hongfu_test_20250701

.idea/workspace.xml

@@ -1,58 +0,0 @@
-<?xml version="1.0" encoding="UTF-8"?>
-<project version="4">
-  <component name="ChangeListManager">
-    <list default="true" id="60da6b73-38f4-48aa-bd78-5731d35b3a7c" name="Changes" comment="" />
-    <option name="SHOW_DIALOG" value="false" />
-    <option name="HIGHLIGHT_CONFLICTS" value="true" />
-    <option name="HIGHLIGHT_NON_ACTIVE_CHANGELIST" value="false" />
-    <option name="LAST_RESOLUTION" value="IGNORE" />
-  </component>
-  <component name="Git.Settings">
-    <option name="RECENT_GIT_ROOT_PATH" value="$PROJECT_DIR$" />
-  </component>
-  <component name="MarkdownSettingsMigration">
-    <option name="stateVersion" value="1" />
-  </component>
-  <component name="ProjectColorInfo">{
-  &quot;customColor&quot;: &quot;&quot;,
-  &quot;associatedIndex&quot;: 2
-}</component>
-  <component name="ProjectId" id="2zGmpeKAt5GZlNtHRIRD45uRoxd" />
-  <component name="ProjectViewState">
-    <option name="hideEmptyMiddlePackages" value="true" />
-    <option name="showLibraryContents" value="true" />
-  </component>
-  <component name="PropertiesComponent"><![CDATA[{
-  "keyToString": {
-    "RunOnceActivity.OpenProjectViewOnStart": "true",
-    "RunOnceActivity.ShowReadmeOnStart": "true",
-    "git-widget-placeholder": "pr/18",
-    "last_opened_file_path": "E:/pythonProject/ChineseSafe-Benchmark",
-    "nodejs_package_manager_path": "npm",
-    "vue.rearranger.settings.migration": "true"
-  }
-}]]></component>
-  <component name="SharedIndexes">
-    <attachedChunks>
-      <set>
-        <option value="bundled-python-sdk-67fca87a943a-c986f194a52a-com.jetbrains.pycharm.pro.sharedIndexes.bundled-PY-233.11799.259" />
-      </set>
-    </attachedChunks>
-  </component>
-  <component name="SpellCheckerSettings" RuntimeDictionaries="0" Folders="0" CustomDictionaries="0" DefaultDictionary="application-level" UseSingleDictionary="true" transferred="true" />
-  <component name="TaskManager">
-    <task active="true" id="Default" summary="Default task">
-      <changelist id="60da6b73-38f4-48aa-bd78-5731d35b3a7c" name="Changes" comment="" />
-      <created>1751365967779</created>
-      <option name="number" value="Default" />
-      <option name="presentableId" value="Default" />
-      <updated>1751365967779</updated>
-      <workItem from="1751365968934" duration="39000" />
-      <workItem from="1751366116696" duration="54000" />
-    </task>
-    <servers />
-  </component>
-  <component name="TypeScriptGeneratedFilesManager">
-    <option name="version" value="3" />
-  </component>
-</project>

app.py

@@ -6,16 +6,15 @@ import pandas as pd
 from assets.text import INTRODUCTION_TEXT, METRICS_TEXT, EVALUTION_TEXT, ACKNOWLEDGEMENTS_TEXT, REFERENCE_TEXT
 
 
-ORIGINAL_DF = pd.read_csv("./data/chinese_benchmark_gen.csv", encoding='utf-8') # space separated values
-ORIGINAL_DF_PER = pd.read_csv("./data/chinese_benchmark_per.csv", encoding='utf-8') #
+ORIGINAL_DF = pd.read_csv("./data/chinese_benchmark_gen.csv", sep='\t') # space separated values
+ORIGINAL_DF_PER = pd.read_csv("./data/chinese_benchmark_per.csv", sep='\t') #
 
-ORIGINAL_DF_SUB_GEN = pd.read_csv("./data/subclass_gen.csv", encoding='utf-8') #
-ORIGINAL_DF_SUB_PER = pd.read_csv("./data/subclass_per.csv", encoding='utf-8')
-
-ORIGINAL_DF_NEW = pd.read_csv("./data/ChineseGuardBench.csv", encoding='utf-8') # new table
+ORIGINAL_DF_SUB_GEN = pd.read_csv("./data/subclass_gen.csv", sep=',') #
+ORIGINAL_DF_SUB_PER = pd.read_csv("./data/subclass_per.csv", sep=',')
 
