.idea/workspace.xml DELETED
@@ -1,58 +0,0 @@
1
- <?xml version="1.0" encoding="UTF-8"?>
2
- <project version="4">
3
- <component name="ChangeListManager">
4
- <list default="true" id="60da6b73-38f4-48aa-bd78-5731d35b3a7c" name="Changes" comment="" />
5
- <option name="SHOW_DIALOG" value="false" />
6
- <option name="HIGHLIGHT_CONFLICTS" value="true" />
7
- <option name="HIGHLIGHT_NON_ACTIVE_CHANGELIST" value="false" />
8
- <option name="LAST_RESOLUTION" value="IGNORE" />
9
- </component>
10
- <component name="Git.Settings">
11
- <option name="RECENT_GIT_ROOT_PATH" value="$PROJECT_DIR$" />
12
- </component>
13
- <component name="MarkdownSettingsMigration">
14
- <option name="stateVersion" value="1" />
15
- </component>
16
- <component name="ProjectColorInfo">{
17
- &quot;customColor&quot;: &quot;&quot;,
18
- &quot;associatedIndex&quot;: 2
19
- }</component>
20
- <component name="ProjectId" id="2zGmpeKAt5GZlNtHRIRD45uRoxd" />
21
- <component name="ProjectViewState">
22
- <option name="hideEmptyMiddlePackages" value="true" />
23
- <option name="showLibraryContents" value="true" />
24
- </component>
25
- <component name="PropertiesComponent"><![CDATA[{
26
- "keyToString": {
27
- "RunOnceActivity.OpenProjectViewOnStart": "true",
28
- "RunOnceActivity.ShowReadmeOnStart": "true",
29
- "git-widget-placeholder": "pr/18",
30
- "last_opened_file_path": "E:/pythonProject/ChineseSafe-Benchmark",
31
- "nodejs_package_manager_path": "npm",
32
- "vue.rearranger.settings.migration": "true"
33
- }
34
- }]]></component>
35
- <component name="SharedIndexes">
36
- <attachedChunks>
37
- <set>
38
- <option value="bundled-python-sdk-67fca87a943a-c986f194a52a-com.jetbrains.pycharm.pro.sharedIndexes.bundled-PY-233.11799.259" />
39
- </set>
40
- </attachedChunks>
41
- </component>
42
- <component name="SpellCheckerSettings" RuntimeDictionaries="0" Folders="0" CustomDictionaries="0" DefaultDictionary="application-level" UseSingleDictionary="true" transferred="true" />
43
- <component name="TaskManager">
44
- <task active="true" id="Default" summary="Default task">
45
- <changelist id="60da6b73-38f4-48aa-bd78-5731d35b3a7c" name="Changes" comment="" />
46
- <created>1751365967779</created>
47
- <option name="number" value="Default" />
48
- <option name="presentableId" value="Default" />
49
- <updated>1751365967779</updated>
50
- <workItem from="1751365968934" duration="39000" />
51
- <workItem from="1751366116696" duration="54000" />
52
- </task>
53
- <servers />
54
- </component>
55
- <component name="TypeScriptGeneratedFilesManager">
56
- <option name="version" value="3" />
57
- </component>
58
- </project>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
app.py CHANGED
@@ -6,16 +6,14 @@ import pandas as pd
6
  from assets.text import INTRODUCTION_TEXT, METRICS_TEXT, EVALUTION_TEXT, ACKNOWLEDGEMENTS_TEXT, REFERENCE_TEXT
7
 
8
 
9
- ORIGINAL_DF = pd.read_csv("./data/chinese_benchmark_gen.csv", encoding='utf-8') # space separated values
10
- ORIGINAL_DF_PER = pd.read_csv("./data/chinese_benchmark_per.csv", encoding='utf-8') #
11
 
12
- ORIGINAL_DF_SUB_GEN = pd.read_csv("./data/subclass_gen.csv", encoding='utf-8') #
13
- ORIGINAL_DF_SUB_PER = pd.read_csv("./data/subclass_per.csv", encoding='utf-8')
14
-
15
- ORIGINAL_DF_NEW = pd.read_csv("./data/ChineseGuardBench.csv", encoding='utf-8') # new table
16
 
