diff --git a/.github/actions/inductor-xpu-e2e-test/action.yml b/.github/actions/inductor-xpu-e2e-test/action.yml
index f9084fe9b..ccbabf407 100644
--- a/.github/actions/inductor-xpu-e2e-test/action.yml
+++ b/.github/actions/inductor-xpu-e2e-test/action.yml
@@ -81,6 +81,7 @@ runs:
         source .github/scripts/env.sh
         cp .github/scripts/inductor_xpu_test.sh ../pytorch
         cd ../pytorch
+
         # check param
         function contains() {
             contains_status="echo 'Start $2 ...'"
@@ -136,8 +137,7 @@ runs:
           done
         done
 
-    - name: Summary E2E Accuracy Test (${{ inputs.suite }} ${{ inputs.dt }} ${{ inputs.mode }} ${{ inputs.scenario }})
-      if: contains(inputs.scenario, 'accuracy')
+    - name: Summary E2E Test (${{ inputs.suite }} ${{ inputs.dt }} ${{ inputs.mode }} ${{ inputs.scenario }})
       shell: bash
       env:
         HUGGING_FACE_HUB_TOKEN: ${{ inputs.hf_token }}
@@ -150,60 +150,15 @@ runs:
           cat $var >> inductor_log/summary_accuracy.csv
         done
 
-        # check param
-        function contains() {
-            contains_status="echo 'Start $2 ...'"
-            {
-              [[ $1 =~ (^|,)$2($|,) ]]
-            } || {
-              echo "[Warning] $2 is not suppotted type! Skipped!"
-              contains_status="continue"
-            }
-        }
-        #Accuracy summary 
         source activate e2e_ci
         cd ${{ github.workspace }}
-        cp .github/scripts/inductor_accuracy_summary.py ../pytorch
+        cp .github/scripts/inductor_summary.py ../pytorch
         cd ../pytorch
         pip install styleFrame scipy pandas
         set -xe
-        for suite in $(echo ${{ inputs.suite }} |sed 's/,/ /g')
-        do
-          contains "huggingface,timm_models,torchbench" $suite
-          $contains_status
-          dt=$(echo ${{ inputs.dt }} |sed 's/,/ /g')
-          mode=$(echo ${{ inputs.mode }} |sed 's/,/ /g')
-          python inductor_accuracy_summary.py -p ${dt} -s ${suite} -m ${mode}
-        done
-
-    - name: Summary E2E Performance Test (${{ inputs.suite }} ${{ inputs.dt }} ${{ inputs.mode }} ${{ inputs.scenario }})
-      if: contains(inputs.scenario, 'performance')
-      shell: bash
-      env:
-        HUGGING_FACE_HUB_TOKEN: ${{ inputs.hf_token }}
-      run: |
-        # check param
-        function contains() {
-            contains_status="echo 'Start $2 ...'"
-            {
-              [[ $1 =~ (^|,)$2($|,) ]]
-            } || {
-              echo "[Warning] $2 is not suppotted type! Skipped!"
-              contains_status="continue"
-            }
-        }
-
-        source activate e2e_ci
-        cp .github/scripts/inductor_perf_summary.py ../pytorch
-        cd ../pytorch
-        source /opt/intel/oneapi/pytorch-gpu-dev-0.5/oneapi-vars.sh
-        pip install styleFrame scipy pandas
-        set -xe
-        for suite in $(echo ${{ inputs.suite }} |sed 's/,/ /g')
-        do
-          contains "huggingface,timm_models,torchbench" $suite
-          $contains_status
-          dt=$(echo ${{ inputs.dt }} |sed 's/,/ /g')
-          mode=$(echo ${{ inputs.mode }} |sed 's/,/ /g')
-          python inductor_perf_summary.py -p ${dt} -s ${suite} -m ${mode}
-        done
+        dt=$(echo ${{ inputs.dt }} |sed 's/,/ /g')
+        mode=$(echo ${{ inputs.mode }} |sed 's/,/ /g')
+        suite=$(echo ${{ inputs.suite }} |sed 's/,/ /g')
+        scenario=$(echo ${{ inputs.scenario }} |sed 's/,/ /g')
+        python inductor_summary.py -p ${dt} -s ${suite} -m ${mode} -sc ${scenario}
+        
diff --git a/.github/scripts/inductor_accuracy_summary.py b/.github/scripts/inductor_accuracy_summary.py
deleted file mode 100644
index 131df9831..000000000
--- a/.github/scripts/inductor_accuracy_summary.py
+++ /dev/null
@@ -1,350 +0,0 @@
-import argparse
-import pandas as pd
-from scipy.stats import gmean
-from styleframe import StyleFrame, Styler, utils
-
-parser = argparse.ArgumentParser(description="Generate report")
-parser.add_argument('-s', '--suite', default='huggingface', choices=["torchbench", "huggingface", "timm_models"], type=str, help='model suite name')
-parser.add_argument('-p', '--precision', default=["amp_fp16", "float32"], nargs='*', type=str, help='precision')
-parser.add_argument('-r', '--reference', type=str, help='reference log files')
-parser.add_argument('-m', '--mode', default=["inference", "training"], nargs='*', type=str, help='mode name')
-args = parser.parse_args()
-
-passrate_values = {}
-
-failure_style = Styler(bg_color='#FF0000', font_color=utils.colors.black)
-regression_style = Styler(bg_color='#F0E68C', font_color=utils.colors.red)
-improve_style = Styler(bg_color='#00FF00', font_color=utils.colors.black)
-
-
-def percentage(part, whole, decimals=2):
-    if whole == 0:
-        return 0
-    return round(100 * float(part) / float(whole), decimals)
-
-def get_passing_entries(df, column_name):
-    return df[df[column_name].notnull()]
-
-def caculate_passrate(df, key_word):
-    # some models may be new for reference
-    df = get_passing_entries(df, key_word)
-    total = len(df.index)
-    passing = df[df[key_word].fillna('').str.contains('pass')][key_word].count()
-    perc = int(percentage(passing, total, decimals=0))
-    return f"{perc}%, {passing}/{total}"
-
-def get_acc_csv(precision, mode):
-    target_path = 'inductor_log/' + args.suite + '/' + precision + '/inductor_' + args.suite + '_' + precision + '_' + mode + '_xpu_accuracy.csv'
-    target_ori_data = pd.read_csv(target_path)
-    target_data = target_ori_data.copy()
-    target_data.sort_values(by=['name'])
-    
-    if args.reference is not None:
-        reference_file_path = args.reference + '/inductor_log/' + args.suite + '/' + precision + '/inductor_' + args.suite + '_' + precision + '_' + mode + '_xpu_accuracy.csv'
-        reference_ori_data = pd.read_csv(reference_file_path)
-        reference_data = reference_ori_data.copy()
-        reference_data.sort_values(by=['name'])
-        data = pd.merge(target_data,reference_data,on=['name'],how= 'outer')
-        return data
-    else:
-        return target_data
-
-def process(input, precision, mode):
-    global passrate_values
-    if input is not None:
-        if args.reference is None:
-            data_new = input[['name', 'batch_size', 'accuracy']].rename(columns={'name': 'name', 'batch_size': 'batch_size', 'accuracy': 'accuracy'})
-            passrate_values['target_' + str(precision) + '_' + str(mode)] = caculate_passrate(data_new, 'accuracy')
-            data = StyleFrame({'name': list(data_new['name']),
-                            'batch_size': list(data_new['batch_size']),
-                            'accuracy': list(data_new['accuracy'])})
-            data.set_column_width(1, 10)
-            data.set_column_width(2, 18)
-            data.set_column_width(3, 18)
-            data.set_row_height(rows=data.row_indexes, height=15)
-        else:
-            data_new=input[['name','batch_size_x','accuracy_x']].rename(columns={'name':'name','batch_size_x':'batch_size_new','accuracy_x':'accuracy_new'})
-            passrate_values['target_' + str(precision) + '_' + str(mode)] = caculate_passrate(data_new, 'accuracy_new')        
-            data_old=input[['batch_size_y','accuracy_y']].rename(columns={'batch_size_y':'batch_size_old','accuracy_y':'accuracy_old'}) 
-            passrate_values['reference_' + str(precision) + '_' + str(mode)] = caculate_passrate(data_old, 'accuracy_old')
-            data_comp = pd.DataFrame((data_new['accuracy_new'] != 'pass') & (data_old['accuracy_old'] == 'pass'),columns=['Accuracy regression'])
-            combined_data = pd.DataFrame({
-                'name': list(data_new['name']),
-                'batch_size_new': list(data_new['batch_size_new']),
-                'accuracy_new': list(data_new['accuracy_new']),
-                'batch_size_old': list(data_old['batch_size_old']),
-                'accuracy_old': list(data_old['accuracy_old']),
-                'Accuracy regression': list(data_comp['Accuracy regression'])
-                })   
-            data = StyleFrame(combined_data)
-            data.set_column_width(1, 10)
-            data.set_column_width(2, 18) 
-            data.set_column_width(3, 18) 
-            data.set_column_width(4, 18)
-            data.set_column_width(5, 15)
-            data.set_column_width(6, 20)
-            data.apply_style_by_indexes(indexes_to_style=data[(data['Accuracy regression'] == 'regression')],styler_obj=regression_style)
-            data.set_row_height(rows=data.row_indexes, height=15)        
-        return data
-    else:
-        return pd.DataFrame()
-
-def update_details(precision, mode, excel):
-    h = {"A": 'Model suite', "B": '', "C": "target", "D": '', "E": args.reference, "F": '', "G": 'Result Comp'}
-    if args.reference is None:
-        h = {"A": 'Model suite', "B": '', "C": "target", "D": ''}
-    head = StyleFrame(pd.DataFrame(h, index=[0]))
-    head.set_column_width(1, 15)
-    head.set_row_height(rows=[1], height=15)
-
-    head.to_excel(excel_writer=excel, sheet_name=precision + '_' + mode, index=False, startrow=0, header=False)
-    target_raw_data = get_acc_csv(precision, mode)
-    target_data = process(target_raw_data, precision, mode)
-    target_data.to_excel(excel_writer=excel, sheet_name=precision + '_' + mode, index=False, startrow=1, startcol=1)
-
-def update_summary(excel):
-    data = {
-        'Test Secnario': ['AMP_BF16 Inference', 'AMP_BF16 Training', 'AMP_FP16 Inference', 'AMP_FP16 Training', 'BF16 Inference', 'BF16 Training', 'FP16 Inference', 'FP16 Training', 'FP32 Inference', 'FP32 Training'],
-        'Comp Item': ['Pass Rate',  'Pass Rate',  'Pass Rate',  'Pass Rate',  'Pass Rate',  'Pass Rate',  'Pass Rate',  'Pass Rate',  'Pass Rate',  'Pass Rate'],
-        'compiler': ['inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor'],
-        'torchbench': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
-        'huggingface': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
-        'timm_models ': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
-        'refer_torchbench ': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
-        'refer_huggingface ': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
-        'refer_timm_models ': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ']
-    }
-    summary = pd.DataFrame(data)
-
-    if 'huggingface' in args.suite:
-        if 'amp_bf16' in args.precision:
-            if 'inference' in args.mode:
-                # passrate
-                summary.iloc[0:1, 4:5] = passrate_values['target_amp_bf16_inference']
-            if 'training' in args.mode:
-                summary.iloc[1:2, 4:5] = passrate_values['target_amp_bf16_training']
-                
-        if 'amp_fp16' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[2:3, 4:5] = passrate_values['target_amp_fp16_inference']
-            if 'training' in args.mode:
-                summary.iloc[3:4, 4:5] = passrate_values['target_amp_fp16_training']
-
-        if 'bfloat16' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[4:5, 4:5] = passrate_values['target_bfloat16_inference']
-            if 'training' in args.mode:
-                summary.iloc[5:6, 4:5] = passrate_values['target_bfloat16_training']
-        if 'float16' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[6:7, 4:5] = passrate_values['target_float16_inference']
-            if 'training' in args.mode:
-                summary.iloc[7:8, 4:5] = passrate_values['target_float16_training']
-        if 'float32' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[8:9, 4:5] = passrate_values['target_float32_inference']
-            if 'training' in args.mode:
-                summary.iloc[9:10, 4:5] = passrate_values['target_float32_training']
-
-        if args.reference is not None:
-            if 'amp_bf16' in args.precision:
-                if 'inference' in args.mode:
-                # passrate
-                    summary.iloc[0:1, 7:8] = passrate_values['reference_amp_bf16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[1:2, 7:8] = passrate_values['reference_amp_bf16_training']
-            if 'amp_fp16' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[2:3, 7:8] = passrate_values['reference_amp_fp16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[3:4, 7:8] = passrate_values['reference_amp_fp16_training']
-
-            if 'bfloat16' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[4:5, 7:8] = passrate_values['reference_bfloat16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[5:6, 7:8] = passrate_values['reference_bfloat16_training']
-            if 'float16' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[6:7, 7:8] = passrate_values['reference_float16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[7:8, 7:8] = passrate_values['reference_float16_training']
-            if 'float32' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[8:9, 7:8] = passrate_values['reference_float32_inference']
-                if 'training' in args.mode:
-                    summary.iloc[9:10, 7:8] = passrate_values['reference_float32_training']
-        
-        print("====================Summary=============================")
-        print(summary)
-        
-        sf = StyleFrame(summary)
-        for i in range(1, 10):
-            sf.set_column_width(i, 22)
-        for j in range(1, 12):
-            sf.set_row_height(j, 30)
-        sf.to_excel(sheet_name='Summary', excel_writer=excel)
-
-    if 'timm_models' in args.suite:
-        if 'amp_bf16' in args.precision:
-            if 'inference' in args.mode:
-                # passrate
-                summary.iloc[0:1, 5:6] = passrate_values['target_amp_bf16_inference']
-            if 'training' in args.mode:
-                summary.iloc[1:2, 5:6] = passrate_values['target_amp_bf16_training']
-        if 'amp_fp16' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[2:3, 5:6] = passrate_values['target_amp_fp16_inference']