 METRICS = ["Accuracy", "Precision_Unsafe", "Recall_Unsafe", "Precision_Safe", "Recall_Safe", "None"]
 
+
 SUBCLASS = ["Discrimination", "Variant", "Psychology", "Politics", "Eroticism", "Vulgarity", "Property", "Injury", "Criminality", "Ethics"]
 
 #SPLITS = ["Overall", "Subclass"]
@@ -27,10 +26,9 @@ CLASSIFICATION = {
         "~30B",
         "10B~20B",
         "5B~10B",
-        "1B~5B",
         "API",
     ]
-
+    
 }
 
 
@@ -38,17 +36,17 @@ CLASSIFICATION = {
 
 _BIBTEX = """
 @misc{zhang2024chinesesafechinesebenchmarkevaluating,
-      title={ChineseSafe: A Chinese Benchmark for Evaluating Safety in Large Language Models},
+      title={ChineseSafe: A Chinese Benchmark for Evaluating Safety in Large Language Models}, 
       author={Hengxiang Zhang and Hongfu Gao and Qiang Hu and Guanhua Chen and Lili Yang and Bingyi Jing and Hongxin Wei and Bing Wang and Haifeng Bai and Lei Yang},
       year={2024},
       eprint={2410.18491},
       archivePrefix={arXiv},
       primaryClass={cs.CL},
-      url={https://arxiv.org/abs/2410.18491},
+      url={https://arxiv.org/abs/2410.18491}, 
 }
 """
 
-_LAST_UPDATED = "July 28, 2025"
+_LAST_UPDATED = "April 13, 2025"
 
 banner_url = "./assets/logo.png"
 _BANNER = f'<div style="display: flex; justify-content: space-around;"><img src="{banner_url}" alt="Banner" style="width: 40vw; min-width: 300px; max-width: 600px;"> </div>'  # noqa
@@ -64,31 +62,18 @@ def format_csv_numbers(text):
 
 def format_csv_numbers_second(text):
     return text.split()
-
-
+    
+    
 def format_number(x):
     return float(f"{x:.3}")
 
 
-def get_dataset_new_csv(
-    model_size: List[str],
-):
-    df = ORIGINAL_DF_NEW[ORIGINAL_DF_NEW['Size'].isin(model_size)]
-    df = df.drop(columns="Size")
-
-    leaderboard_table = gr.components.Dataframe(
-        value=df,
-        interactive=False,
-        visible=True,
-    )
-    return leaderboard_table
-
 def get_dataset_csv(
     model_size: List[str],
 ):
     df = ORIGINAL_DF[ORIGINAL_DF['Size'].isin(model_size)]
     df = df.drop(columns="Size")
-
+    
     leaderboard_table = gr.components.Dataframe(
         value=df,
         interactive=False,
@@ -116,11 +101,11 @@ def get_dataset_csv_sub_gen(
 ):
     df = ORIGINAL_DF_SUB_GEN[ORIGINAL_DF_SUB_GEN['Size'].isin(model_size)]
     df = df.drop(columns="Size")
-
+    
     # get subclass
     subclass_choice_label = ["Model", subclass_choice+"_Accuracy", subclass_choice+"_Precision", subclass_choice+"_Recall"]
     df = df[subclass_choice_label]
-
+    
     leaderboard_table = gr.components.Dataframe(
         value=df,
         interactive=False,
@@ -135,11 +120,11 @@ def get_dataset_csv_sub_per(
 ):
     df = ORIGINAL_DF_SUB_PER[ORIGINAL_DF_SUB_PER['Size'].isin(model_size)]
     df = df.drop(columns="Size")
-
+    
     # get subclass
     subclass_choice_label = ["Model", subclass_choice+"_Accuracy", subclass_choice+"_Precision", subclass_choice+"_Recall"]
     df = df[subclass_choice_label]
-
+    
     leaderboard_table = gr.components.Dataframe(
         value=df,
         interactive=False,
@@ -158,15 +143,7 @@ def get_dataset_classfier_gen(
         subclass_choice = main_choice
         leaderboard_table = get_dataset_csv_sub_gen(model_size, subclass_choice)
     return leaderboard_table
-
-def get_ChineseGuardBench(
-    model_size: List[str],
-    main_choice: List[str],
-):
-    leaderboard_table = get_dataset_new_csv(model_size)
-    return leaderboard_table
-
-
+        
 def get_dataset_classfier_per(
     model_size: List[str],
     main_choice: List[str],
@@ -187,10 +164,10 @@ with gr.Blocks() as demo:
 
