': float(mean_ttft.group(1)),
+                'Mean TPOT (ms)': float(mean_tpot.group(1)),
+                'Total Throughput (tok/s)': float(total_throughput.group(1))
+            })
+    
+    return results
+
+def create_excel_repeating_format(results, output_file):
+    """Create Excel file with repeating format: each ISL/OSL combination repeats display, ensure ISL row has complete border"""
+    
+    # Sort by ISL and CONC
+    results.sort(key=lambda x: (x['ISL'], -x['CONC']))
+    
+    # Create new workbook
+    wb = Workbook()
+    ws = wb.active
+    ws.title = "Benchmark Results"
+    
+    # Define border style - ensure all borders are set
+    full_border = Border(
+        left=Side(style='thin'),
+        right=Side(style='thin'),
+        top=Side(style='thin'),
+        bottom=Side(style='thin')
+    )
+    
+    # Define fill styles
+    header_fill = PatternFill(start_color="D3D3D3", end_color="D3D3D3", fill_type="solid")
+    isl_header_fills = [
+        PatternFill(start_color="F8D7DA", end_color="F8D7DA", fill_type="solid"),
+        PatternFill(start_color="D4EDDA", end_color="D4EDDA", fill_type="solid"),
+        PatternFill(start_color="FFF3CD", end_color="FFF3CD", fill_type="solid"),
+        PatternFill(start_color="D1ECF1", end_color="D1ECF1", fill_type="solid")
+    ]
+    
+    # Define font styles
+    header_font = Font(bold=True, size=11, color="000000")
+    isl_title_font = Font(bold=True, size=12, color="000000")
+    data_font = Font(size=10)
+    center_alignment = Alignment(horizontal='center', vertical='center')
+    
+    current_row = 1
+    grouped_results = {}
+    for result in results:
+        key = (result['ISL'], result['OSL'])
+        if key not in grouped_results:
+            grouped_results[key] = []
+        grouped_results[key].append(result)
+    
+    # Sort by ISL
+    sorted_keys = sorted(grouped_results.keys())
+    
+    # Create independent 4-row area for each ISL/OSL combination
+    for idx, (isl, osl) in enumerate(sorted_keys):
+        data_list = grouped_results[(isl, osl)]
+        data_list.sort(key=lambda x: -x['CONC'])  # Sort by CONC descending
+        
+        # Row 1: ISL/OSL title row (merge 4 cells) - ensure complete border
+        ws.merge_cells(start_row=current_row, start_column=1, end_row=current_row, end_column=4)
+        
+        cell = ws.cell(row=current_row, column=1, value=f"ISL={isl}, OSL={osl}")
+        cell.fill = isl_header_fills[idx % len(isl_header_fills)]
+        cell.font = isl_title_font
+        cell.alignment = center_alignment
+        cell.border = full_border  # Ensure complete border
+        
+        # Set border for each cell of merged cells individually (ensure display)
+        for col in range(1, 5):
+            ws.cell(row=current_row, column=col).border = full_border
+        
+        current_row += 1
+        
+        # Row 2: Header row
+        for col, header in enumerate(["Concurrency", "Mean TTFT (ms)", "Mean TPOT (ms)", "Total Throughput (tok/s)"], 1):
+            cell = ws.cell(row=current_row, column=col, value=header)
+            cell.fill = header_fill
+            cell.font = header_font
+            cell.alignment = center_alignment
+            cell.border = full_border
+        
+        current_row += 1
+        
+        # Row 3-4: Data rows (maximum 2 concurrency values)
+        for i, data in enumerate(data_list):  # Take first 2 concurrency values
+            # Concurrency
+            cell = ws.cell(row=current_row, column=1, value=data['CONC'])
+            cell.font = data_font
+            cell.alignment = center_alignment
+            cell.border = full_border
+            
+            # Mean TTFT
+            cell = ws.cell(row=current_row, column=2, value=data['Mean TTFT (ms)'])
+            cell.font = data_font
+            cell.alignment = center_alignment
+            cell.border = full_border
+            
+            # Mean TPOT
+            cell = ws.cell(row=current_row, column=3, value=data['Mean TPOT (ms)'])