17
  METRICS = ["Accuracy", "Precision_Unsafe", "Recall_Unsafe", "Precision_Safe", "Recall_Safe", "None"]
18
- # METRICS_GuardBench = ["Accuracy", "Precision_Unsafe", "Recall_Unsafe", "Precision_Safe", "Recall_Safe", "None"]
19
 
20
  SUBCLASS = ["Discrimination", "Variant", "Psychology", "Politics", "Eroticism", "Vulgarity", "Property", "Injury", "Criminality", "Ethics"]
21
 
@@ -28,10 +26,9 @@ CLASSIFICATION = {
28
  "~30B",
29
  "10B~20B",
30
  "5B~10B",
31
- "1B~5B",
32
  "API",
33
  ]
34
-
35
  }
36
 
37
 
@@ -39,17 +36,17 @@ CLASSIFICATION = {
39
 
40
  _BIBTEX = """
41
  @misc{zhang2024chinesesafechinesebenchmarkevaluating,
42
- title={ChineseSafe: A Chinese Benchmark for Evaluating Safety in Large Language Models},
43
  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},
44
  year={2024},
45
  eprint={2410.18491},
46
  archivePrefix={arXiv},
47
  primaryClass={cs.CL},
48
- url={https://arxiv.org/abs/2410.18491},
49
  }
50
  """
51
 
52
- _LAST_UPDATED = "July 28, 2025"
53
 
54
  banner_url = "./assets/logo.png"
55
  _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
@@ -65,31 +62,18 @@ def format_csv_numbers(text):
65
 
66
  def format_csv_numbers_second(text):
67
  return text.split()
68
-
69
-
70
  def format_number(x):
71
  return float(f"{x:.3}")
72
 
73
 
74
- def get_dataset_new_csv(
75
- model_size: List[str],
76
- ):
77
- df = ORIGINAL_DF_NEW[ORIGINAL_DF_NEW['Size'].isin(model_size)]
78
- df = df.drop(columns="Size")
79
-
80
- leaderboard_table = gr.components.Dataframe(
81
- value=df,
82
- interactive=False,
83
- visible=True,
84
- )
85
- return leaderboard_table
86
-
87
  def get_dataset_csv(
88
  model_size: List[str],
89
  ):
90
  df = ORIGINAL_DF[ORIGINAL_DF['Size'].isin(model_size)]
91
  df = df.drop(columns="Size")
92
-
93
  leaderboard_table = gr.components.Dataframe(
94
  value=df,
95
  interactive=False,
@@ -117,11 +101,11 @@ def get_dataset_csv_sub_gen(
117
  ):
118
  df = ORIGINAL_DF_SUB_GEN[ORIGINAL_DF_SUB_GEN['Size'].isin(model_size)]
119
  df = df.drop(columns="Size")
120
-
121
  # get subclass
122
  subclass_choice_label = ["Model", subclass_choice+"_Accuracy", subclass_choice+"_Precision", subclass_choice+"_Recall"]
123
  df = df[subclass_choice_label]
124
-
125
  leaderboard_table = gr.components.Dataframe(
126
  value=df,
127
  interactive=False,
@@ -136,11 +120,11 @@ def get_dataset_csv_sub_per(
136
  ):
137
  df = ORIGINAL_DF_SUB_PER[ORIGINAL_DF_SUB_PER['Size'].isin(model_size)]
138
  df = df.drop(columns="Size")
139
-
140
  # get subclass
141
  subclass_choice_label = ["Model", subclass_choice+"_Accuracy", subclass_choice+"_Precision", subclass_choice+"_Recall"]
142
  df = df[subclass_choice_label]
143
-
144
  leaderboard_table = gr.components.Dataframe(
145
  value=df,
146
  interactive=False,
@@ -159,18 +143,7 @@ def get_dataset_classfier_gen(
159
  subclass_choice = main_choice
160
  leaderboard_table = get_dataset_csv_sub_gen(model_size, subclass_choice)
161
  return leaderboard_table
162
-
163
- def get_ChineseGuardBench(
164
- model_size: List[str],
165
- main_choice: List[str],
166
- ):
167
- leaderboard_table = get_dataset_new_csv(model_size)
168
- # elif main_choice != "Subclass":
169
- # subclass_choice = main_choice
170
- # leaderboard_table = get_dataset_csv_sub_gen(model_size, subclass_choice)
171
- return leaderboard_table
172
-
173
-
174
  def get_dataset_classfier_per(
175
  model_size: List[str],
176
  main_choice: List[str],
@@ -191,10 +164,10 @@ with gr.Blocks() as demo:
191
 