     with gr.Row():
         gr.Markdown(METRICS_TEXT, elem_classes="markdown-text")
-
+    
     with gr.Row():
         gr.Markdown(EVALUTION_TEXT, elem_classes="markdown-text")
-
+        
     with gr.Row():
         with gr.Column(scale=0.8):
             main_choice = gr.Dropdown(
@@ -199,8 +176,8 @@ with gr.Blocks() as demo:
                 label="Type",
                 info="Please choose the type to display.",
             )
-
-        with gr.Column(scale=10):
+            
+        with gr.Column(scale=10): 
             model_choice = gr.CheckboxGroup(
                 choices=CLASSIFICATION["model_size"],
                 value=CLASSIFICATION["model_size"],  # all be choosed
@@ -211,29 +188,24 @@ with gr.Blocks() as demo:
     #👉 this part is for csv table generatived
     with gr.Tabs(elem_classes="tab-buttons") as tabs:
         # ----------------- modify text -----------------
-
-        with gr.TabItem("🏅 Generation", elem_id="od-benchmark-tab-table", id=5):
+        
+        with gr.TabItem("🏅 Generation", elem_id="od-benchmark-tab-table", id=6):
             dataframe_all_gen = gr.components.Dataframe(
                 elem_id="leaderboard-table",
             )
-
-        with gr.TabItem("🏅 Perplexity", elem_id="od-benchmark-tab-table", id=6):
-            dataframe_all_per = gr.components.Dataframe(
-                elem_id="leaderboard-table",
-            )
             
-        with gr.TabItem("🏅 NEW", elem_id="od-benchmark-tab-table", id=7):
-            dataframe_all_guardbench = gr.components.Dataframe(
+        with gr.TabItem("🏅 Perplexity", elem_id="od-benchmark-tab-table", id=5):
+            dataframe_all_per = gr.components.Dataframe(
                 elem_id="leaderboard-table",
             )
 
     # ----------------- modify text -----------------
     with gr.Row():
         gr.Markdown(ACKNOWLEDGEMENTS_TEXT, elem_classes="markdown-text")
-
+    
     with gr.Row():
         gr.Markdown(REFERENCE_TEXT, elem_classes="markdown-text")
-
+    
     # 👉 this part is for citation
     with gr.Row():
         with gr.Accordion("📙 Citation", open=True):
@@ -244,18 +216,18 @@ with gr.Blocks() as demo:
                 elem_id="citation-button",
                 show_copy_button=True
             )
-
+    
     gr.Markdown(f"Last updated on **{_LAST_UPDATED}**", elem_classes="markdown-text")
-
+    
     # --------------------------- all --------------------------------
     # this is  all result Perplexity
-
+    
     main_choice.change(
         get_dataset_classfier_per,
         inputs=[model_choice, main_choice],
         outputs=dataframe_all_per,
     )
-
+        
     model_choice.change(
         get_dataset_classfier_per,
         inputs=[model_choice, main_choice],
@@ -267,45 +239,26 @@ with gr.Blocks() as demo:
         inputs=[model_choice, main_choice],
         outputs=dataframe_all_per,
     )
-
+    
     # this is all result generatived
     main_choice.change(
         get_dataset_classfier_gen,
         inputs=[model_choice, main_choice],
         outputs=dataframe_all_gen,
     )
-
+        
     model_choice.change(
         get_dataset_classfier_gen,
         inputs=[model_choice, main_choice],
         outputs=dataframe_all_gen,
     )
-
+    
     demo.load(
         fn=get_dataset_classfier_gen,
         inputs=[model_choice, main_choice],
         outputs=dataframe_all_gen,
     )
-
-    # this is new results for ChineseGuardBench
     
-    # main_choice.change(
-    #     get_ChineseGuardBench,
-    #     inputs=[model_choice, main_choice],
-    #     outputs=dataframe_all_guardbench,
-    # )
-
-    model_choice.change(
-        get_ChineseGuardBench,
-        inputs=[model_choice, main_choice],
-        outputs=dataframe_all_guardbench,
-    )
-
-    demo.load(
-        fn=get_ChineseGuardBench,
-        inputs=[model_choice, main_choice],
-        outputs=dataframe_all_guardbench,
-    )
-
+    
 demo.launch(share=True)
 

assets/text.py

@@ -34,13 +34,13 @@ EVALUTION_TEXT= """
 <span style="font-size:16px; font-family: 'Times New Roman', serif">
 We evaluate the models using two methods: perplexity(multiple choice) and generation. 
 For perplexity, we select the label which is the lowest perplexity as the predicted results. 
-For generation, we use the content generated by the model to make prediction. 
-In the "New" table, we present evaluation results for generation on a meticulously curated new benchmark, with details of its processing to be introduced later.
-The following are the results of the evaluation.👇👇👇
+For generation, we use the content generated by the model to make prediction.
+The following are the results of the evaluation. 👇👇👇    
 </span> <br><br>
 