-            if 'training' in args.mode:
-                summary.iloc[3:4, 5:6] = passrate_values['target_amp_fp16_training']
-
-        if 'bfloat16' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[4:5, 5:6] = passrate_values['target_bfloat16_inference']
-            if 'training' in args.mode:
-                summary.iloc[5:6, 5:6] = passrate_values['target_bfloat16_training']
-        if 'float16' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[6:7, 5:6] = passrate_values['target_float16_inference']
-            if 'training' in args.mode:
-                summary.iloc[7:8, 5:6] = passrate_values['target_float16_training']
-        if 'float32' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[8:9, 5:6] = passrate_values['target_float32_inference']
-            if 'training' in args.mode:
-                summary.iloc[9:10, 5:6] = passrate_values['target_float32_training']
-
-        if args.reference is not None:
-            if 'amp_bf16' in args.precision:
-                if 'inference' in args.mode:
-                    # passrate
-                    summary.iloc[0:1, 8:9] = passrate_values['reference_amp_bf16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[1:2, 8:9] = passrate_values['reference_amp_bf16_training']
-            if 'amp_fp16' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[2:3, 8:9] = passrate_values['reference_amp_fp16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[3:4, 8:9] = passrate_values['reference_amp_fp16_training']
-
-            if 'bfloat16' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[4:5, 8:9] = passrate_values['reference_bfloat16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[5:6, 8:9] = passrate_values['reference_bfloat16_training']
-            if 'float16' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[6:7, 8:9] = passrate_values['reference_float16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[7:8, 8:9] = passrate_values['reference_float16_training']
-            if 'float32' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[8:9, 8:9] = passrate_values['reference_float32_inference']
-                if 'training' in args.mode:
-                    summary.iloc[9:10, 8:9] = passrate_values['reference_float32_training']
-        
-        print("====================Summary=============================")
-        print(summary)
-        
-        sf = StyleFrame(summary)
-        for i in range(1, 10):
-            sf.set_column_width(i, 22)
-        for j in range(1, 12):
-            sf.set_row_height(j, 30)
-        sf.to_excel(sheet_name='Summary', excel_writer=excel)
-    
-    if 'torchbench' in args.suite:
-        if 'amp_bf16' in args.precision:
-            if 'inference' in args.mode:
-                # passrate
-                summary.iloc[0:1, 3:4] = passrate_values['target_amp_bf16_inference']
-            if 'training' in args.mode:
-                summary.iloc[1:2, 3:4] = passrate_values['target_amp_bf16_training']
-        if 'amp_fp16' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[2:3, 3:4] = passrate_values['target_amp_fp16_inference']
-            if 'training' in args.mode:
-                summary.iloc[3:4, 3:4] = passrate_values['target_amp_fp16_training']
-
-        if 'bfloat16' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[4:5, 3:4] = passrate_values['target_bfloat16_inference']
-            if 'training' in args.mode:
-                summary.iloc[5:6, 3:4] = passrate_values['target_bfloat16_training']
-        if 'float16' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[6:7, 3:4] = passrate_values['target_float16_inference']
-            if 'training' in args.mode:
-                summary.iloc[7:8, 3:4] = passrate_values['target_float16_training']
-        if 'float32' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[8:9, 3:4] = passrate_values['target_float32_inference']
-            if 'training' in args.mode:
-                summary.iloc[9:10, 3:4] = passrate_values['target_float32_training']
-
-        if args.reference is not None:
-            if 'amp_bf16' in args.precision:
-                if 'inference' in args.mode:
-                    # passrate
-                    summary.iloc[0:1, 6:7] = passrate_values['reference_amp_bf16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[1:2, 6:7] = passrate_values['reference_amp_bf16_training']
-
-            if 'amp_fp16' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[2:3, 6:7] = passrate_values['reference_amp_fp16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[3:4, 6:7] = passrate_values['reference_amp_fp16_training']
-
-            if 'bfloat16' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[4:5, 6:7] = passrate_values['reference_bfloat16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[5:6, 6:7] = passrate_values['reference_bfloat16_training']
-            if 'float16' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[6:7, 6:7] = passrate_values['reference_float16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[7:8, 6:7] = passrate_values['reference_float16_training']
-            if 'float32' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[8:9, 6:7] = passrate_values['reference_float32_inference']
-                if 'training' in args.mode:
-                    summary.iloc[9:10, 6:7] = passrate_values['reference_float32_training']
-        
-        print("====================Summary=============================")
-        print(summary)
-        
-        sf = StyleFrame(summary)
-        for i in range(1, 10):
-            sf.set_column_width(i, 22)
-        for j in range(1, 12):
-            sf.set_row_height(j, 30)
-        sf.to_excel(sheet_name='Summary', excel_writer=excel)
-
-def generate_report(excel, precision_list, mode_list):
-    for p in precision_list:
-        for m in mode_list:
-            update_details(p, m, excel)
-    update_summary(excel)
-
-
-def excel_postprocess(file, precison, mode):
-    wb = file.book
-    # details
-    for p in precison:
-        for m in mode:
-            wdt = wb[p + '_' + m]
-            wdt.merge_cells(start_row=1, end_row=2, start_column=1, end_column=1)
-            wdt.merge_cells(start_row=1, end_row=1, start_column=3, end_column=4)
-            wdt.merge_cells(start_row=1, end_row=1, start_column=5, end_column=6)
-            wdt.merge_cells(start_row=1, end_row=1, start_column=7, end_column=7)
-    wb.save(file)
-
-
-if __name__ == '__main__':
-    summary_path = 'inductor_log/' + str(args.suite) + '/Inductor_' + args.suite + '_E2E_Test_Acc_Report.xlsx'
-    excel = StyleFrame.ExcelWriter(summary_path)
-    print(f"=========Acc Check Summary file located in {summary_path}==================")
-    generate_report(excel, args.precision, args.mode)
-    excel_postprocess(excel, args.precision, args.mode)
diff --git a/.github/scripts/inductor_perf_summary.py b/.github/scripts/inductor_perf_summary.py
deleted file mode 100644
index 02a68eea7..000000000
--- a/.github/scripts/inductor_perf_summary.py
+++ /dev/null
@@ -1,498 +0,0 @@
-import argparse
-
-import pandas as pd
-from scipy.stats import gmean
-from styleframe import StyleFrame, Styler, utils
-
-parser = argparse.ArgumentParser(description="Generate report")
-parser.add_argument('-s', '--suite', default='huggingface', choices=["torchbench", "huggingface", "timm_models"], type=str, help='model suite name')
-parser.add_argument('-p', '--precision', default=["amp_fp16", "float32"], nargs='*', type=str, help='precision')
-parser.add_argument('-r', '--reference', type=str, help='reference log files')
-parser.add_argument('-m', '--mode', default=["inference", "training"], nargs='*', type=str, help='mode name')
-parser.add_argument('--html_off', action='store_true', help='turn off html file generate')
-args = parser.parse_args()
-
-passrate_values = {}
-geomean_values = {}
-
-new_performance_regression=pd.DataFrame()
-
-failure_style = Styler(bg_color='#FF0000', font_color=utils.colors.black)
-regression_style = Styler(bg_color='#F0E68C', font_color=utils.colors.red)
-improve_style = Styler(bg_color='#00FF00', font_color=utils.colors.black)
-
-# refer https://github.com/pytorch/pytorch/blob/main/benchmarks/dynamo/runner.py#L757-L778
-
-
-def percentage(part, whole, decimals=2):
-    if whole == 0:
-        return 0
-    return round(100 * float(part) / float(whole), decimals)
-
-
-def get_passing_entries(df, column_name):
-    return df[column_name][df[column_name] > 0]
-
-
-def caculate_geomean(df, column_name):
-    cleaned_df = get_passing_entries(df, column_name).clip(1)
-    if cleaned_df.empty:
-        return "0.0x"
-    return f"{gmean(cleaned_df):.2f}x"
-
-
-def caculate_passrate(df, compiler):
-    total = len(df.index)
-    passing = df[df[compiler] > 0.0][compiler].count()
-    perc = int(percentage(passing, total, decimals=0))
-    return f"{perc}%, {passing}/{total}"
-
-def get_perf_csv(precision, mode):
-    target_path = 'inductor_log/' + args.suite + '/' + precision + '/inductor_' + args.suite + '_' + precision + '_' + mode + '_xpu_performance.csv'
-    target_ori_data = pd.read_csv(target_path, header=0, encoding='utf-8')
-    target_data = target_ori_data.copy()
-    target_data.sort_values(by=['name'])
-    
-    if args.reference is not None:
-        reference_file_path = args.reference + '/inductor_log/' + args.suite + '/' + precision + '/inductor_' + args.suite + '_' + precision + '_' + mode + '_xpu_performance.csv'
-        reference_ori_data = pd.read_csv(reference_file_path, header=0, encoding='utf-8')
-        reference_data = reference_ori_data.copy()
-        reference_data.sort_values(by=['name'])
-        data = pd.merge(target_data,reference_data,on=['name'],how= 'outer')
-        return data
-    else:
-        return target_data
-
-def process(input, precision, mode):
-    global geomean_values, passrate_values
-    if input is not None:
-        if args.reference is None:
-            data_new = input[['name', 'batch_size', 'speedup', 'abs_latency','compilation_latency']].rename(columns={'name': 'name', 'batch_size': 'batch_size', 'speedup': 'speedup', "abs_latency": 'inductor', "compilation_latency": 'compilation_latency'})
-            data_new['inductor'] = data_new['inductor'].apply(pd.to_numeric, errors='coerce').div(1000)
-            data_new['speedup'] = data_new['speedup'].apply(pd.to_numeric, errors='coerce').fillna(0.0)
-            data_new['eager'] = data_new['speedup'] * data_new['inductor']
-            geomean_values['target_' + str(precision) + '_' + str(mode)] = caculate_geomean(data_new, 'speedup')
-            passrate_values['target_' + str(precision) + '_' + str(mode)] = caculate_passrate(data_new, 'inductor')
-            data = StyleFrame({'name': list(data_new['name']),
-                            'batch_size': list(data_new['batch_size']),
-                            'speedup': list(data_new['speedup']),
-                            'inductor': list(data_new['inductor']),
-                            'eager': list(data_new['eager']),
-                            'compilation_latency': list(data_new['compilation_latency'])})
-            data.set_column_width(1, 10)
-            data.set_column_width(2, 18)
-            data.set_column_width(3, 18)
-            data.set_column_width(4, 18)
-            data.set_column_width(5, 15)
-            data.set_column_width(6, 20)
-            data.set_row_height(rows=data.row_indexes, height=15)
-        else:
-            data_new=input[['name','batch_size_x','speedup_x','abs_latency_x','compilation_latency_x']].rename(columns={'name':'name','batch_size_x':'batch_size_new','speedup_x':'speed_up_new',"abs_latency_x":'inductor_new',"compilation_latency_x":'compilation_latency_new'})
-            data_new['inductor_new']=data_new['inductor_new'].astype(float).div(1000)
-            data_new['speed_up_new']=data_new['speed_up_new'].apply(pd.to_numeric, errors='coerce').fillna(0.0)
-            data_new['eager_new'] = data_new['speed_up_new'] * data_new['inductor_new']
-            geomean_values['target_' + str(precision) + '_' + str(mode)] = caculate_geomean(data_new, 'speed_up_new')
-            passrate_values['target_' + str(precision) + '_' + str(mode)] = caculate_passrate(data_new, 'inductor_new')        
-            data_old=input[['batch_size_y','speedup_y','abs_latency_y','compilation_latency_y']].rename(columns={'batch_size_y':'batch_size_old','speedup_y':'speed_up_old',"abs_latency_y":'inductor_old',"compilation_latency_y":'compilation_latency_old'}) 
-            data_old['inductor_old']=data_old['inductor_old'].astype(float).div(1000)
-            data_old['speed_up_old']=data_old['speed_up_old'].apply(pd.to_numeric, errors='coerce').fillna(0.0)
-            data_old['eager_old'] = data_old['speed_up_old'] * data_old['inductor_old']
-            input['speedup_x']=input['speedup_x'].apply(pd.to_numeric, errors='coerce').fillna(0.0)
-            input['speedup_y']=input['speedup_y'].apply(pd.to_numeric, errors='coerce').fillna(0.0)
-            geomean_values['reference_' + str(precision) + '_' + str(mode)] = caculate_geomean(data_old, 'speed_up_old')
-            passrate_values['reference_' + str(precision) + '_' + str(mode)] = caculate_passrate(data_old, 'inductor_old')        
-            data_ratio= pd.DataFrame(round(input['speedup_x'] / input['speedup_y'],2),columns=['Ratio Speedup(New/old)'])
-            data_ratio['Eager Ratio(old/new)'] = pd.DataFrame(round(data_old['eager_old'] / data_new['eager_new'],2))
-            data_ratio['Inductor Ratio(old/new)'] = pd.DataFrame(round(data_old['inductor_old'] / data_new['inductor_new'],2))
-            data_ratio['Compilation_latency_Ratio(old/new)'] = pd.DataFrame(round(data_old['compilation_latency_old'] / data_new['compilation_latency_new'],2))
-                
-            combined_data = pd.DataFrame({
-                'name': list(data_new['name']),
-                'batch_size_new': list(data_new['batch_size_new']),
-                'speed_up_new': list(data_new['speed_up_new']),