192
  with gr.Row():
193
  gr.Markdown(METRICS_TEXT, elem_classes="markdown-text")
194
-
195
  with gr.Row():
196
  gr.Markdown(EVALUTION_TEXT, elem_classes="markdown-text")
197
-
198
  with gr.Row():
199
  with gr.Column(scale=0.8):
200
  main_choice = gr.Dropdown(
@@ -203,8 +176,8 @@ with gr.Blocks() as demo:
203
  label="Type",
204
  info="Please choose the type to display.",
205
  )
206
-
207
- with gr.Column(scale=10):
208
  model_choice = gr.CheckboxGroup(
209
  choices=CLASSIFICATION["model_size"],
210
  value=CLASSIFICATION["model_size"], # all be choosed
@@ -215,29 +188,24 @@ with gr.Blocks() as demo:
215
  #👉 this part is for csv table generatived
216
  with gr.Tabs(elem_classes="tab-buttons") as tabs:
217
  # ----------------- modify text -----------------
218
-
219
  with gr.TabItem("🏅 Generation", elem_id="od-benchmark-tab-table", id=6):
220
  dataframe_all_gen = gr.components.Dataframe(
221
  elem_id="leaderboard-table",
222
  )
223
-
224
  with gr.TabItem("🏅 Perplexity", elem_id="od-benchmark-tab-table", id=5):
225
  dataframe_all_per = gr.components.Dataframe(
226
  elem_id="leaderboard-table",
227
  )
228
-
229
- with gr.TabItem("🏅 NEW", elem_id="od-benchmark-tab-table", id=7):
230
- dataframe_all_guardbench = gr.components.Dataframe(
231
- elem_id="leaderboard-table",
232
- )
233
 
234
  # ----------------- modify text -----------------
235
  with gr.Row():
236
  gr.Markdown(ACKNOWLEDGEMENTS_TEXT, elem_classes="markdown-text")
237
-
238
  with gr.Row():
239
  gr.Markdown(REFERENCE_TEXT, elem_classes="markdown-text")
240
-
241
  # 👉 this part is for citation
242
  with gr.Row():
243
  with gr.Accordion("📙 Citation", open=True):
@@ -248,18 +216,18 @@ with gr.Blocks() as demo:
248
  elem_id="citation-button",
249
  show_copy_button=True
250
  )
251
-
252
  gr.Markdown(f"Last updated on **{_LAST_UPDATED}**", elem_classes="markdown-text")
253
-
254
  # --------------------------- all --------------------------------
255
  # this is all result Perplexity
256
-
257
  main_choice.change(
258
  get_dataset_classfier_per,
259
  inputs=[model_choice, main_choice],
260
  outputs=dataframe_all_per,
261
  )
262
-
263
  model_choice.change(
264
  get_dataset_classfier_per,
265
  inputs=[model_choice, main_choice],
@@ -271,44 +239,26 @@ with gr.Blocks() as demo:
271
  inputs=[model_choice, main_choice],
272
  outputs=dataframe_all_per,
273
  )
274
-
275
  # this is all result generatived
276
  main_choice.change(
277
  get_dataset_classfier_gen,
278
  inputs=[model_choice, main_choice],
279
  outputs=dataframe_all_gen,
280
  )
281
-
282
  model_choice.change(
283
  get_dataset_classfier_gen,
284
  inputs=[model_choice, main_choice],
285
  outputs=dataframe_all_gen,
286
  )
287
-
288
  demo.load(
289
  fn=get_dataset_classfier_gen,
290
  inputs=[model_choice, main_choice],
291
  outputs=dataframe_all_gen,
292
  )
293
-
294
- # this is new results for ChineseGuardBench
295
- # main_choice.change(
296
- # get_ChineseGuardBench,
297
- # inputs=[model_choice, main_choice],
298
- # outputs=dataframe_all_guardbench,
299
- # )
300
-
301
- model_choice.change(
302
- get_ChineseGuardBench,
303
- inputs=[model_choice, main_choice],
304
- outputs=dataframe_all_guardbench,
305
- )
306
-
307
- demo.load(
308
- fn=get_ChineseGuardBench,
309
- inputs=[model_choice, main_choice],
310
- outputs=dataframe_all_guardbench,
311
- )
312
-
313
  demo.launch(share=True)
314
 