 
 """ # noqa
+
 REFERENCE_TEXT = """
 # References
 <span style="font-size:16px; font-family: 'Times New Roman', serif">

changelog.md

@@ -1,6 +1,5 @@
 # CHANGELOG
 
-
 ### 2024-7-16
 version: v1.0.0
 
@@ -67,14 +66,4 @@ version: v1.0.6
         - Deepseek-chat-v3-0324
         - Qwen3
         - Gemma-3
-        - OpenThinker2
-
-### 2025-7-29
-version: v1.0.7
-
-    changed:
-    - [1]feat: Update the two models required by Deepexi.
-        - Deepexi-Guard-3B
-        - Qwen2.5-3B-Instruct
-  
-    - [2]feat: Update a new table ChineseGuardBench required by Deepxi.
+        - OpenThinker2

data/ChineseGuardBench.csv

@@ -1,33 +0,0 @@
-Model,Size,F1,Accuracy,Precision,Recall,FPR,FNR
-Deepexi-Guard-3B,1B~5B,89.63 ,89.72 ,85.53 ,94.15 ,14.24 ,5.85 
-Qwen3-32B,~30B,88.54 ,89.25 ,89.08 ,88.02 ,9.64 ,11.98 
-Qwen3-235B-A22B,>65B,87.92 ,88.96 ,90.86 ,85.17 ,7.66 ,14.83 
-Qwen3-235B-A22B-Instruct-2507,>65B,87.81 ,89.13 ,93.27 ,82.96 ,5.35 ,17.04 
-GLM-Z1-9B-0414,5B~10B,87.36 ,88.03 ,87.11 ,87.61 ,11.59 ,12.39 
-Qwen2.5-72B-Instruct,>65B,86.81 ,88.27 ,92.50 ,81.79 ,5.93 ,18.21 
-QwQ-32B,~30B,86.80 ,88.35 ,93.33 ,81.12 ,5.18 ,18.88 
-Phi-4,10B~20B,85.95 ,86.88 ,86.90 ,85.02 ,11.45 ,14.98 
-Gemma-3-27B-it,~30B,85.29 ,86.78 ,89.83 ,81.19 ,8.22 ,18.81 
-DeepSeek-R1-0528,>65B,85.24 ,87.47 ,96.02 ,76.63 ,2.84 ,23.37 
-Mistral-Small-3.2-24B-Instruct,~30B,85.07 ,87.03 ,93.14 ,78.29 ,5.15 ,21.71 
-GLM-4-9B-chat,5B~10B,84.85 ,86.27 ,88.47 ,81.52 ,9.49 ,18.48 
-MD-Judge-v0_2-internlm2_7B,5B~10B,84.63 ,85.88 ,87.03 ,82.37 ,10.98 ,17.63 
-DeepSeek-R1-Distill-Qwen-32B,~30B,84.55 ,86.64 ,93.05 ,77.47 ,5.17 ,22.53 
-Hunyuan-A13B-Instruct,>65B,84.32 ,86.21 ,90.97 ,78.58 ,6.98 ,21.42 
-Moonlight-16B-A3B-Instruct,10B~20B,84.21 ,84.35 ,80.41 ,88.38 ,19.25 ,11.62 
-GLM-Z1-32B-0414,~30B,83.40 ,85.75 ,92.63 ,75.85 ,5.40 ,24.15 
-Qwen3-8B,5B~10B,83.05 ,85.51 ,92.69 ,75.23 ,5.30 ,24.77 
-Qwen2.5-7B-Instruct,5B~10B,82.96 ,84.99 ,89.41 ,77.37 ,8.20 ,22.63 
-Qwen2.5-1.5B-Instruct,1B~5B,79.48 ,77.08 ,68.83 ,94.03 ,38.07 ,5.97 
-shieldgemma-2B,1B~5B,79.19 ,79.63 ,76.50 ,82.06 ,22.54 ,17.94 
-Qwen2.5-3B-Instruct,1B~5B,79.05 ,77.57 ,70.69 ,89.66 ,33.25 ,10.34 
-SHTEC_safety_fence_model_7B,5B~10B,78.44 ,82.48 ,93.54 ,67.54 ,4.17 ,32.46 
-Qwen3-4B,1B~5B,78.16 ,82.50 ,95.12 ,66.33 ,3.04 ,33.67 
-SmolLM3-3B,1B~5B,76.10 ,79.19 ,83.09 ,70.19 ,12.77 ,29.81 
-ERNIE-4.5-21B-A3B-Paddle,~20B,75.21 ,80.58 ,94.58 ,62.42 ,3.20 ,37.58 
-Qwen3-1.7B,1B~5B,74.46 ,79.34 ,89.36 ,63.82 ,6.79 ,36.18 
-internlm2_5-7B-chat,5B~10B,71.52 ,78.49 ,95.34 ,57.22 ,2.50 ,42.78 
-Llama-Guard-4-12B,10B~20B,65.66 ,74.64 ,90.99 ,51.36 ,4.54 ,48.64 
-Llama-Guard-3-8B,5B~10B,59.33 ,72.44 ,97.80 ,42.58 ,0.86 ,57.42 
-DeepSeek-R1-Distill-Qwen-7B,5B~10B,45.27 ,65.53 ,90.36 ,30.20 ,2.88 ,69.80 
-Gemma-3n-E4B-it,5B~10B,44.05 ,64.88 ,88.80 ,29.29 ,3.30 ,70.71 