-                'inductor_new': list(data_new['inductor_new']),
-                'eager_new': list(data_new['eager_new']),
-                'compilation_latency_new': list(data_new['compilation_latency_new']),
-                'batch_size_old': list(data_old['batch_size_old']),
-                'speed_up_old': list(data_old['speed_up_old']),
-                'inductor_old': list(data_old['inductor_old']),
-                'eager_old': list(data_old['eager_old']),
-                'compilation_latency_old': list(data_old['compilation_latency_old']),
-                'Ratio Speedup(New/old)': list(data_ratio['Ratio Speedup(New/old)']),
-                'Eager Ratio(old/new)': list(data_ratio['Eager Ratio(old/new)']),
-                'Inductor Ratio(old/new)': list(data_ratio['Inductor Ratio(old/new)']),
-                'Compilation_latency_Ratio(old/new)': list(data_ratio['Compilation_latency_Ratio(old/new)'])
-                })   
-            data = StyleFrame(combined_data)
-            data.set_column_width(1, 10)
-            data.set_column_width(2, 18) 
-            data.set_column_width(3, 18) 
-            data.set_column_width(4, 18)
-            data.set_column_width(5, 15)
-            data.set_column_width(6, 20)
-            data.set_column_width(7, 18)
-            data.set_column_width(8, 18) 
-            data.set_column_width(9, 18) 
-            data.set_column_width(10, 15)
-            data.set_column_width(11, 20)
-            data.set_column_width(12, 28)
-            data.set_column_width(13, 28) 
-            data.set_column_width(14, 28)
-            data.set_column_width(15, 32)
-            data.apply_style_by_indexes(indexes_to_style=data[data['batch_size_new'] == 0], styler_obj=failure_style)
-            data.apply_style_by_indexes(indexes_to_style=data[(data['Inductor Ratio(old/new)'] > 0) & (data['Inductor Ratio(old/new)'] < 0.9)],styler_obj=regression_style)
-            global new_performance_regression
-            regression = data.loc[(data['Inductor Ratio(old/new)'] > 0) & (data['Inductor Ratio(old/new)'] < 0.9)]
-            regression = regression.copy()
-            regression.loc[0] = list(regression.shape[1]*'*')
-            new_performance_regression = pd.concat([new_performance_regression,regression])
-            data.apply_style_by_indexes(indexes_to_style=data[data['Inductor Ratio(old/new)'] > 1.1],styler_obj=improve_style)
-            data.set_row_height(rows=data.row_indexes, height=15)        
-        return data
-    else:
-        return pd.DataFrame()
-
-def update_details(precision, mode, excel):
-    h = {"A": 'Model suite', "B": '', "C": "target", "D": '', "E": '', "F": '', "G": '',"H": args.reference, "I": '', "J": '',"K": '',"L":'',"M": 'Result Comp',"N": '',"O": '',"P":''}
-    if args.reference is None:
-        h = {"A": 'Model suite', "B": '', "C": "target", "D": '', "E": '', "F": '', "G": ''}
-    head = StyleFrame(pd.DataFrame(h, index=[0]))
-    head.set_column_width(1, 15)
-    head.set_row_height(rows=[1], height=15)
-
-    head.to_excel(excel_writer=excel, sheet_name=precision + '_' + mode, index=False, startrow=0, header=False)
-    target_raw_data = get_perf_csv(precision, mode)
-    #print("target_raw_data", target_raw_data)
-    target_data = process(target_raw_data, precision, mode)
-    target_data.to_excel(excel_writer=excel, sheet_name=precision + '_' + mode, index=False, startrow=1, startcol=1)
-
-def update_summary(excel):
-    data = {
-        'Test Secnario': ['AMP_BF16 Inference', ' ', 'AMP_BF16 Training', ' ', 'AMP_FP16 Inference', ' ', 'AMP_FP16 Training', ' ', 'BF16 Inference', ' ', 'BF16 Training', ' ', 'FP16 Inference', ' ', 'FP16 Training', ' ', 'FP32 Inference', ' ', 'FP32 Training', ' '],
-        'Comp Item': ['Pass Rate', 'Geomean Speedup', 'Pass Rate', 'Geomean Speedup', 'Pass Rate', 'Geomean Speedup', 'Pass Rate', 'Geomean Speedup', 'Pass Rate', 'Geomean Speedup', 'Pass Rate', 'Geomean Speedup', 'Pass Rate', 'Geomean Speedup', 'Pass Rate', 'Geomean Speedup', 'Pass Rate', 'Geomean Speedup', 'Pass Rate', 'Geomean Speedup'],
-        'Date': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
-        'Compiler': ['inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor'],
-        'torchbench': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
-        'huggingface': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
-        'timm_models ': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
-        'ref_torchbench ': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
-        'refer_huggingface ': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
-        'refer_timm_models ': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ']
-    }
-    summary = pd.DataFrame(data)
-
-    if 'huggingface' in args.suite:
-        if 'amp_bf16' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[0:1, 5:6] = passrate_values['target_amp_bf16_inference']
-                summary.iloc[1:2, 5:6] = geomean_values['target_amp_bf16_inference']
-            if 'training' in args.mode:
-                summary.iloc[2:3, 5:6] = passrate_values['target_amp_bf16_training']
-                summary.iloc[3:4, 5:6] = geomean_values['target_amp_bf16_training']
-        if 'amp_fp16' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[4:5, 5:6] = passrate_values['target_amp_fp16_inference']
-                summary.iloc[5:6, 5:6] = geomean_values['target_amp_fp16_inference']
-            if 'training' in args.mode:
-                summary.iloc[6:7, 5:6] = passrate_values['target_amp_fp16_training']
-                summary.iloc[7:8, 5:6] = geomean_values['target_amp_fp16_training']
-
-        if 'bfloat16' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[8:9, 5:6] = passrate_values['target_bfloat16_inference']
-                summary.iloc[9:10, 5:6] = geomean_values['target_bfloat16_inference']
-            if 'training' in args.mode:
-                summary.iloc[10:11, 5:6] = passrate_values['target_bfloat16_training']
-                summary.iloc[11:12, 5:6] = geomean_values['target_bfloat16_training']
-        if 'float16' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[12:13, 5:6] = passrate_values['target_float16_inference']
-                summary.iloc[13:14, 5:6] = geomean_values['target_float16_inference']
-            if 'training' in args.mode:
-                summary.iloc[14:15, 5:6] = passrate_values['target_float16_training']
-                summary.iloc[15:16, 5:6] = geomean_values['target_float16_training']
-        if 'float32' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[16:17, 5:6] = passrate_values['target_float32_inference']
-                summary.iloc[17:18, 5:6] = geomean_values['target_float32_inference']
-            if 'training' in args.mode:
-                summary.iloc[18:19, 5:6] = passrate_values['target_float32_training']
-                summary.iloc[19:20, 5:6] = geomean_values['target_float32_training']
-
-        if args.reference is not None:
-            if 'amp_bf16' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[0:1, 8:9] = passrate_values['reference_amp_bf16_inference']
-                    summary.iloc[1:2, 8:9] = geomean_values['reference_amp_bf16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[2:3, 8:9] = passrate_values['reference_amp_bf16_training']
-                    summary.iloc[3:4, 8:9] = geomean_values['reference_amp_bf16_training']
-            if 'amp_fp16' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[4:5, 8:9] = passrate_values['reference_amp_fp16_inference']
-                    summary.iloc[5:6, 8:9] = geomean_values['reference_amp_fp16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[6:7, 8:9] = passrate_values['reference_amp_fp16_training']
-                    summary.iloc[7:8, 8:9] = geomean_values['reference_amp_fp16_training']
-
-            if 'bfloat16' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[8:9, 8:9] = passrate_values['reference_bfloat16_inference']
-                    summary.iloc[9:10, 8:9] = geomean_values['reference_bfloat16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[10:11, 8:9] = passrate_values['reference_bfloat16_training']
-                    summary.iloc[11:12, 8:9] = geomean_values['reference_bfloat16_training']
-            if 'float16' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[12:13, 8:9] = passrate_values['reference_float16_inference']
-                    summary.iloc[13:14, 8:9] = geomean_values['reference_float16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[14:15, 8:9] = passrate_values['reference_float16_training']
-                    summary.iloc[15:16, 8:9] = geomean_values['reference_float16_training']
-            if 'float32' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[16:17, 8:9] = passrate_values['reference_float32_inference']
-                    summary.iloc[17:18, 8:9] = geomean_values['reference_float32_inference']
-                if 'training' in args.mode:
-                    summary.iloc[18:19, 8:9] = passrate_values['reference_float32_training']
-                    summary.iloc[19:20, 8:9] = geomean_values['reference_float32_training']
-        
-        print("====================Summary=============================")
-        print(summary)
-        
-        sf = StyleFrame(summary)
-        for i in range(1, 11):
-            sf.set_column_width(i, 22)
-        for j in range(1, 22):
-            sf.set_row_height(j, 30)
-        sf.to_excel(sheet_name='Summary', excel_writer=excel)
-
-    if 'timm_models' in args.suite:
-        if 'amp_bf16' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[0:1, 6:7] = passrate_values['target_amp_bf16_inference']
-                summary.iloc[1:2, 6:7] = geomean_values['target_amp_bf16_inference']
-            if 'training' in args.mode:
-                summary.iloc[2:3, 6:7] = passrate_values['target_amp_bf16_training']
-                summary.iloc[3:4, 6:7] = geomean_values['target_amp_bf16_training']
-        if 'amp_fp16' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[4:5, 6:7] = passrate_values['target_amp_fp16_inference']
-                summary.iloc[5:6, 6:7] = geomean_values['target_amp_fp16_inference']
-            if 'training' in args.mode:
-                summary.iloc[6:7, 6:7] = passrate_values['target_amp_fp16_training']
-                summary.iloc[7:8, 6:7] = geomean_values['target_amp_fp16_training']
-
-        if 'bfloat16' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[8:9, 6:7] = passrate_values['target_bfloat16_inference']
-                summary.iloc[9:10, 6:7] = geomean_values['target_bfloat16_inference']
-            if 'training' in args.mode:
-                summary.iloc[10:11, 6:7] = passrate_values['target_bfloat16_training']
-                summary.iloc[11:12, 6:7] = geomean_values['target_bfloat16_training']
-        if 'float16' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[12:13, 6:7] = passrate_values['target_float16_inference']
-                summary.iloc[13:14, 6:7] = geomean_values['target_float16_inference']
-            if 'training' in args.mode:
-                summary.iloc[14:15, 6:7] = passrate_values['target_float16_training']
-                summary.iloc[15:16, 6:7] = geomean_values['target_float16_training']
-        if 'float32' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[16:17, 6:7] = passrate_values['target_float32_inference']
-                summary.iloc[17:18, 6:7] = geomean_values['target_float32_inference']
-            if 'training' in args.mode:
-                summary.iloc[18:19, 6:7] = passrate_values['target_float32_training']
-                summary.iloc[19:20, 6:7] = geomean_values['target_float32_training']
-
-        if args.reference is not None:
-            if 'amp_bf16' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[0:1, 9:10] = passrate_values['reference_amp_bf16_inference']
-                    summary.iloc[1:2, 9:10] = geomean_values['reference_amp_bf16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[2:3, 9:10] = passrate_values['reference_amp_bf16_training']
-                    summary.iloc[3:4, 9:10] = geomean_values['reference_amp_bf16_training']
-
-            if 'amp_fp16' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[4:5, 9:10] = passrate_values['reference_amp_fp16_inference']
-                    summary.iloc[5:6, 9:10] = geomean_values['reference_amp_fp16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[6:7, 9:10] = passrate_values['reference_amp_fp16_training']
-                    summary.iloc[7:8, 9:10] = geomean_values['reference_amp_fp16_training']
-
-            if 'bfloat16' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[8:9, 9:10] = passrate_values['reference_bfloat16_inference']
-                    summary.iloc[9:10, 9:10] = geomean_values['reference_bfloat16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[10:11, 9:10] = passrate_values['reference_bfloat16_training']
-                    summary.iloc[11:12, 9:10] = geomean_values['reference_bfloat16_training']
-            if 'float16' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[12:13, 9:10] = passrate_values['reference_float16_inference']
-                    summary.iloc[13:14, 9:10] = geomean_values['reference_float16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[14:15, 9:10] = passrate_values['reference_float16_training']
-                    summary.iloc[15:16, 9:10] = geomean_values['reference_float16_training']
-            if 'float32' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[16:17, 9:10] = passrate_values['reference_float32_inference']
-                    summary.iloc[17:18, 9:10] = geomean_values['reference_float32_inference']
-                if 'training' in args.mode:
-                    summary.iloc[18:19, 9:10] = passrate_values['reference_float32_training']
-                    summary.iloc[19:20, 9:10] = geomean_values['reference_float32_training']
-        
-        print("====================Summary=============================")