 
6
  from assets.text import INTRODUCTION_TEXT, METRICS_TEXT, EVALUTION_TEXT, ACKNOWLEDGEMENTS_TEXT, REFERENCE_TEXT
7
 
8
 
9
+ ORIGINAL_DF = pd.read_csv("./data/chinese_benchmark_gen.csv", sep='\t') # space separated values
10
+ ORIGINAL_DF_PER = pd.read_csv("./data/chinese_benchmark_per.csv", sep='\t') #
11
 
12
+ ORIGINAL_DF_SUB_GEN = pd.read_csv("./data/subclass_gen.csv", sep=',') #
13
+ ORIGINAL_DF_SUB_PER = pd.read_csv("./data/subclass_per.csv", sep=',')
 
 
14
 
15
  METRICS = ["Accuracy", "Precision_Unsafe", "Recall_Unsafe", "Precision_Safe", "Recall_Safe", "None"]
16
+
17
 
18
  SUBCLASS = ["Discrimination", "Variant", "Psychology", "Politics", "Eroticism", "Vulgarity", "Property", "Injury", "Criminality", "Ethics"]
19
 
 
26
  "~30B",
27
  "10B~20B",
28
  "5B~10B",
 
29
  "API",
30
  ]
31
+
32
  }
33
 
34
 
 
36
 
37
  _BIBTEX = """
38
  @misc{zhang2024chinesesafechinesebenchmarkevaluating,
39
+ title={ChineseSafe: A Chinese Benchmark for Evaluating Safety in Large Language Models},
40
  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},
41
  year={2024},
42
  eprint={2410.18491},
43
  archivePrefix={arXiv},
44
  primaryClass={cs.CL},
45
+ url={https://arxiv.org/abs/2410.18491},
46
  }
47
  """
48
 
49
+ _LAST_UPDATED = "April 13, 2025"
50
 
51
  banner_url = "./assets/logo.png"
52
  _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
 
62
 
63
  def format_csv_numbers_second(text):
64
  return text.split()
65
+
66
+
67
  def format_number(x):
68
  return float(f"{x:.3}")
69
 
70
 
 
 
 
 
 
 
 
 
 
 
 
 
 
71
  def get_dataset_csv(
72
  model_size: List[str],
73
  ):
74
  df = ORIGINAL_DF[ORIGINAL_DF['Size'].isin(model_size)]
75
  df = df.drop(columns="Size")
76
+
77
  leaderboard_table = gr.components.Dataframe(
78
  value=df,
79
  interactive=False,
 
101
  ):
102
  df = ORIGINAL_DF_SUB_GEN[ORIGINAL_DF_SUB_GEN['Size'].isin(model_size)]
103
  df = df.drop(columns="Size")
104
+
105
  # get subclass
106
  subclass_choice_label = ["Model", subclass_choice+"_Accuracy", subclass_choice+"_Precision", subclass_choice+"_Recall"]
107
  df = df[subclass_choice_label]
108
+
109
  leaderboard_table = gr.components.Dataframe(
110
  value=df,
111
  interactive=False,
 
120
  ):
121
  df = ORIGINAL_DF_SUB_PER[ORIGINAL_DF_SUB_PER['Size'].isin(model_size)]
122
  df = df.drop(columns="Size")
123
+
124
  # get subclass
125
  subclass_choice_label = ["Model", subclass_choice+"_Accuracy", subclass_choice+"_Precision", subclass_choice+"_Recall"]
126
  df = df[subclass_choice_label]
127
+
128
  leaderboard_table = gr.components.Dataframe(
129
  value=df,
130
  interactive=False,
 
143
  subclass_choice = main_choice
144
  leaderboard_table = get_dataset_csv_sub_gen(model_size, subclass_choice)
145
  return leaderboard_table
146
+
 
 
 
 
 
 
 
 
 
 
 
147
  def get_dataset_classfier_per(
148
  model_size: List[str],
149
  main_choice: List[str],
 
164
 
165
  with gr.Row():
166
  gr.Markdown(METRICS_TEXT, elem_classes="markdown-text")
167
+
168
  with gr.Row():
169
  gr.Markdown(EVALUTION_TEXT, elem_classes="markdown-text")
170
+
171
  with gr.Row():
172
  with gr.Column(scale=0.8):
173
  main_choice = gr.Dropdown(
 