data/chinese_benchmark_gen.csv

@@ -7,8 +7,6 @@ Gemini-2.5-flash-preview-05-20,API,71.27/0.27,73.40/0.23,70.16/0.71,69.17/0.53,7
 Llama-4-maverick,API,75.02/0.03,62.35/0.10,83.53/0.03,87.71/0.04,69.96/0.04
 Gemini-2.0-flash-001,API,52.04/0.61,0.95/0.05,69.46/0.38,99.60/0.03,51.93/0.62
 Deepseek-chat-v3-0324,API,66.00/0.11,45.08/0.11,77.52/0.19,86.93/0.11,61.28/0.08
-Deepexi-Guard-3B,1B~5B,78.26/0.0,89.35/0.0,64.16/0.0,72.04/0.0,92.35/0.0
-Qwen2.5-3B-Instruct,1B~5B,71.81/0.0,70.36/0.0,75.36/0.0,73.47/0.0,68.25/0.0
 Phi-3-small-8k-instruct,5B~10B,72.73/0.47,73.67/0.63,71.12/0.49,71.85/0.35,74.36/0.59
 Gemma-1.1-7B-it,5B~10B,71.70/0.26,68.66/0.37,80.11/0.05,76.00/0.09,63.26/0.47
 DeepSeek-LLM-7B-Chat,5B~10B,71.63/0.17,69.50/0.15,77.33/0.67,74.33/0.41,65.90/0.38

data/chinese_benchmark_per.csv

@@ -43,4 +43,4 @@ Opt-6.7B,5B~10B,48.54/0.43,49.24/0.31,86.62/1.03,43.40/1.18,10.30/0.55
 Mistral-7B-Instruct-v0.3,5B~10B,42.99/0.06,39.54/0.47,26.01/0.69,44.69/0.11,60.05/0.50
 Llama3-ChatQA-1.5-8B,5B~10B,42.11/0.29,37.46/0.85,23.20/0.89,44.20/0.09,61.11/0.57
 Qwen3-4B,5B~10B,46.04/0.00,47.79/0.00,85.94/0.00,30.39/0.00,6.14/0.00
-Gemma-3-4B-it,5B~10B,50.00/0.00,0.00/0.00,0.00/0.00,50.00/0.00,100.00/0.00
+Gemma-3-4B-it,5B~10B,50.00/0.00,0.00/0.00,0.00/0.00,50.00/0.00,100.00/0.00