-        print(summary)
-        
-        sf = StyleFrame(summary)
-        for i in range(1, 11):
-            sf.set_column_width(i, 22)
-        for j in range(1, 22):
-            sf.set_row_height(j, 30)
-        sf.to_excel(sheet_name='Summary', excel_writer=excel)
-    
-    if 'torchbench' in args.suite:
-        if 'amp_bf16' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[0:1, 4:5] = passrate_values['target_amp_bf16_inference']
-                summary.iloc[1:2, 4:5] = geomean_values['target_amp_bf16_inference']
-            if 'training' in args.mode:
-                summary.iloc[2:3, 4:5] = passrate_values['target_amp_bf16_training']
-                summary.iloc[3:4, 4:5] = geomean_values['target_amp_bf16_training']
-        if 'amp_fp16' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[4:5, 4:5] = passrate_values['target_amp_fp16_inference']
-                summary.iloc[5:6, 4:5] = geomean_values['target_amp_fp16_inference']
-            if 'training' in args.mode:
-                summary.iloc[6:7, 4:5] = passrate_values['target_amp_fp16_training']
-                summary.iloc[7:8, 4:5] = geomean_values['target_amp_fp16_training']
-
-        if 'bfloat16' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[8:9, 4:5] = passrate_values['target_bfloat16_inference']
-                summary.iloc[9:10, 4:5] = geomean_values['target_bfloat16_inference']
-            if 'training' in args.mode:
-                summary.iloc[10:11, 4:5] = passrate_values['target_bfloat16_training']
-                summary.iloc[11:12, 4:5] = geomean_values['target_bfloat16_training']
-        if 'float16' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[12:13, 4:5] = passrate_values['target_float16_inference']
-                summary.iloc[13:14, 4:5] = geomean_values['target_float16_inference']
-            if 'training' in args.mode:
-                summary.iloc[14:15, 4:5] = passrate_values['target_float16_training']
-                summary.iloc[15:16, 4:5] = geomean_values['target_float16_training']
-        if 'float32' in args.precision:
-            if 'inference' in args.mode:
-                summary.iloc[16:17, 4:5] = passrate_values['target_float32_inference']
-                summary.iloc[17:18, 4:5] = geomean_values['target_float32_inference']
-            if 'training' in args.mode:
-                summary.iloc[18:19, 4:5] = passrate_values['target_float32_training']
-                summary.iloc[19:20, 4:5] = geomean_values['target_float32_training']
-
-        if args.reference is not None:
-            if 'amp_bf16' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[0:1, 7:8] = passrate_values['reference_amp_bf16_inference']
-                    summary.iloc[1:2, 7:8] = geomean_values['reference_amp_bf16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[2:3, 7:8] = passrate_values['reference_amp_bf16_training']
-                    summary.iloc[3:4, 7:8] = geomean_values['reference_amp_bf16_training']
-            if 'amp_fp16' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[4:5, 7:8] = passrate_values['reference_amp_fp16_inference']
-                    summary.iloc[5:6, 7:8] = geomean_values['reference_amp_fp16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[6:7, 7:8] = passrate_values['reference_amp_fp16_training']
-                    summary.iloc[7:8, 7:8] = geomean_values['reference_amp_fp16_training']
-
-            if 'bfloat16' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[8:9, 7:8] = passrate_values['reference_bfloat16_inference']
-                    summary.iloc[9:10, 7:8] = geomean_values['reference_bfloat16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[10:11, 7:8] = passrate_values['reference_bfloat16_training']
-                    summary.iloc[11:12, 7:8] = geomean_values['reference_bfloat16_training']
-            if 'float16' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[12:13, 7:8] = passrate_values['reference_float16_inference']
-                    summary.iloc[13:14, 7:8] = geomean_values['reference_float16_inference']
-                if 'training' in args.mode:
-                    summary.iloc[14:15, 7:8] = passrate_values['reference_float16_training']
-                    summary.iloc[15:16, 7:8] = geomean_values['reference_float16_training']
-            if 'float32' in args.precision:
-                if 'inference' in args.mode:
-                    summary.iloc[16:17, 7:8] = passrate_values['reference_float32_inference']
-                    summary.iloc[17:18, 7:8] = geomean_values['reference_float32_inference']
-                if 'training' in args.mode:
-                    summary.iloc[18:19, 7:8] = passrate_values['reference_float32_training']
-                    summary.iloc[19:20, 7:8] = geomean_values['reference_float32_training']
-        
-        print("====================Summary=============================")
-        print(summary)
-        
-        sf = StyleFrame(summary)
-        for i in range(1, 11):
-            sf.set_column_width(i, 22)
-        for j in range(1, 22):
-            sf.set_row_height(j, 30)
-        sf.to_excel(sheet_name='Summary', excel_writer=excel)
-
-
-def generate_report(excel, precision_list, mode_list):
-    for p in precision_list:
-        for m in mode_list:
-            update_details(p, m, excel)
-    update_summary(excel)
-
-
-def excel_postprocess(file, precison, mode):
-    wb = file.book
-    # Summary
-    ws = wb['Summary']
-    for i in range(2, 21, 2):
-        ws.merge_cells(start_row=i, end_row=i + 1, start_column=1, end_column=1)
-    # details
-    for p in precison:
-        for m in mode:
-            wdt = wb[p + '_' + m]
-            wdt.merge_cells(start_row=1, end_row=2, start_column=1, end_column=1)
-            wdt.merge_cells(start_row=1, end_row=1, start_column=3, end_column=7)
-            wdt.merge_cells(start_row=1, end_row=1, start_column=8, end_column=12)
-            wdt.merge_cells(start_row=1, end_row=1, start_column=13, end_column=16)
-    wb.save(file)
-
-def html_generate(html_off):
-    if not html_off:
-        try:
-            content = pd.read_excel('inductor_log/' + args.suite + '/Inductor_' + args.suite + '_' + args.precision + '_' + args.mode + '_E2E_Test_Report' + '.xlsx')
-            #summary= pd.DataFrame(content[-1]).to_html(classes="table",index = False)
-            #swinfo= pd.DataFrame(content[1]).to_html(classes="table",index = False)
-            #refer_swinfo_html = ''
-            #if args.reference is not None:
-            #    refer_swinfo = pd.read_table(args.reference+'/inductor_log/version.txt', sep = '\:', header = None,names=['item', 'commit'],engine='python')
-            #    refer_swinfo_html = refer_swinfo.to_html(classes="table",index = False)            
-            #mt_failures= pd.DataFrame(content[2]).to_html(classes="table",index = False)
-            #st_failures= pd.DataFrame(content[3]).to_html(classes="table",index = False)
-            perf_regression= new_performance_regression.to_html(classes="table",index = False)
-            #failures_regression= new_failures.to_html(classes="table",index = False)
-            #with open('inductor_log/inductor_model_bench.html',mode = "a") as f,open('inductor_log/inductor_perf_regression.html',mode = "a") as perf_f,open('inductor_log/inductor_failures.html',mode = "a") as failure_f:
-            with open('inductor_log/' + args.suite + '/inductor_perf_regression.html',mode = "a") as perf_f:
-                #f.write(html_head()+"<p>Summary</p>"+summary+"<p>SW info</p>"+swinfo+"<p>Multi-threads Failures</p>"+mt_failures+"<p>Single-thread Failures</p>"+st_failures+"<p>new_perf_regression</p>"+perf_regression+"<p>new_failures</p>"+failures_regression+f"<p>image: docker pull ccr-registry.caas.intel.com/pytorch/pt_inductor:{args.image_tag}</p>"+html_tail())
-                perf_f.write("<p>new_perf_regression </p>"+perf_regression)
-                #failure_f.write(f"<p>new_failures in {str((datetime.now() - timedelta(days=2)).date())}</p>"+failures_regression)
-            #f.close()
-            perf_f.close()
-            #failure_f.close()              
-        except:
-            print("html_generate_failed")
-            pass
-
-
-if __name__ == '__main__':
-    excel = StyleFrame.ExcelWriter('inductor_log/' + str(args.suite) + '/Inductor_' + args.suite + '_E2E_Test_Perf_Report.xlsx')
-    generate_report(excel, args.precision, args.mode)
-    excel_postprocess(excel, args.precision, args.mode)
-    html_generate(args.html_off)
diff --git a/.github/scripts/inductor_summary.py b/.github/scripts/inductor_summary.py
new file mode 100644
index 000000000..e11147664
--- /dev/null
+++ b/.github/scripts/inductor_summary.py
@@ -0,0 +1,700 @@
+import argparse
+
+import pandas as pd
+from scipy.stats import gmean
+from styleframe import StyleFrame, Styler, utils
+
+parser = argparse.ArgumentParser(description="Generate report")
+parser.add_argument('-s', '--suite', default=["huggingface"], nargs='*', type=str, help='model suite name')
+parser.add_argument('-p', '--precision', default=["amp_fp16", "float32"], nargs='*', type=str, help='precision')
+parser.add_argument('-r', '--reference', type=str, help='reference log files')
+parser.add_argument('-m', '--mode', default=["inference", "training"], nargs='*', type=str, help='mode name')
+parser.add_argument('-sc', '--scenario', default=["performance"], nargs='*', type=str, help='Test scenario set')
+args = parser.parse_args()
+
+passrate_values = {}
+geomean_values = {}
+
+new_performance_regression=pd.DataFrame()
+
+failure_style = Styler(bg_color='#FF0000', font_color=utils.colors.black)
+regression_style = Styler(bg_color='#F0E68C', font_color=utils.colors.red)
+improve_style = Styler(bg_color='#00FF00', font_color=utils.colors.black)
+
+# refer https://github.com/pytorch/pytorch/blob/main/benchmarks/dynamo/runner.py#L757-L778
+
+
+def percentage(part, whole, decimals=2):
+    if whole == 0:
+        return 0
+    return round(100 * float(part) / float(whole), decimals)
+
+
+def get_passing_entries(df, column_name, scenario):
+    if scenario == 'performance':
+        return df[column_name][df[column_name] > 0]
+    else:
+        return df[df[column_name].notnull()]
+
+
+def caculate_geomean(df, column_name, scenario):
+    cleaned_df = get_passing_entries(df, column_name, scenario).clip(1)
+    if cleaned_df.empty:
+        return "0.0x"
+    return f"{gmean(cleaned_df):.2f}x"
+
+
+def caculate_passrate(df, key_word, scenario):
+    total = len(df.index)
+    if scenario == 'performance':
+        passing = df[df[key_word] > 0.0][key_word].count()
+    else:
+        df = get_passing_entries(df, key_word, scenario)
+        passing = df[df[key_word].fillna('').str.contains('pass')][key_word].count()
+    perc = int(percentage(passing, total, decimals=0))
+    return f"{perc}%, {passing}/{total}"
+
+def get_perf_csv(scenario, precision, mode, suite):
+    target_path = 'inductor_log/' + suite + '/' + precision + '/inductor_' + suite + '_' + precision + '_' + mode + '_xpu_' + scenario + '.csv'
+    target_ori_data = pd.read_csv(target_path, header=0, encoding='utf-8')
+    target_data = target_ori_data.copy()
+    target_data.sort_values(by=['name'])
+    
+    if args.reference is not None:
+        reference_file_path = args.reference + '/inductor_log/' + suite + '/' + precision + '/inductor_' + suite + '_' + precision + '_' + mode + '_xpu_' + scenario + '.csv'
+        reference_ori_data = pd.read_csv(reference_file_path, header=0, encoding='utf-8')
+        reference_data = reference_ori_data.copy()
+        reference_data.sort_values(by=['name'])
+        data = pd.merge(target_data,reference_data,on=['name'],how= 'outer')
+        return data
+    else:
+        return target_data
+
+def process(input, scenario, precision, mode):
+    global geomean_values, passrate_values
+    processed_data = pd.DataFrame()
+    if input is not None and scenario == 'performance':
+        if args.reference is None:
+            data_new = input[['name', 'batch_size', 'speedup', 'abs_latency','compilation_latency']].rename(columns={'name': 'name', 'batch_size': 'batch_size', 'speedup': 'speedup', "abs_latency": 'inductor', "compilation_latency": 'compilation_latency'})
+            data_new['inductor'] = data_new['inductor'].apply(pd.to_numeric, errors='coerce').div(1000)
+            data_new['speedup'] = data_new['speedup'].apply(pd.to_numeric, errors='coerce').fillna(0.0)
+            data_new['eager'] = data_new['speedup'] * data_new['inductor']
+            geomean_values['target_' + str(precision) + '_' + str(mode)] = caculate_geomean(data_new, 'speedup', scenario)
+            passrate_values['target_' + str(precision) + '_' + str(mode)] = caculate_passrate(data_new, 'inductor', scenario)
+            processed_data = StyleFrame({'name': list(data_new['name']),
+                            'batch_size': list(data_new['batch_size']),
+                            'speedup': list(data_new['speedup']),
+                            'inductor': list(data_new['inductor']),
+                            'eager': list(data_new['eager']),
+                            'compilation_latency': list(data_new['compilation_latency'])})
+            processed_data.set_column_width(1, 10)
+            processed_data.set_column_width(2, 18)
+            processed_data.set_column_width(3, 18)
+            processed_data.set_column_width(4, 18)
+            processed_data.set_column_width(5, 15)
+            processed_data.set_column_width(6, 20)
+            processed_data.set_row_height(rows=processed_data.row_indexes, height=15)
+        else:
+            data_new=input[['name','batch_size_x','speedup_x','abs_latency_x','compilation_latency_x']].rename(columns={'name':'name','batch_size_x':'batch_size_new','speedup_x':'speed_up_new',"abs_latency_x":'inductor_new',"compilation_latency_x":'compilation_latency_new'})
+            data_new['inductor_new']=data_new['inductor_new'].astype(float).div(1000)