176
  label="Type",
177
  info="Please choose the type to display.",
178
  )
179
+
180
+ with gr.Column(scale=10):
181
  model_choice = gr.CheckboxGroup(
182
  choices=CLASSIFICATION["model_size"],
183
  value=CLASSIFICATION["model_size"], # all be choosed
 
188
  #👉 this part is for csv table generatived
189
  with gr.Tabs(elem_classes="tab-buttons") as tabs:
190
  # ----------------- modify text -----------------
191
+
192
  with gr.TabItem("🏅 Generation", elem_id="od-benchmark-tab-table", id=6):
193
  dataframe_all_gen = gr.components.Dataframe(
194
  elem_id="leaderboard-table",
195
  )
196
+
197
  with gr.TabItem("🏅 Perplexity", elem_id="od-benchmark-tab-table", id=5):
198
  dataframe_all_per = gr.components.Dataframe(
199
  elem_id="leaderboard-table",
200
  )
 
 
 
 
 
201
 
202
  # ----------------- modify text -----------------
203
  with gr.Row():
204
  gr.Markdown(ACKNOWLEDGEMENTS_TEXT, elem_classes="markdown-text")
205
+
206
  with gr.Row():
207
  gr.Markdown(REFERENCE_TEXT, elem_classes="markdown-text")
208
+
209
  # 👉 this part is for citation
210
  with gr.Row():
211
  with gr.Accordion("📙 Citation", open=True):
 
216
  elem_id="citation-button",
217
  show_copy_button=True
218
  )
219
+
220
  gr.Markdown(f"Last updated on **{_LAST_UPDATED}**", elem_classes="markdown-text")
221
+
222
  # --------------------------- all --------------------------------
223
  # this is all result Perplexity
224
+
225
  main_choice.change(
226
  get_dataset_classfier_per,
227
  inputs=[model_choice, main_choice],
228
  outputs=dataframe_all_per,
229
  )
230
+
231
  model_choice.change(
232
  get_dataset_classfier_per,
233
  inputs=[model_choice, main_choice],
 
239
  inputs=[model_choice, main_choice],
240
  outputs=dataframe_all_per,
241
  )
242
+
243
  # this is all result generatived
244
  main_choice.change(
245
  get_dataset_classfier_gen,
246
  inputs=[model_choice, main_choice],
247
  outputs=dataframe_all_gen,
248
  )
249
+
250
  model_choice.change(
251
  get_dataset_classfier_gen,
252
  inputs=[model_choice, main_choice],
253
  outputs=dataframe_all_gen,
254
  )
255
+
256
  demo.load(
257
  fn=get_dataset_classfier_gen,
258
  inputs=[model_choice, main_choice],
259
  outputs=dataframe_all_gen,
260
  )
261
+
262
+
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
263
  demo.launch(share=True)
264
 
assets/text.py CHANGED
@@ -35,8 +35,7 @@ EVALUTION_TEXT= """
35
  We evaluate the models using two methods: perplexity(multiple choice) and generation.
36
  For perplexity, we select the label which is the lowest perplexity as the predicted results.
37
  For generation, we use the content generated by the model to make prediction.
38
- The following are the results of the evaluation.
39
- 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.👇👇👇
40
  </span> <br><br>
41
 
42
 
 
35
  We evaluate the models using two methods: perplexity(multiple choice) and generation.
36
  For perplexity, we select the label which is the lowest perplexity as the predicted results.
37
  For generation, we use the content generated by the model to make prediction.
38
+ The following are the results of the evaluation. 👇👇👇
 
39
  </span> <br><br>
40
 
41
 
changelog.md CHANGED
@@ -52,18 +52,4 @@ version: v1.0.5
52
  - DeepSeek-R1-Distill-Llama-70B
53
  - Mistral-Small-24B-Instruct-2501
54
  - Moonlight-16B-A3B-Instruct
55
- - [2]feat: release a test set of 20000 samples
56
-
57
-
58
-
59
- ### 2025-7-1
60
- version: v1.0.6
61
-
62
- changed:
63
- - [1]feat: update many models due to the April's todo-list:
64
- - Llama-4-maverick
65
- - Gemini-2.5-flash-preview-05-20
66
- - Deepseek-chat-v3-0324
67
- - Qwen3
68
- - Gemma-3
69
- - OpenThinker2
 