+            cell.font = data_font
+            cell.alignment = center_alignment
+            cell.border = full_border
+            
+            # Total Throughput
+            cell = ws.cell(row=current_row, column=4, value=data['Total Throughput (tok/s)'])
+            cell.font = data_font
+            cell.alignment = center_alignment
+            cell.border = full_border
+            
+            current_row += 1
+        
+        # Add empty row to separate different ISL/OSL combinations (no border)
+        current_row += 1
+    
+    # Remove last empty row
+    if current_row > 1:
+        current_row -= 1
+    
+    # Auto adjust column width
+    for col_idx in range(1, 5):
+        column_letter = get_column_letter(col_idx)
+        
+        if col_idx == 1:  # Concurrency column
+            ws.column_dimensions[column_letter].width = 12
+        elif col_idx == 2 or col_idx == 3:  # Mean TTFT/TPOT columns
+            ws.column_dimensions[column_letter].width = 15
+        else:  # Total Throughput column
+            ws.column_dimensions[column_letter].width = 22
+    
+    # Save file
+    wb.save(output_file)
+    return output_file
+
+def print_csv_repeating_format(results):
+    """Print data in CSV table format (repeating format)"""
+    
+    # Group by ISL
+    grouped_results = {}
+    for result in results:
+        key = (result['ISL'], result['OSL'])
+        if key not in grouped_results:
+            grouped_results[key] = []
+        grouped_results[key].append(result)
+    
+    # Sort by ISL
+    sorted_keys = sorted(grouped_results.keys())
+    
+    print("\n" + "=" * 60)
+    print("DEEPSEEK-V3.2 BENCHMARK RESULTS")
+    print("=" * 60)
+    
+    # Print independent table for each ISL/OSL combination
+    for idx, (isl, osl) in enumerate(sorted_keys):
+        data_list = grouped_results[(isl, osl)]
+        data_list.sort(key=lambda x: -x['CONC'])  # Sort by CONC descending
+        
+        print(f"\n┌{'ISL=' + str(isl) + ', OSL=' + str(osl):^52}┐")
+        print("├────────────────────────────────────────────────────┤")
+        print("│ Concurrency │ Mean TTFT │ Mean TPOT │  Throughput  │")
+        print("├─────────────┼───────────┼───────────┼──────────────┤")
+        
+        # Print maximum 2 rows of data
+        for i, data in enumerate(data_list):
+            print(f"│ {data['CONC']:^11} │ {data['Mean TTFT (ms)']:>9.2f} │ {data['Mean TPOT (ms)']:>9.2f} │ {data['Total Throughput (tok/s)']:>12.2f} │")
+        
+        # If data has less than 2 rows, fill with empty rows
+        while len(data_list) < 2 and i < 1:
+            print("│             │           │           │              │")
+            i += 1
+        
+        print("└─────────────┴───────────┴───────────┴──────────────┘")
+    
+    print("=" * 60)
+
+def main():
+    # Input file path
+    parser = argparse.ArgumentParser(
+        description='benchmark infomation',
+    )
+    
+    # Add parameter
+    parser.add_argument(
+        'input_file',
+        nargs='?',  # Optional parameter
+        default='result.txt',
+        help='the path of input file, default:result.txt.'
+    )
+    
+    args = parser.parse_args()
+    log_file_path = args.input_file
+    
+    if not os.path.isfile(log_file_path):
+        print(f"file not exist - {log_file_path}")
+        sys.exit(1)
+
+    output_file = "benchmark.xlsx"
+    
+    try:
+        # Parse log file
+        results = parse_log_file(log_file_path)
+        
+        if results:
+            # Print data in CSV table format
+            print_csv_repeating_format(results)
+            
+            # Create Excel file
+            create_excel_repeating_format(results, output_file)
+            
+        else:
+            print("No valid benchmark results found")
+            
+    except FileNotFoundError:
+        print(f"Error: cannot find log file {log_file_path}")
+    except Exception as e:
+        print(f"Error during processing: {e}")
+
+if __name__ == "__main__":
+    main()
\ No newline at end of file