+            data_new['speed_up_new']=data_new['speed_up_new'].apply(pd.to_numeric, errors='coerce').fillna(0.0)
+            data_new['eager_new'] = data_new['speed_up_new'] * data_new['inductor_new']
+            geomean_values['target_' + str(precision) + '_' + str(mode)] = caculate_geomean(data_new, 'speed_up_new', scenario)
+            passrate_values['target_' + str(precision) + '_' + str(mode)] = caculate_passrate(data_new, 'inductor_new', scenario)        
+            data_old=input[['batch_size_y','speedup_y','abs_latency_y','compilation_latency_y']].rename(columns={'batch_size_y':'batch_size_old','speedup_y':'speed_up_old',"abs_latency_y":'inductor_old',"compilation_latency_y":'compilation_latency_old'}) 
+            data_old['inductor_old']=data_old['inductor_old'].astype(float).div(1000)
+            data_old['speed_up_old']=data_old['speed_up_old'].apply(pd.to_numeric, errors='coerce').fillna(0.0)
+            data_old['eager_old'] = data_old['speed_up_old'] * data_old['inductor_old']
+            input['speedup_x']=input['speedup_x'].apply(pd.to_numeric, errors='coerce').fillna(0.0)
+            input['speedup_y']=input['speedup_y'].apply(pd.to_numeric, errors='coerce').fillna(0.0)
+            geomean_values['reference_' + str(precision) + '_' + str(mode)] = caculate_geomean(data_old, 'speed_up_old', scenario)
+            passrate_values['reference_' + str(precision) + '_' + str(mode)] = caculate_passrate(data_old, 'inductor_old', scenario)        
+            data_ratio= pd.DataFrame(round(input['speedup_x'] / input['speedup_y'],2),columns=['Ratio Speedup(New/old)'])
+            data_ratio['Eager Ratio(old/new)'] = pd.DataFrame(round(data_old['eager_old'] / data_new['eager_new'],2))
+            data_ratio['Inductor Ratio(old/new)'] = pd.DataFrame(round(data_old['inductor_old'] / data_new['inductor_new'],2))
+            data_ratio['Compilation_latency_Ratio(old/new)'] = pd.DataFrame(round(data_old['compilation_latency_old'] / data_new['compilation_latency_new'],2))
+                
+            combined_data = pd.DataFrame({
+                'name': list(data_new['name']),
+                'batch_size_new': list(data_new['batch_size_new']),
+                'speed_up_new': list(data_new['speed_up_new']),
+                'inductor_new': list(data_new['inductor_new']),
+                'eager_new': list(data_new['eager_new']),
+                'compilation_latency_new': list(data_new['compilation_latency_new']),
+                'batch_size_old': list(data_old['batch_size_old']),
+                'speed_up_old': list(data_old['speed_up_old']),
+                'inductor_old': list(data_old['inductor_old']),
+                'eager_old': list(data_old['eager_old']),
+                'compilation_latency_old': list(data_old['compilation_latency_old']),
+                'Ratio Speedup(New/old)': list(data_ratio['Ratio Speedup(New/old)']),
+                'Eager Ratio(old/new)': list(data_ratio['Eager Ratio(old/new)']),
+                'Inductor Ratio(old/new)': list(data_ratio['Inductor Ratio(old/new)']),
+                'Compilation_latency_Ratio(old/new)': list(data_ratio['Compilation_latency_Ratio(old/new)'])
+                }) 
+            processed_data = StyleFrame(combined_data)
+            processed_data.set_column_width(1, 10)
+            processed_data.set_column_width(2, 18) 
+            processed_data.set_column_width(3, 18) 
+            processed_data.set_column_width(4, 18)
+            processed_data.set_column_width(5, 15)
+            processed_data.set_column_width(6, 20)
+            processed_data.set_column_width(7, 18)
+            processed_data.set_column_width(8, 18) 
+            processed_data.set_column_width(9, 18) 
+            processed_data.set_column_width(10, 15)
+            processed_data.set_column_width(11, 20)
+            processed_data.set_column_width(12, 28)
+            processed_data.set_column_width(13, 28) 
+            processed_data.set_column_width(14, 28)
+            processed_data.set_column_width(15, 32)
+            processed_data.apply_style_by_indexes(indexes_to_style=processed_data[processed_data['batch_size_new'] == 0], styler_obj=failure_style)
+            processed_data.apply_style_by_indexes(indexes_to_style=processed_data[(processed_data['Inductor Ratio(old/new)'] > 0) & (processed_data['Inductor Ratio(old/new)'] < 0.9)],styler_obj=regression_style)
+            global new_performance_regression
+            regression = processed_data.loc[(processed_data['Inductor Ratio(old/new)'] > 0) & (processed_data['Inductor Ratio(old/new)'] < 0.9)]
+            regression = regression.copy()
+            regression.loc[0] = list(regression.shape[1]*'*')
+            new_performance_regression = pd.concat([new_performance_regression,regression])
+            processed_data.apply_style_by_indexes(indexes_to_style=processed_data[processed_data['Inductor Ratio(old/new)'] > 1.1],styler_obj=improve_style)
+            processed_data.set_row_height(rows=processed_data.row_indexes, height=15)        
+    if input is not None and scenario == 'accuracy':
+        if args.reference is None:
+            data_new = input[['name', 'batch_size', 'accuracy']].rename(columns={'name': 'name', 'batch_size': 'batch_size', 'accuracy': 'accuracy'})
+            passrate_values['target_' + str(precision) + '_' + str(mode)] = caculate_passrate(data_new, 'accuracy', scenario)
+            processed_data = data_new
+            processed_data = StyleFrame({'name': list(data_new['name']),
+                            'batch_size': list(data_new['batch_size']),
+                             'accuracy': list(data_new['accuracy'])})
+            processed_data.set_column_width(1, 10)
+            processed_data.set_column_width(2, 18)
+            processed_data.set_column_width(3, 18)
+            processed_data.set_row_height(rows=processed_data.row_indexes, height=15)
+        else:
+            data_new=input[['name','batch_size_x','accuracy_x']].rename(columns={'name':'name','batch_size_x':'batch_size_new','accuracy_x':'accuracy_new'})
+            passrate_values['target_' + str(precision) + '_' + str(mode)] = caculate_passrate(data_new, 'accuracy_new', scenario)        
+            data_old=input[['batch_size_y','accuracy_y']].rename(columns={'batch_size_y':'batch_size_old','accuracy_y':'accuracy_old'}) 
+            passrate_values['reference_' + str(precision) + '_' + str(mode)] = caculate_passrate(data_old, 'accuracy_old', scenario)
+            data_comp = pd.DataFrame((data_new['accuracy_new'] != 'pass') & (data_old['accuracy_old'] == 'pass'),columns=['Accuracy regression'])
+            combined_data = pd.DataFrame({
+                'name': list(data_new['name']),
+                'batch_size_new': list(data_new['batch_size_new']),
+                'accuracy_new': list(data_new['accuracy_new']),
+                'batch_size_old': list(data_old['batch_size_old']),
+                'accuracy_old': list(data_old['accuracy_old']),
+                'Accuracy regression': list(data_comp['Accuracy regression'])
+                })
+            processed_data = StyleFrame(combined_data)
+            processed_data.set_column_width(1, 10)
+            processed_data.set_column_width(2, 18) 
+            processed_data.set_column_width(3, 18) 
+            processed_data.set_column_width(4, 18)
+            processed_data.set_column_width(5, 15)
+            processed_data.set_column_width(6, 20)
+            processed_data.apply_style_by_indexes(indexes_to_style=processed_data[(processed_data['Accuracy regression'] == 'regression')],styler_obj=regression_style)
+            processed_data.set_row_height(rows=processed_data.row_indexes, height=15)        
+    return processed_data
+
+def update_details(scenario, precision, mode, suite, excel):
+    if scenario == 'performance':
+        h = {"A": 'Model suite', "B": '', "C": "target", "D": '', "E": '', "F": '', "G": '',"H": args.reference, "I": '', "J": '',"K": '',"L":'',"M": 'Result Comp',"N": '',"O": '',"P":''}
+        if args.reference is None:
+            h = {"A": 'Model suite', "B": '', "C": "target", "D": '', "E": '', "F": '', "G": ''}
+    else:
+        h = {"A": 'Model suite', "B": '', "C": "target", "D": '', "E": args.reference, "F": '', "G": 'Result Comp'}
+        if args.reference is None:
+            h = {"A": 'Model suite', "B": '', "C": "target", "D": ''}
+    head = StyleFrame(pd.DataFrame(h, index=[0]))
+    head.set_column_width(1, 15)
+    head.set_row_height(rows=[1], height=15)
+    head.to_excel(excel_writer=excel, sheet_name=suite + '_' + precision + '_' + mode[0:3] + '_' + scenario[0:3], index=False, startrow=0, header=False)
+    target_raw_data = get_perf_csv(scenario, precision, mode, suite)
+    target_data = process(target_raw_data, scenario, precision, mode)
+    target_data.to_excel(excel_writer=excel, sheet_name=suite + '_' + precision + '_' + mode[0:3] + '_' + scenario[0:3], index=False, startrow=1, startcol=1)
+
+def update_summary(excel, scenario, suite):
+    if scenario == 'performance':
+        data = {
+            'Test Secnario': ['AMP_BF16 Inference', ' ', 'AMP_BF16 Training', ' ', 'AMP_FP16 Inference', ' ', 'AMP_FP16 Training', ' ', 'BF16 Inference', ' ', 'BF16 Training', ' ', 'FP16 Inference', ' ', 'FP16 Training', ' ', 'FP32 Inference', ' ', 'FP32 Training', ' '],
+            'Comp Item': ['Pass Rate', 'Geomean Speedup', 'Pass Rate', 'Geomean Speedup', 'Pass Rate', 'Geomean Speedup', 'Pass Rate', 'Geomean Speedup', 'Pass Rate', 'Geomean Speedup', 'Pass Rate', 'Geomean Speedup', 'Pass Rate', 'Geomean Speedup', 'Pass Rate', 'Geomean Speedup', 'Pass Rate', 'Geomean Speedup', 'Pass Rate', 'Geomean Speedup'],
+            'Date': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
+            'Compiler': ['inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor'],
+            'torchbench': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
+            'huggingface': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
+            'timm_models ': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
+            'ref_torchbench ': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
+            'refer_huggingface ': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
+            'refer_timm_models ': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ']
+        }
+        summary = pd.DataFrame(data)
+
+        if suite == 'huggingface':
+            if 'amp_bf16' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[0:1, 5:6] = passrate_values['target_amp_bf16_inference']
+                    summary.iloc[1:2, 5:6] = geomean_values['target_amp_bf16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[2:3, 5:6] = passrate_values['target_amp_bf16_training']
+                    summary.iloc[3:4, 5:6] = geomean_values['target_amp_bf16_training']
+            if 'amp_fp16' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[4:5, 5:6] = passrate_values['target_amp_fp16_inference']
+                    summary.iloc[5:6, 5:6] = geomean_values['target_amp_fp16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[6:7, 5:6] = passrate_values['target_amp_fp16_training']
+                    summary.iloc[7:8, 5:6] = geomean_values['target_amp_fp16_training']
+
+            if 'bfloat16' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[8:9, 5:6] = passrate_values['target_bfloat16_inference']
+                    summary.iloc[9:10, 5:6] = geomean_values['target_bfloat16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[10:11, 5:6] = passrate_values['target_bfloat16_training']
+                    summary.iloc[11:12, 5:6] = geomean_values['target_bfloat16_training']
+            if 'float16' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[12:13, 5:6] = passrate_values['target_float16_inference']
+                    summary.iloc[13:14, 5:6] = geomean_values['target_float16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[14:15, 5:6] = passrate_values['target_float16_training']
+                    summary.iloc[15:16, 5:6] = geomean_values['target_float16_training']
+            if 'float32' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[16:17, 5:6] = passrate_values['target_float32_inference']
+                    summary.iloc[17:18, 5:6] = geomean_values['target_float32_inference']
+                if 'training' in args.mode:
+                    summary.iloc[18:19, 5:6] = passrate_values['target_float32_training']
+                    summary.iloc[19:20, 5:6] = geomean_values['target_float32_training']
+
+            if args.reference is not None:
+                if 'amp_bf16' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[0:1, 8:9] = passrate_values['reference_amp_bf16_inference']
+                        summary.iloc[1:2, 8:9] = geomean_values['reference_amp_bf16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[2:3, 8:9] = passrate_values['reference_amp_bf16_training']
+                        summary.iloc[3:4, 8:9] = geomean_values['reference_amp_bf16_training']
+                if 'amp_fp16' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[4:5, 8:9] = passrate_values['reference_amp_fp16_inference']
+                        summary.iloc[5:6, 8:9] = geomean_values['reference_amp_fp16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[6:7, 8:9] = passrate_values['reference_amp_fp16_training']
+                        summary.iloc[7:8, 8:9] = geomean_values['reference_amp_fp16_training']
+
+                if 'bfloat16' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[8:9, 8:9] = passrate_values['reference_bfloat16_inference']
+                        summary.iloc[9:10, 8:9] = geomean_values['reference_bfloat16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[10:11, 8:9] = passrate_values['reference_bfloat16_training']
+                        summary.iloc[11:12, 8:9] = geomean_values['reference_bfloat16_training']
+                if 'float16' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[12:13, 8:9] = passrate_values['reference_float16_inference']