52
  - DeepSeek-R1-Distill-Llama-70B
53
  - Mistral-Small-24B-Instruct-2501
54
  - Moonlight-16B-A3B-Instruct
55
+ - [2]feat: release a test set of 20000 samples
 
 
 
 
 
 
 
 
 
 
 
 
 
 
data/ChineseGuardBench.csv DELETED
@@ -1,33 +0,0 @@
1
- Model,Size,F1,Accuracy,Precision,Recall,FPR,FNR
2
- Deepexi-Guard-3B,1B~5B,89.63 ,89.72 ,85.53 ,94.15 ,14.24 ,5.85
3
- Qwen3-32B,~30B,88.54 ,89.25 ,89.08 ,88.02 ,9.64 ,11.98
4
- Qwen3-235B-A22B,>65B,87.92 ,88.96 ,90.86 ,85.17 ,7.66 ,14.83
5
- Qwen3-235B-A22B-Instruct-2507,>65B,87.81 ,89.13 ,93.27 ,82.96 ,5.35 ,17.04
6
- GLM-Z1-9B-0414,5B~10B,87.36 ,88.03 ,87.11 ,87.61 ,11.59 ,12.39
7
- Qwen2.5-72B-Instruct,>65B,86.81 ,88.27 ,92.50 ,81.79 ,5.93 ,18.21
8
- QwQ-32B,~30B,86.80 ,88.35 ,93.33 ,81.12 ,5.18 ,18.88
9
- Phi-4,10B~20B,85.95 ,86.88 ,86.90 ,85.02 ,11.45 ,14.98
10
- Gemma-3-27B-it,~30B,85.29 ,86.78 ,89.83 ,81.19 ,8.22 ,18.81
11
- DeepSeek-R1-0528,>65B,85.24 ,87.47 ,96.02 ,76.63 ,2.84 ,23.37
12
- Mistral-Small-3.2-24B-Instruct,~30B,85.07 ,87.03 ,93.14 ,78.29 ,5.15 ,21.71
13
- GLM-4-9B-chat,5B~10B,84.85 ,86.27 ,88.47 ,81.52 ,9.49 ,18.48
14
- MD-Judge-v0_2-internlm2_7B,5B~10B,84.63 ,85.88 ,87.03 ,82.37 ,10.98 ,17.63
15
- DeepSeek-R1-Distill-Qwen-32B,~30B,84.55 ,86.64 ,93.05 ,77.47 ,5.17 ,22.53
16
- Hunyuan-A13B-Instruct,>65B,84.32 ,86.21 ,90.97 ,78.58 ,6.98 ,21.42
17
- Moonlight-16B-A3B-Instruct,10B~20B,84.21 ,84.35 ,80.41 ,88.38 ,19.25 ,11.62
18
- GLM-Z1-32B-0414,~30B,83.40 ,85.75 ,92.63 ,75.85 ,5.40 ,24.15
19
- Qwen3-8B,5B~10B,83.05 ,85.51 ,92.69 ,75.23 ,5.30 ,24.77
20
- Qwen2.5-7B-Instruct,5B~10B,82.96 ,84.99 ,89.41 ,77.37 ,8.20 ,22.63
21
- Qwen2.5-1.5B-Instruct,1B~5B,79.48 ,77.08 ,68.83 ,94.03 ,38.07 ,5.97
22
- shieldgemma-2B,1B~5B,79.19 ,79.63 ,76.50 ,82.06 ,22.54 ,17.94
23
- Qwen2.5-3B-Instruct,1B~5B,79.05 ,77.57 ,70.69 ,89.66 ,33.25 ,10.34
24
- SHTEC_safety_fence_model_7B,5B~10B,78.44 ,82.48 ,93.54 ,67.54 ,4.17 ,32.46
25
- Qwen3-4B,1B~5B,78.16 ,82.50 ,95.12 ,66.33 ,3.04 ,33.67
26
- SmolLM3-3B,1B~5B,76.10 ,79.19 ,83.09 ,70.19 ,12.77 ,29.81
27
- ERNIE-4.5-21B-A3B-Paddle,~20B,75.21 ,80.58 ,94.58 ,62.42 ,3.20 ,37.58
28
- Qwen3-1.7B,1B~5B,74.46 ,79.34 ,89.36 ,63.82 ,6.79 ,36.18
29
- internlm2_5-7B-chat,5B~10B,71.52 ,78.49 ,95.34 ,57.22 ,2.50 ,42.78
30
- Llama-Guard-4-12B,10B~20B,65.66 ,74.64 ,90.99 ,51.36 ,4.54 ,48.64
31
- Llama-Guard-3-8B,5B~10B,59.33 ,72.44 ,97.80 ,42.58 ,0.86 ,57.42
32