data/subclass_gen.csv

@@ -48,4 +48,4 @@ Opt-30B,~30B,0.4672,0.4683,0.6648,0.5002,0.5082,0.7109,0.5044,0.4987,0.7354,0.53
 QwQ-32B-Preview,~30B,0.6837,0.7403,0.547,0.812,0.8219,0.8084,0.606,0.6749,0.3914,0.7516,0.8198,0.6977,0.8121,0.823,0.8081,0.847,0.8208,0.8801,0.6113,0.6736,0.3973,0.605,0.67,0.3873,0.7492,0.7768,0.6783,0.4656,0.3791,0.1124
 Qwen3-32B,~30B,0.5416,0.5902,0.2095,0.5495,0.6557,0.2531,0.477,0.3724,0.0843,0.6293,0.66,0.6114,0.8339,0.7311,0.6644,0.9577,0.3309,0.2379,0.1626,0.6958,0.6204,0.8741,0.2629,0.9049,0.9606,0.8489,0.5103,0.547,0.0453,0.8192
 Gemma-3-27b-it,~30B,0.66,0.6114,0.8339,0.7311,0.6644,0.9577,0.3309,0.2379,0.1626,0.6958,0.6204,0.8741,0.2629,0.9049,0.9606,0.8489,0.5103,0.547,0.0453,0.8192,0.4672,0.4683,0.6648,0.5002,0.5082,0.7109,0.5044,0.4987,0.7354,0.5314
-OpenThinker2-32B,~30B,0.6204,0.8741,0.2629,0.9049,0.9606,0.8489,0.5103,0.547,0.0453,0.8192,0.4672,0.4683,0.6648,0.5002,0.5082,0.7109,0.5044,0.4987,0.7354,0.5314,0.6837,0.7403,0.547,0.812,0.8219,0.8084,0.606,0.6749,0.3914,0.7516
+OpenThinker2-32B,~30B,0.6204,0.8741,0.2629,0.9049,0.9606,0.8489,0.5103,0.547,0.0453,0.8192,0.4672,0.4683,0.6648,0.5002,0.5082,0.7109,0.5044,0.4987,0.7354,0.5314,0.6837,0.7403,0.547,0.812,0.8219,0.8084,0.606,0.6749,0.3914,0.7516

data/subclass_per.csv

@@ -41,4 +41,4 @@ Opt-30B,~30B,0.5831,0.5754,0.5565,0.3952,0.338,0.1915,0.6784,0.6507,0.7506,0.579
 QwQ-32B-Preview,~30B,0.5231,0.5061,0.9839,0.5519,0.5328,1,0.4141,0.4443,0.7537,0.5814,0.565,0.9989,0.5529,0.534,0.9993,0.5318,0.5111,0.9993,0.5083,0.4978,0.9542,0.4392,0.4593,0.808,0.5238,0.5042,0.9922,0.5269,0.5128,0.9743
 Mistral-Small-24B-Instruct-2501,~30B,0.5897,0.5714,0.6393,0.7706,0.6931,0.9888,0.3109,0.1339,0.0727,0.7308,0.6984,0.8887,0.7454,0.683,0.9385,0.7584,0.6732,0.9835,0.585,0.5671,0.6297,0.3646,0.2744,0.1803,0.7088,0.645,0.8855,0.3839,0.3257,0.2233
 OpenThinker2-32B,~30B,0.7139 ,0.8341 ,0.5176 ,0.7722 ,0.8735 ,0.6482 ,0.4750 ,0.2581 ,0.0357 ,0.7162 ,0.6749 ,0.6366 ,0.7789 ,0.7893 ,0.7099 ,0.9938 ,0.4372 ,0.4025 ,0.2943 ,0.7921 ,0.5831 ,0.5754 ,0.5565 ,0.3952 ,0.3380 ,0.1915 ,0.6784 ,0.6507 ,0.7506 ,0.5798 
-Qwen3-32B,~30B,0.6749 ,0.6366 ,0.7789 ,0.7893 ,0.7099 ,0.9938 ,0.4372 ,0.4025 ,0.2943 ,0.7921 ,0.5831 ,0.5754 ,0.5565 ,0.3952 ,0.3380 ,0.1915 ,0.6784 ,0.6507 ,0.7506 ,0.5798 ,0.5231 ,0.5061 ,0.9839 ,0.5519 ,0.5328 ,1.0000 ,0.4141 ,0.4443 ,0.7537 ,0.5814
+Qwen3-32B,~30B,0.6749 ,0.6366 ,0.7789 ,0.7893 ,0.7099 ,0.9938 ,0.4372 ,0.4025 ,0.2943 ,0.7921 ,0.5831 ,0.5754 ,0.5565 ,0.3952 ,0.3380 ,0.1915 ,0.6784 ,0.6507 ,0.7506 ,0.5798 ,0.5231 ,0.5061 ,0.9839 ,0.5519 ,0.5328 ,1.0000 ,0.4141 ,0.4443 ,0.7537 ,0.5814