+                        summary.iloc[13:14, 8:9] = geomean_values['reference_float16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[14:15, 8:9] = passrate_values['reference_float16_training']
+                        summary.iloc[15:16, 8:9] = geomean_values['reference_float16_training']
+                if 'float32' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[16:17, 8:9] = passrate_values['reference_float32_inference']
+                        summary.iloc[17:18, 8:9] = geomean_values['reference_float32_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[18:19, 8:9] = passrate_values['reference_float32_training']
+                        summary.iloc[19:20, 8:9] = geomean_values['reference_float32_training']
+
+        if suite == 'timm_models':
+            if 'amp_bf16' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[0:1, 6:7] = passrate_values['target_amp_bf16_inference']
+                    summary.iloc[1:2, 6:7] = geomean_values['target_amp_bf16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[2:3, 6:7] = passrate_values['target_amp_bf16_training']
+                    summary.iloc[3:4, 6:7] = geomean_values['target_amp_bf16_training']
+            if 'amp_fp16' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[4:5, 6:7] = passrate_values['target_amp_fp16_inference']
+                    summary.iloc[5:6, 6:7] = geomean_values['target_amp_fp16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[6:7, 6:7] = passrate_values['target_amp_fp16_training']
+                    summary.iloc[7:8, 6:7] = geomean_values['target_amp_fp16_training']
+
+            if 'bfloat16' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[8:9, 6:7] = passrate_values['target_bfloat16_inference']
+                    summary.iloc[9:10, 6:7] = geomean_values['target_bfloat16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[10:11, 6:7] = passrate_values['target_bfloat16_training']
+                    summary.iloc[11:12, 6:7] = geomean_values['target_bfloat16_training']
+            if 'float16' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[12:13, 6:7] = passrate_values['target_float16_inference']
+                    summary.iloc[13:14, 6:7] = geomean_values['target_float16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[14:15, 6:7] = passrate_values['target_float16_training']
+                    summary.iloc[15:16, 6:7] = geomean_values['target_float16_training']
+            if 'float32' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[16:17, 6:7] = passrate_values['target_float32_inference']
+                    summary.iloc[17:18, 6:7] = geomean_values['target_float32_inference']
+                if 'training' in args.mode:
+                    summary.iloc[18:19, 6:7] = passrate_values['target_float32_training']
+                    summary.iloc[19:20, 6:7] = geomean_values['target_float32_training']
+
+            if args.reference is not None:
+                if 'amp_bf16' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[0:1, 9:10] = passrate_values['reference_amp_bf16_inference']
+                        summary.iloc[1:2, 9:10] = geomean_values['reference_amp_bf16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[2:3, 9:10] = passrate_values['reference_amp_bf16_training']
+                        summary.iloc[3:4, 9:10] = geomean_values['reference_amp_bf16_training']
+
+                if 'amp_fp16' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[4:5, 9:10] = passrate_values['reference_amp_fp16_inference']
+                        summary.iloc[5:6, 9:10] = geomean_values['reference_amp_fp16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[6:7, 9:10] = passrate_values['reference_amp_fp16_training']
+                        summary.iloc[7:8, 9:10] = geomean_values['reference_amp_fp16_training']
+
+                if 'bfloat16' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[8:9, 9:10] = passrate_values['reference_bfloat16_inference']
+                        summary.iloc[9:10, 9:10] = geomean_values['reference_bfloat16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[10:11, 9:10] = passrate_values['reference_bfloat16_training']
+                        summary.iloc[11:12, 9:10] = geomean_values['reference_bfloat16_training']
+                if 'float16' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[12:13, 9:10] = passrate_values['reference_float16_inference']
+                        summary.iloc[13:14, 9:10] = geomean_values['reference_float16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[14:15, 9:10] = passrate_values['reference_float16_training']
+                        summary.iloc[15:16, 9:10] = geomean_values['reference_float16_training']
+                if 'float32' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[16:17, 9:10] = passrate_values['reference_float32_inference']
+                        summary.iloc[17:18, 9:10] = geomean_values['reference_float32_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[18:19, 9:10] = passrate_values['reference_float32_training']
+                        summary.iloc[19:20, 9:10] = geomean_values['reference_float32_training']
+        
+        if suite == 'torchbench':
+            if 'amp_bf16' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[0:1, 4:5] = passrate_values['target_amp_bf16_inference']
+                    summary.iloc[1:2, 4:5] = geomean_values['target_amp_bf16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[2:3, 4:5] = passrate_values['target_amp_bf16_training']
+                    summary.iloc[3:4, 4:5] = geomean_values['target_amp_bf16_training']
+            if 'amp_fp16' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[4:5, 4:5] = passrate_values['target_amp_fp16_inference']
+                    summary.iloc[5:6, 4:5] = geomean_values['target_amp_fp16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[6:7, 4:5] = passrate_values['target_amp_fp16_training']
+                    summary.iloc[7:8, 4:5] = geomean_values['target_amp_fp16_training']
+
+            if 'bfloat16' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[8:9, 4:5] = passrate_values['target_bfloat16_inference']
+                    summary.iloc[9:10, 4:5] = geomean_values['target_bfloat16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[10:11, 4:5] = passrate_values['target_bfloat16_training']
+                    summary.iloc[11:12, 4:5] = geomean_values['target_bfloat16_training']
+            if 'float16' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[12:13, 4:5] = passrate_values['target_float16_inference']
+                    summary.iloc[13:14, 4:5] = geomean_values['target_float16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[14:15, 4:5] = passrate_values['target_float16_training']
+                    summary.iloc[15:16, 4:5] = geomean_values['target_float16_training']
+            if 'float32' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[16:17, 4:5] = passrate_values['target_float32_inference']
+                    summary.iloc[17:18, 4:5] = geomean_values['target_float32_inference']
+                if 'training' in args.mode:
+                    summary.iloc[18:19, 4:5] = passrate_values['target_float32_training']
+                    summary.iloc[19:20, 4:5] = geomean_values['target_float32_training']
+
+            if args.reference is not None:
+                if 'amp_bf16' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[0:1, 7:8] = passrate_values['reference_amp_bf16_inference']
+                        summary.iloc[1:2, 7:8] = geomean_values['reference_amp_bf16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[2:3, 7:8] = passrate_values['reference_amp_bf16_training']
+                        summary.iloc[3:4, 7:8] = geomean_values['reference_amp_bf16_training']
+                if 'amp_fp16' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[4:5, 7:8] = passrate_values['reference_amp_fp16_inference']
+                        summary.iloc[5:6, 7:8] = geomean_values['reference_amp_fp16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[6:7, 7:8] = passrate_values['reference_amp_fp16_training']
+                        summary.iloc[7:8, 7:8] = geomean_values['reference_amp_fp16_training']
+
+                if 'bfloat16' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[8:9, 7:8] = passrate_values['reference_bfloat16_inference']
+                        summary.iloc[9:10, 7:8] = geomean_values['reference_bfloat16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[10:11, 7:8] = passrate_values['reference_bfloat16_training']
+                        summary.iloc[11:12, 7:8] = geomean_values['reference_bfloat16_training']
+                if 'float16' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[12:13, 7:8] = passrate_values['reference_float16_inference']
+                        summary.iloc[13:14, 7:8] = geomean_values['reference_float16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[14:15, 7:8] = passrate_values['reference_float16_training']
+                        summary.iloc[15:16, 7:8] = geomean_values['reference_float16_training']
+                if 'float32' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[16:17, 7:8] = passrate_values['reference_float32_inference']
+                        summary.iloc[17:18, 7:8] = geomean_values['reference_float32_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[18:19, 7:8] = passrate_values['reference_float32_training']
+                        summary.iloc[19:20, 7:8] = geomean_values['reference_float32_training']
+            
+        print(f"===================={scenario}Summary=============================")
+        print(summary)
+        
+        sf = StyleFrame(summary)
+        for i in range(1, 11):
+            sf.set_column_width(i, 22)
+        for j in range(1, 22):
+            sf.set_row_height(j, 30)
+        sf.to_excel(sheet_name=suite + '_'  + scenario + '_Summary', excel_writer=excel)
+
+    else:
+        data = {
+        'Test Secnario': ['AMP_BF16 Inference', 'AMP_BF16 Training', 'AMP_FP16 Inference', 'AMP_FP16 Training', 'BF16 Inference', 'BF16 Training', 'FP16 Inference', 'FP16 Training', 'FP32 Inference', 'FP32 Training'],
+        'Comp Item': ['Pass Rate',  'Pass Rate',  'Pass Rate',  'Pass Rate',  'Pass Rate',  'Pass Rate',  'Pass Rate',  'Pass Rate',  'Pass Rate',  'Pass Rate'],
+        'compiler': ['inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor', 'inductor'],
+        'torchbench': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
+        'huggingface': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
+        'timm_models ': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
+        'refer_torchbench ': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
+        'refer_huggingface ': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' '],
+        'refer_timm_models ': [' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ', ' ']
+        }
+        summary = pd.DataFrame(data)
+
+        if suite == 'huggingface':
+            if 'amp_bf16' in args.precision:
+                if 'inference' in args.mode:
+                    # passrate
+                    summary.iloc[0:1, 4:5] = passrate_values['target_amp_bf16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[1:2, 4:5] = passrate_values['target_amp_bf16_training']
+                    
+            if 'amp_fp16' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[2:3, 4:5] = passrate_values['target_amp_fp16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[3:4, 4:5] = passrate_values['target_amp_fp16_training']
+
+            if 'bfloat16' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[4:5, 4:5] = passrate_values['target_bfloat16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[5:6, 4:5] = passrate_values['target_bfloat16_training']
+            if 'float16' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[6:7, 4:5] = passrate_values['target_float16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[7:8, 4:5] = passrate_values['target_float16_training']
+            if 'float32' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[8:9, 4:5] = passrate_values['target_float32_inference']
+                if 'training' in args.mode:
+                    summary.iloc[9:10, 4:5] = passrate_values['target_float32_training']
+
+            if args.reference is not None:
+                if 'amp_bf16' in args.precision:
+                    if 'inference' in args.mode:
+                    # passrate
+                        summary.iloc[0:1, 7:8] = passrate_values['reference_amp_bf16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[1:2, 7:8] = passrate_values['reference_amp_bf16_training']
+                if 'amp_fp16' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[2:3, 7:8] = passrate_values['reference_amp_fp16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[3:4, 7:8] = passrate_values['reference_amp_fp16_training']
+
+                if 'bfloat16' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[4:5, 7:8] = passrate_values['reference_bfloat16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[5:6, 7:8] = passrate_values['reference_bfloat16_training']
+                if 'float16' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[6:7, 7:8] = passrate_values['reference_float16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[7:8, 7:8] = passrate_values['reference_float16_training']
+                if 'float32' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[8:9, 7:8] = passrate_values['reference_float32_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[9:10, 7:8] = passrate_values['reference_float32_training']
+
+        if suite == 'timm_models':
+            if 'amp_bf16' in args.precision:
+                if 'inference' in args.mode:
+                    # passrate
+                    summary.iloc[0:1, 5:6] = passrate_values['target_amp_bf16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[1:2, 5:6] = passrate_values['target_amp_bf16_training']