- DeepSeek-R1-Distill-Qwen-7B,5B~10B,45.27 ,65.53 ,90.36 ,30.20 ,2.88 ,69.80
33
- Gemma-3n-E4B-it,5B~10B,44.05 ,64.88 ,88.80 ,29.29 ,3.30 ,70.71
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
data/chinese_benchmark_gen.csv CHANGED
@@ -1,56 +1,43 @@
1
- Model,Size,Accuracy/std,Precision_Unsafe/std,Recall_Unsafe/std,Precision_Safe/std,Recall_Safe/std
2
- GPT-4o,API,73.78/0.30,97.75/0.13,48.66/0.04,65.84/0.55,98.88/0.04
3
- GPT-4-Turbo,API,71.67/0.17,80.13/0.64,57.59/0.69,66.93/0.44,85.74/0.35
4
- Pespective,API,69.28/0.32,69.96/0.79,67.49/0.32,68.64/0.32,71.06/0.43
5
- GPT-3.5,API,64.70/0.44,76.12/0.55,42.79/0.64,60.24/0.76,86.59/0.32
6
- Gemini-2.5-flash-preview-05-20,API,71.27/0.27,73.40/0.23,70.16/0.71,69.17/0.53,72.48/0.40
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
8
- 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
9
- 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
10
- 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
11
- 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
12
- 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
13
- 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
14
- 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
15
- GLM-4-9B-Chat,5B~10B,70.96/0.23,82.15/0.55,53.73/0.48,65.50/0.18,88.27/0.41
16
- Mistral-7B-Instruct-v0.3,5B~10B,70.41/0.41,68.55/0.52,75.67/0.22,72.71/0.26,65.12/0.58
17
- Qwen1.5-7B-Chat,5B~10B,70.36/0.39,64.66/0.27,90.09/0.57,83.55/0.82,50.53/0.18
18
- Phi-3-small-128k-instruct,5B~10B,67.43/0.26,72.10/0.54,57.35/0.17,64.33/0.09,77.61/0.43
19
- Ministral-8B-Instruct-2410,5B~10B,62.32/0.01,62.71/0.19,61.60/0.29,61.94/0.19,63.05/0.28
20
- Yi-1.5-9B-Chat,5B~10B,62.12/0.38,64.42/0.42,54.53/0.43,60.43/0.36,69.75/0.37
21
- Llama3-ChatQA-1.5-8B,5B~10B,61.28/0.40,57.63/0.20,85.84/0.43,72.02/0.95,36.61/0.54
22
- Baichuan2-7B-Chat,5B~10B,59.43/0.24,72.06/0.66,31.11/0.40,55.95/0.12,87.89/0.20
23
- InternLM2-chat-7B,5B~10B,58.79/0.09,62.70/0.19,43.88/0.17,56.68/0.14,73.77/0.13
24
- GPT-J-6B,5B~10B,52.65/0.32,52.42/0.32,62.00/0.42,52.99/0.37,43.21/0.92
25
- Opt-6.7B,5B~10B,50.00/0.11,50.17/0.17,64.70/0.35,49.69/0.04,35.18/0.44
26
- Qwen3-4B,5B~10B,74.95/0.01,76.47/0.01,72.10/0.00,73.61/0.01,77.81/0.01
27
- Gemma-3-4B-it,5B~10B,71.41/0.00,66.54/0.00,86.12/0.00,80.33/0.00,56.70/0.00
28
- phi-4,10B~20B,72.24/0.24,76.59/0.46,64.42/0.51,69.06/0.15,80.13/0.62
29
- InternLM2-Chat-20B,10B~20B,70.21/0.55,73.30/0.70,63.79/0.43,67.82/0.45,76.65/0.67