+            if 'amp_fp16' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[2:3, 5:6] = passrate_values['target_amp_fp16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[3:4, 5:6] = passrate_values['target_amp_fp16_training']
+
+            if 'bfloat16' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[4:5, 5:6] = passrate_values['target_bfloat16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[5:6, 5:6] = passrate_values['target_bfloat16_training']
+            if 'float16' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[6:7, 5:6] = passrate_values['target_float16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[7:8, 5:6] = passrate_values['target_float16_training']
+            if 'float32' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[8:9, 5:6] = passrate_values['target_float32_inference']
+                if 'training' in args.mode:
+                    summary.iloc[9:10, 5:6] = passrate_values['target_float32_training']
+
+            if args.reference is not None:
+                if 'amp_bf16' in args.precision:
+                    if 'inference' in args.mode:
+                        # passrate
+                        summary.iloc[0:1, 8:9] = passrate_values['reference_amp_bf16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[1:2, 8:9] = passrate_values['reference_amp_bf16_training']
+                if 'amp_fp16' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[2:3, 8:9] = passrate_values['reference_amp_fp16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[3:4, 8:9] = passrate_values['reference_amp_fp16_training']
+
+                if 'bfloat16' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[4:5, 8:9] = passrate_values['reference_bfloat16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[5:6, 8:9] = passrate_values['reference_bfloat16_training']
+                if 'float16' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[6:7, 8:9] = passrate_values['reference_float16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[7:8, 8:9] = passrate_values['reference_float16_training']
+                if 'float32' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[8:9, 8:9] = passrate_values['reference_float32_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[9:10, 8:9] = passrate_values['reference_float32_training']
+
+        if suite == 'torchbench':
+            if 'amp_bf16' in args.precision:
+                if 'inference' in args.mode:
+                    # passrate
+                    summary.iloc[0:1, 3:4] = passrate_values['target_amp_bf16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[1:2, 3:4] = passrate_values['target_amp_bf16_training']
+            if 'amp_fp16' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[2:3, 3:4] = passrate_values['target_amp_fp16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[3:4, 3:4] = passrate_values['target_amp_fp16_training']
+
+            if 'bfloat16' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[4:5, 3:4] = passrate_values['target_bfloat16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[5:6, 3:4] = passrate_values['target_bfloat16_training']
+            if 'float16' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[6:7, 3:4] = passrate_values['target_float16_inference']
+                if 'training' in args.mode:
+                    summary.iloc[7:8, 3:4] = passrate_values['target_float16_training']
+            if 'float32' in args.precision:
+                if 'inference' in args.mode:
+                    summary.iloc[8:9, 3:4] = passrate_values['target_float32_inference']
+                if 'training' in args.mode:
+                    summary.iloc[9:10, 3:4] = passrate_values['target_float32_training']
+
+            if args.reference is not None:
+                if 'amp_bf16' in args.precision:
+                    if 'inference' in args.mode:
+                        # passrate
+                        summary.iloc[0:1, 6:7] = passrate_values['reference_amp_bf16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[1:2, 6:7] = passrate_values['reference_amp_bf16_training']
+
+                if 'amp_fp16' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[2:3, 6:7] = passrate_values['reference_amp_fp16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[3:4, 6:7] = passrate_values['reference_amp_fp16_training']
+
+                if 'bfloat16' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[4:5, 6:7] = passrate_values['reference_bfloat16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[5:6, 6:7] = passrate_values['reference_bfloat16_training']
+                if 'float16' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[6:7, 6:7] = passrate_values['reference_float16_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[7:8, 6:7] = passrate_values['reference_float16_training']
+                if 'float32' in args.precision:
+                    if 'inference' in args.mode:
+                        summary.iloc[8:9, 6:7] = passrate_values['reference_float32_inference']
+                    if 'training' in args.mode:
+                        summary.iloc[9:10, 6:7] = passrate_values['reference_float32_training']
+            
+        print(f"===================={scenario} Summary=============================")
+        print(summary)
+        
+        sf = StyleFrame(summary)
+        for i in range(1, 10):
+            sf.set_column_width(i, 22)
+        for j in range(1, 12):
+            sf.set_row_height(j, 30)
+        sf.to_excel(sheet_name=suite + '_'  + scenario + '_Summary', excel_writer=excel)
+
+
+def generate_report(excel, scenario_list, precision_list, mode_list, suite_list):
+    for sc in scenario_list:
+        for s in suite_list:
+            for p in precision_list:
+                for m in mode_list:
+                    update_details(sc, p, m, s, excel)
+            update_summary(excel, sc, s)
+
+
+def excel_postprocess(file, scenario, precison, mode, suite):
+    wb = file.book
+    # Summary
+    #ws = wb['Summary']
+    #for i in range(2, 21, 2):
+    #    ws.merge_cells(start_row=i, end_row=i + 1, start_column=1, end_column=1)
+    # details
+    for sc in scenario:
+        for s in suite:
+            for p in precison:
+                for m in mode:
+                    wdt = wb[s + '_' + p + '_' + m[0:3] + '_' + sc[0:3]]
+                    wdt.merge_cells(start_row=1, end_row=2, start_column=1, end_column=1)
+                    wdt.merge_cells(start_row=1, end_row=1, start_column=3, end_column=7)
+                    wdt.merge_cells(start_row=1, end_row=1, start_column=8, end_column=12)
+                    wdt.merge_cells(start_row=1, end_row=1, start_column=13, end_column=16)
+            wb.save(file)
+
+
+if __name__ == '__main__':
+    excel = StyleFrame.ExcelWriter('inductor_log/Inductor_E2E_Test_Report.xlsx')
+    generate_report(excel, args.scenario, args.precision, args.mode, args.suite)
+    excel_postprocess(excel, args.scenario, args.precision, args.mode, args.suite)
diff --git a/.github/workflows/nightly_ondemand.yml b/.github/workflows/nightly_ondemand.yml
index 4f56edb78..959915dcb 100644
--- a/.github/workflows/nightly_ondemand.yml
+++ b/.github/workflows/nightly_ondemand.yml
@@ -62,7 +62,7 @@ on:
 permissions: read-all
 
 concurrency:
-  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref_name }}-${{ github.ref_type == 'branch' && github.sha }}-${{ github.event_name == 'workflow_dispatch' }}-${{ github.event_name == 'schedule' }}-${{ inputs.suite }}-${{ inputs.dt }}-${{ inputs.mode }}-${{ inputs.scenario }}-${{ inputs.triton }}-${{ inputs.model }}
+  group: ${{ github.workflow }}-${{ github.sha }}-${{ github.event_name }}-${{ inputs.pytorch }}-${{ inputs.keep_torch_xpu_ops }}-${{ inputs.ut }}-${{ inputs.triton }}-${{ inputs.suite }}-${{ inputs.dt }}-${{ inputs.mode }}-${{ inputs.scenario }}-${{ inputs.model }}-${{ inputs.python }}
   cancel-in-progress: true
 
 jobs:
@@ -288,7 +288,7 @@ jobs:
 
   Tests-Failure-And-Report:
     if: ${{ ! cancelled() }}
-    runs-on: pvc_e2e
+    runs-on: [ self-hosted, Linux ]
     permissions:
       issues: write
     env:
diff --git a/.github/workflows/nightly_ondemand_rolling.yml b/.github/workflows/nightly_ondemand_rolling.yml
new file mode 100644
index 000000000..4016e7e36
--- /dev/null
+++ b/.github/workflows/nightly_ondemand_rolling.yml
@@ -0,0 +1,372 @@
+name: Nightly-OnDemand Tests Rolling
+
+on:
+  schedule:
+    # GMT+8 21:30 every workday
+    - cron: '30 13 * * 0-4'
+    # GMT+8 0:30 Saturday
+    - cron: '30 16 * * 5'
+  workflow_dispatch:
+    inputs:
+      pytorch:
+        required: false
+        type: string
+        default: 'main'
+        description: Pytorch branch/commit
+      keep_torch_xpu_ops:
+        required: false
+        type: string
+        default: 'false'
+        description: Keep torch-xpu-ops pin. `true` means use pined commit
+      ut:
+        required: false
+        type: string
+        default: 'torch_xpu'
+        description: UT scope. `op_example,op_extended,op_ut,torch_xpu`. Delimiter is comma
+      triton:
+        required: false
+        type: string
+        default: ''
+        description: Triton commit. Use pytorch pined commit by default
+      suite:
+        required: true
+        type: string
+        default: 'huggingface'
+        description: Dynamo benchmarks test suite. `huggingface,timm_models,torchbench`. Delimiter is comma
+      dt:
+        required: true
+        type: string
+        default: 'float32'
+        description: Data precision of the test. `float32,bfloat16,float16,amp_bf16,amp_fp16`. Delimiter is comma
+      mode:
+        required: true
+        type: string
+        default: 'inference'
+        description: Test mode. `inference,training`. Delimiter is comma
+      scenario:
+        required: true
+        type: string
+        default: 'accuracy'
+        description: Test scenario. `accuracy,performance`. Delimiter is comma
+      model:
+        required: false
+        type: string
+        default: ''
+        description: Model. Will only run this one mode if set
+      python:
+        required: false
+        type: string
+        default: '3.10'
+        description: Python version
+
+permissions: read-all
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.sha }}-${{ github.event_name }}-${{ inputs.pytorch }}-${{ inputs.keep_torch_xpu_ops }}-${{ inputs.ut }}-${{ inputs.triton }}-${{ inputs.suite }}-${{ inputs.dt }}-${{ inputs.mode }}-${{ inputs.scenario }}-${{ inputs.model }}-${{ inputs.python }}
+  cancel-in-progress: true
+
+jobs:
+  Linux-Nightly-Ondemand-UT-Tests-Rolling:
+    if: github.event_name == 'schedule' || ${{ inputs.ut_suite }}
+    uses: ./.github/workflows/_linux_ut.yml
+    env:
+      ZE_AFFINITY_MASK: 0
+    with: 
+      keep_torch_xpu_ops: ${{ github.event_name == 'schedule' && 'false' || inputs.keep_torch_xpu_ops }}
+      ut: ${{ github.event_name == 'schedule' && 'op_example,op_extended,op_ut,torch_xpu' || inputs.ut }}
+      pytorch: ${{ github.event_name == 'schedule' && 'main' || inputs.pytorch }}
+      python: ${{ github.event_name == 'schedule' && '3.10' || inputs.python }}
+      runner: pvc_rolling
+  
+  Linux-Weekly-UT-Tests-ABI-0-Rolling:
+    if: github.event_name == 'schedule' && github.event.schedule == '30 16 * * 5'
+    uses: ./.github/workflows/_linux_ut.yml
+    env:
+      ZE_AFFINITY_MASK: 0
+    with: 
+      abi: 0
+      ut: op_example,op_extended,op_ut,torch_xpu
+      runner: pvc_rolling
+
+  Linux-Nightly-Ondemand-E2E-Tests-Rolling:
+    runs-on: pvc_rolling
+    # Don't run on forked repos
+    if: github.repository_owner == 'intel'
+    timeout-minutes: 3600
+    env:
+      pytorch: ${{ github.event_name == 'schedule' && 'main' || inputs.pytorch }}
+      keep_torch_xpu_ops: ${{ github.event_name == 'schedule' && 'false' || inputs.keep_torch_xpu_ops }}
+      ut: ${{ github.event_name == 'schedule' && 'op_example,op_extended,op_ut,torch_xpu' || inputs.ut }}
+      python: ${{ github.event_name == 'schedule' && '3.10' || inputs.python }}
+    outputs:
+      TORCH_BRANCH_ID: ${{ steps.pinned.outputs.TORCH_BRANCH_ID }}
+      TORCH_COMMIT_ID: ${{ steps.pinned.outputs.TORCH_COMMIT_ID }}
+      DRIVER_VERSION: ${{ steps.pinned.outputs.DRIVER_VERSION }}
+      BUNDLE_VERSION: ${{ steps.pinned.outputs.BUNDLE_VERSION }}
+      OS_PRETTY_NAME: ${{ steps.pinned.outputs.OS_PRETTY_NAME }}
+      GCC_VERSION: ${{ steps.pinned.outputs.GCC_VERSION }}
+      TORCHBENCH_COMMIT_ID: ${{ steps.pinned.outputs.TORCHBENCH_COMMIT_ID }}
+      TORCHVISION_COMMIT_ID: ${{ steps.pinned.outputs.TORCHVISION_COMMIT_ID }}
+      TORCHAUDIO_COMMIT_ID: ${{ steps.pinned.outputs.TORCHAUDIO_COMMIT_ID }}
+      TRANSFORMERS_VERSION: ${{ steps.pinned.outputs.TRANSFORMERS_VERSION }}
+      TIMM_COMMIT_ID: ${{ steps.pinned.outputs.TIMM_COMMIT_ID }}
+      TRITON_COMMIT_ID: ${{ steps.pinned.outputs.TRITON_COMMIT_ID }}
+      TIMEOUT_MODELS: ${{ steps.summary.outputs.TIMEOUT_MODELS }}
+    steps:
+      - name: Checkout torch-xpu-ops
+        uses: actions/checkout@v4
+      - name: Prepare Conda ENV
+        run: |
+          which conda && conda clean -ay
+          conda remove --all -y -n e2e_ci || rm -rf $(dirname ${CONDA_EXE})/../envs/e2e_ci
+          conda create -n e2e_ci python=${{ env.python }} cmake ninja -y
+          source activate e2e_ci
+          pip install mkl-static mkl-include
+          pip install pandas scipy tqdm
+      - name: Prepare Stock Pytorch
+        run: |
+          pwd
+          cd ../ && rm -rf pytorch
+          source activate e2e_ci
+          git clone https://github.com/pytorch/pytorch pytorch
+          cd pytorch && git checkout ${{ env.pytorch }} 
+          # apply PRs for stock pytorch
+          pip install requests
+          python ../torch-xpu-ops/.github/scripts/apply_torch_pr.py
+          git status && git show -s
+          git submodule sync && git submodule update --init --recursive
+          if [[ ${{ env.keep_torch_xpu_ops }} == 'true' ]]; then
+            echo "Don't replace torch-xpu-ops!"
+          else
+            rm -rf third_party/torch-xpu-ops && cp -r ../torch-xpu-ops third_party/
+            # Workaround for torch-xpu-ops ci test
+            sed -i "s/checkout --quiet \${TORCH_XPU_OPS_COMMIT}/log -n 1/g" caffe2/CMakeLists.txt
+          fi
+      - name: Identify pinned versions
+        id: pinned
+        run: |
+          cd ../pytorch
+          if [ -z ${{ inputs.triton }} ]; then
+            echo "TRITON_COMMIT_ID=$(<.ci/docker/ci_commit_pins/triton-xpu.txt)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
+          else
+            echo "TRITON_COMMIT_ID=${{ inputs.triton }}" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
+          fi
+          echo "TORCH_BRANCH_ID=$(git rev-parse --abbrev-ref HEAD)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
+          echo "TORCH_COMMIT_ID=$(git rev-parse HEAD)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
+          echo "TORCHBENCH_COMMIT_ID=$(<third_party/torch-xpu-ops/.github/ci_commit_pins/torchbench.txt)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
+          echo "TORCHVISION_COMMIT_ID=$(<.github/ci_commit_pins/vision.txt)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
+          echo "TORCHAUDIO_COMMIT_ID=$(<.github/ci_commit_pins/audio.txt)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
+          echo "TRANSFORMERS_VERSION=$(<.ci/docker/ci_commit_pins/huggingface.txt)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
+          echo "TIMM_COMMIT_ID=$(<.ci/docker/ci_commit_pins/timm.txt)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
+          echo "MODEL_ONLY_NAME=${{ inputs.model }}" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
+          source /opt/intel/oneapi/pytorch-gpu-dev-0.5/oneapi-vars.sh
+          echo "DRIVER_VERSION=$(dkms status 2>&1 |grep 'intel-i915-dkms' |sed 's/.*\///;s/,.*//')" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
+          echo "BUNDLE_VERSION=$(dpcpp --version 2>&1 |grep 'DPC++/C++' |sed 's/.*(//;s/).*//')" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
+          . /etc/os-release
+          echo "OS_PRETTY_NAME=${PRETTY_NAME}" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
+          echo "GCC_VERSION=$(gcc -dumpversion)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
+          echo ${GITHUB_ENV}
+      - name: Triton Installation
+        run: |
+          source activate e2e_ci
+          cd ../pytorch
+          TRITON_REPO="https://github.com/intel/intel-xpu-backend-for-triton"
+          echo ${TRITON_REPO}@${TRITON_COMMIT_ID}
+          pip install --force-reinstall "git+${TRITON_REPO}@${TRITON_COMMIT_ID}#subdirectory=python"
+      - name: Build Pytorch XPU
+        run: |
+          source activate e2e_ci
+          source .github/scripts/env.sh
+          cd ../pytorch
+          pip install -r requirements.txt
+          export CMAKE_PREFIX_PATH=${CMAKE_PREFIX_PATH}:${CONDA_PREFIX:-"$(dirname $(which conda))/../"}
+          python setup.py bdist_wheel
+          pip install --force-reinstall dist/*.whl
+      - name: Show GITHUB_ENV
+        run: |
+          echo "$GITHUB_ENV"
+          rm -rf ../pytorch/inductor_log
+          rm -rf /tmp/torchinductor_*
+
+      # Nihglty launch
+      - name: Nightly Huggingface FP32/BF16/FP16 Inference & Training Accuracy Test
+        if: github.event_name == 'schedule' && github.event.schedule == '30 13 * * 0-4'
+        uses: ./.github/actions/inductor-xpu-e2e-test
+        with:
+          suite: huggingface
+          env_prepare: true
+          dt: float32,bfloat16,float16
+          mode: inference,training
+          scenario: accuracy
+          hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
+      - name: Nightly Torchbench BF16 Training Accuracy Test
+        if: github.event_name == 'schedule' && github.event.schedule == '30 13 * * 0-4'
+        uses: ./.github/actions/inductor-xpu-e2e-test
+        with:
+          suite: torchbench
+          dt: bfloat16
+          mode: training
+          scenario: accuracy
+          env_prepare: true
+          hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
+      - name: Nightly Timm_models FP16 Training Accuracy Test
+        if: github.event_name == 'schedule' && github.event.schedule == '30 13 * * 0-4'
+        uses: ./.github/actions/inductor-xpu-e2e-test
+        with:
+          suite: timm_models
+          dt: float16
+          mode: training
+          scenario: accuracy
+          env_prepare: true
+          hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
+      # Weekly launch
+      - name: Weekly Huggingface Full Test
+        if: github.event_name == 'schedule' && github.event.schedule == '30 16 * * 5'
+        uses: ./.github/actions/inductor-xpu-e2e-test
+        with:
+          suite: huggingface
+          env_prepare: true
+          dt: float32,bfloat16,float16,amp_bf16,amp_fp16
+          mode: inference,training
+          scenario: accuracy,performance
+          hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
+      - name: Weekly Torchbench Full Test
+        if: github.event_name == 'schedule' && github.event.schedule == '30 16 * * 5'
+        uses: ./.github/actions/inductor-xpu-e2e-test
+        with:
+          suite: torchbench
+          env_prepare: true
+          dt: float32,bfloat16,float16,amp_bf16,amp_fp16
+          mode: inference,training
+          scenario: accuracy,performance
+          hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
+      - name: Weekly Timm_models Full Test
+        if: github.event_name == 'schedule' && github.event.schedule == '30 16 * * 5'
+        uses: ./.github/actions/inductor-xpu-e2e-test
+        with:
+          suite: timm_models
+          env_prepare: true
+          dt: float32,bfloat16,float16,amp_bf16,amp_fp16
+          mode: inference,training
+          scenario: accuracy,performance
+          hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
+      # On-demand launch
+      - name: OnDemand Test (${{ inputs.suite }} ${{ inputs.dt }} ${{ inputs.mode }} ${{ inputs.scenario }})
+        if: github.event_name != 'schedule'
+        uses: ./.github/actions/inductor-xpu-e2e-test
+        with:
+          suite: ${{ inputs.suite }}
+          env_prepare: true
+          dt: ${{ inputs.dt }}
+          mode: ${{ inputs.mode }}
+          scenario: ${{ inputs.scenario }}
+          hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
+      - name: Summarize archieve files
+        id: summary
+        if: ${{ ! cancelled() }}
+        run: |
+          rm -rf ${{ github.workspace }}/upload_files
+          cp -r ${{ github.workspace }}/../pytorch/inductor_log ${{ github.workspace }}/upload_files
+          mkdir -p ${{ github.workspace }}/../../_backup/ && cd ${{ github.workspace }}/../../_backup/
+          find . -type f -name "*.tgz" -mtime +3 -delete # delete files older than 3 days
+          tar zcf xpu-inductor-${GITHUB_RUN_ID}.tgz -C ${{ github.workspace }}/upload_files/ . # backup logs
+          failed_models=$(grep "Real failed models: *[1-9]" ${{ github.workspace }}/upload_files/summary_accuracy.log |wc -l || true)
+          timeout_models=$(grep "timeout models: *[1-9]" ${{ github.workspace }}/upload_files/summary_accuracy.log |wc -l || true)
+          if [ ${timeout_models} -ne 0 ];then
+            TIMEOUT_MODELS="$(
+              grep -E "timeout models: [1-9]|Summary for" ${{ github.workspace }}/upload_files/summary_accuracy.log |grep "timeout" -B 1
+            )"
+            echo "${TIMEOUT_MODELS}" |sed 's/Summary/\\nSummary/g;s/Timeout/\\nTimeout/g' |tee -a "${GITHUB_OUTPUT}"
+            grep -E "timeout models: [1-9]|Summary for" ${{ github.workspace }}/upload_files/summary_accuracy.log |grep "timeout" -B 1
+          fi
+          if [ ${failed_models} -ne 0 ];then
+            grep -E "Real failed models: [1-9]|Summary for" ${{ github.workspace }}/upload_files/summary_accuracy.log |grep "failed" -B 1
+            exit 1
+          fi
+      - name: Upload Inductor XPU E2E Data
+        if: ${{ ! cancelled() }}
+        uses: actions/upload-artifact@v4
+        with:
+          name: Inductor-XPU-E2E-Data-${{ github.event.pull_request.number || github.sha }}
+          path: ${{ github.workspace }}/upload_files
+
+  Tests-Failure-And-Report:
+    if: ${{ ! cancelled() }}
+    runs-on: [ self-hosted, Linux ]
+    permissions:
+      issues: write
+    env:
+      GH_TOKEN: ${{ github.token }}
+      python: ${{ github.event_name == 'schedule' && '3.10' || inputs.python }}
+    needs: Linux-Nightly-Ondemand-E2E-Tests
+    steps:
+      - name: Report github issue for XPU OPS nightly
+        if: github.repository_owner == 'intel'
+        run: |
+          set -xe
+          # Test env
+          build_url="${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
+          repo="${{ github.repository }}"
+          TORCH_BRANCH_ID="${{ needs.Linux-Nightly-Ondemand-E2E-Tests.outputs.TORCH_BRANCH_ID }}"
+          TORCH_COMMIT_ID="${{ needs.Linux-Nightly-Ondemand-E2E-Tests.outputs.TORCH_COMMIT_ID }}"
+          DRIVER_VERSION="${{ needs.Linux-Nightly-Ondemand-E2E-Tests.outputs.DRIVER_VERSION }}"
+          BUNDLE_VERSION="${{ needs.Linux-Nightly-Ondemand-E2E-Tests.outputs.BUNDLE_VERSION }}"
+          OS_PRETTY_NAME="${{ needs.Linux-Nightly-Ondemand-E2E-Tests.outputs.OS_PRETTY_NAME }}"
+          GCC_VERSION="${{ needs.Linux-Nightly-Ondemand-E2E-Tests.outputs.GCC_VERSION }}"
+          TORCHBENCH_COMMIT_ID="${{ needs.Linux-Nightly-Ondemand-E2E-Tests.outputs.TORCHBENCH_COMMIT_ID }}"
+          TORCHVISION_COMMIT_ID="${{ needs.Linux-Nightly-Ondemand-E2E-Tests.outputs.TORCHVISION_COMMIT_ID }}"
+          TORCHAUDIO_COMMIT_ID="${{ needs.Linux-Nightly-Ondemand-E2E-Tests.outputs.TORCHAUDIO_COMMIT_ID }}"
+          TRANSFORMERS_VERSION="${{ needs.Linux-Nightly-Ondemand-E2E-Tests.outputs.TRANSFORMERS_VERSION }}"
+          TIMM_COMMIT_ID="${{ needs.Linux-Nightly-Ondemand-E2E-Tests.outputs.TIMM_COMMIT_ID }}"
+          TRITON_COMMIT_ID="${{ needs.Linux-Nightly-Ondemand-E2E-Tests.outputs.TRITON_COMMIT_ID }}"
+          TIMEOUT_MODELS="${{ needs.Linux-Nightly-Ondemand-E2E-Tests.outputs.TIMEOUT_MODELS }}"
+          # Test status
+          if [ "${{ needs.Linux-Nightly-Ondemand-E2E-Tests.result }}" == "success" ];then
+            test_status=Success
+          elif [ "${{ needs.Linux-Nightly-Ondemand-E2E-Tests.result }}" == "failure" ];then
+            test_status=Failure
+            cc_comment="CC ${{ secrets.NIGHTLY_EMAIL_LIST }}"
+          else
+            test_status=None
+            exit 0
+          fi
+          # Test Type
+          if [ "${GITHUB_EVENT_NAME}" == "workflow_dispatch" ];then
+            test_type="On-demand"
+            test_issue_id=426
+            cc_comment="CC @${GITHUB_TRIGGERING_ACTOR}"
+          elif [ "${{ github.event.schedule }}" == "30 16 * * 5" ];then
+            test_type="Weekly"
+            test_issue_id=432
+          else
+            test_type="Nightly"
+            test_issue_id=432
+          fi
+          # Test report
+          echo -e "**${test_status}** $test_type Rolling Test on $(date +'%F'), See: $build_url\n" > ${{ github.workspace }}/report.txt
+          printf "Torch-xpu-ops | PyTorch | Triton\n--- | --- | ---\n${GITHUB_WORKFLOW_SHA:0:7} on ${GITHUB_REF_NAME} | " >> ${{ github.workspace }}/report.txt
+          printf "[${TORCH_COMMIT_ID:0:7}](https://github.com/pytorch/pytorch/commit/${TORCH_COMMIT_ID:0:7}) on $TORCH_BRANCH_ID | " >> ${{ github.workspace }}/report.txt
+          echo -e "[${TRITON_COMMIT_ID:0:7}](https://github.com/intel/intel-xpu-backend-for-triton/commit/${TRITON_COMMIT_ID:0:7}) \n" >> ${{ github.workspace }}/report.txt
+          printf "Transformers | Timm | Torchbench | Torchvision | Torchaudio\n--- | --- | --- | --- | ---\n" >> ${{ github.workspace }}/report.txt
+          printf "[${TRANSFORMERS_VERSION:0:7}](https://github.com/huggingface/transformers/commit/${TRANSFORMERS_VERSION:0:7}) | " >> ${{ github.workspace }}/report.txt
+          printf "[${TIMM_COMMIT_ID:0:7}](https://github.com/huggingface/pytorch-image-models/commit/${TIMM_COMMIT_ID:0:7}) | " >> ${{ github.workspace }}/report.txt
+          printf "[${TORCHBENCH_COMMIT_ID:0:7}](https://github.com/pytorch/benchmark/commit/${TORCHBENCH_COMMIT_ID:0:7}) | " >> ${{ github.workspace }}/report.txt
+          printf "[${TORCHVISION_COMMIT_ID:0:7}](https://github.com/pytorch/vision/commit/${TORCHVISION_COMMIT_ID:0:7}) | " >> ${{ github.workspace }}/report.txt
+          echo -e "[${TORCHAUDIO_COMMIT_ID:0:7}](https://github.com/pytorch/audio/commit/${TORCHAUDIO_COMMIT_ID:0:7}) \n" >> ${{ github.workspace }}/report.txt
+          printf "Device | OS | GCC | Python | Driver(DKMS) | Bundle(DPCPP)\n--- | --- | --- | --- | --- | ---\n" >> ${{ github.workspace }}/report.txt
+          echo -e "$RUNNER_NAME | $OS_PRETTY_NAME | $GCC_VERSION | ${{ env.python }} | rolling-$DRIVER_VERSION| $BUNDLE_VERSION \n" >> ${{ github.workspace }}/report.txt
+          if [ "${GITHUB_EVENT_NAME}" == "workflow_dispatch" ];then
+            test_scope="${{ inputs.suite }}/${{ inputs.dt }}/${{ inputs.mode }}/${{ inputs.scenario }}"
+            if [ "${{ inputs.triton }}" != "" ];then
+              test_scope+="; triton=${{ inputs.triton }}"
+            fi
+            if [ "${{ inputs.model }}" != "" ];then
+              test_scope+="; model=${{ inputs.model }}"
+            fi
+            echo -e "Inputs | $test_scope\n--- | --- \n" >> ${{ github.workspace }}/report.txt
+          fi
+          echo "$TIMEOUT_MODELS" >> ${{ github.workspace }}/report.txt
+          echo "$cc_comment" >> ${{ github.workspace }}/report.txt
+          # Report
+          report_txt=$(cat ${{ github.workspace }}/report.txt)
+          gh --repo $repo issue comment $test_issue_id --body "$report_txt"
diff --git a/test/xpu/run_test_with_skip.py b/test/xpu/run_test_with_skip.py
index 2267d940c..3f2e371d4 100644
--- a/test/xpu/run_test_with_skip.py
+++ b/test/xpu/run_test_with_skip.py
@@ -1172,6 +1172,8 @@ def launch_test(test_case, skip_list=None, exe_list=None):
     "test_to_nn_TransformerEncoder_eval_mode_swap_True_set_grad_True_xpu_float32",
     "test_to_nn_TransformerEncoder_train_mode_swap_True_set_grad_True_xpu_float32",
     "test_to_nn_Transformer_swap_True_set_grad_True_xpu_float32",
+    #issue 746, adjust tolerence 
+    "test_non_contiguous_tensors_nn_Conv3d_xpu_float32",
 )
 res += launch_test("test_modules_xpu.py", skip_list)
 
@@ -2905,9 +2907,13 @@ def launch_test(test_case, skip_list=None, exe_list=None):
 res += launch_test("nn/test_convolution_xpu.py", skip_list)
 
 # test_dynamic_shapes
-
-
-res += launch_test("test_dynamic_shapes_xpu.py")
+skip_list = (
+    # issue 746, new ut failures introduced by new pytorch
+    "test_method_fn_add_first_type_int_second_type_float",
+    "test_method_fn_mul_first_type_int_second_type_float",
+    "test_method_fn_sub_first_type_int_second_type_float",
+)
+res += launch_test("test_dynamic_shapes_xpu.py", skip_list)
 
 # test_load_state_dict