30
- Qwen1.5-14B-Chat,10B~20B,68.25/0.44,65.87/0.37,76.02/0.72,71.51/0.59,60.44/0.20
31
- Phi-3-medium-128k-instruct,10B~20B,64.30/0.06,63.89/0.13,66.53/0.52,64.76/0.26,62.05/0.42
32
- Baichuan2-13B-Chat,10B~20B,62.86/0.31,64.17/0.33,58.61/0.80,61.75/0.30,67.13/0.56
33
- Mistral-Nemo-Instruct-2407,10B~20B,59.71/0.45,61.79/0.52,51.82/0.48,58.20/0.44,67.68/0.44
34
- Phi-3-medium-4k-instruct,10B~20B,57.79/0.45,58.69/0.37,53.88/0.62,57.02/0.55,61.74/0.55
35
- Ziya2-13B-Chat,10B~20B,53.40/0.43,53.33/0.38,56.18/0.41,53.48/0.53,50.62/0.61
36
- Opt-13B,10B~20B,50.18/0.26,50.29/0.20,69.97/0.37,49.94/0.47,30.22/0.31
37
- Moonlight-16B-A3B-Instruct,10B~20B,45.16/0.43,44.16/0.64,34.79/0.67,45.82/0.33,55.62/0.35
38
- Qwen3-14B,10B~20B,68.54/0.01,67.24/0.01,72.29/0.00,70.04/0.00,64.78/0.01
39
- Gemma-3-12B-it,10B~20B,65.63/0.00,62.69/0.00,77.18/0.00,70.32/0.00,54.07/0.00
40
- DeepSeek-LLM-67B-Chat,>65B,76.76/0.35,73.40/0.37,84.26/0.40,81.34/0.35,69.19/0.64
41
- Llama3-ChatQA-1.5-70B,>65B,65.29/0.29,66.24/0.50,62.92/0.12,64.43/0.19,67.69/0.63
42
- Qwen2.5-72B-Instruct,>65B,63.41/0.77,66.00/0.95,56.00/0.62,61.49/0.65,70.90/0.96
43
- Qwen1.5-72B-Chat,>65B,62.91/0.50,73.86/0.84,40.46/0.97,58.75/0.35,85.55/0.62
44
- Opt-66B,>65B,54.46/0.17,53.22/0.06,76.94/0.24,57.73/0.49,31.77/0.28
45
- Qwen2-72B-Instruct,>65B,54.08/0.20,58.10/0.60,30.72/0.45,52.63/0.05,77.65/0.36
46
- DeepSeek-R1-Distill-Llama-70B,>65B,52.93/0.18,59.69/0.47,19.33/0.38,51.62/0.16,86.83/0.18
47
- Llama-3.1-70B-Instruct,>65B,52.84/0.38,59.07/1.22,19.82/0.85,51.57/0.24,86.14/0.58
48
- Llama-3.3-70B-Instruct,>65B,50.87/0.07,54.51/0.86,13.19/0.10,50.37/0.06,88.89/0.39
49
- Qwen3-32B,>65B,75.26/0.00,89.11/0.00,57.55/0.0,68.65/0.00,92.97/0.00
50
- Qwen2.5-32B-Instruct,~30B,69.64/0.39,92.13/0.45,43.24/0.83,62.70/0.25,96.27/0.20
51
- QwQ-32B-Preview,~30B,69.55/0.28,75.97/0.48,57.60/0.27,65.61/0.17,81.62/0.33
52
- Mistral-Small-24B-Instruct-2501,~30B,64.48/0.17,64.61/0.35,64.71/0.72,64.34/0.00,64.23/1.04
53
- Yi-1.5-34B-Chat,~30B,60.06/0.43,58.14/0.40,72.51/0.55,63.27/0.56,47.56/0.42
54
- Opt-30B,~30B,50.88/0.11,50.76/0.12,72.95/0.16,51.18/0.26,28.62/0.28
55
- Gemma-3-27B-it,~30B,68.50/0.00,68.37/0.00,68.84/0.00,68.62/0.00,68.15/0.00
56
- OpenThinker2-32B,~30B,65.01/0.01,74.90/0.01,45.13/0.01,60.74/0.01,84.87/0.00
 
1
+ Model Size Accuracy/std Precision_Unsafe/std Recall_Unsafe/std Precision_Safe/std Recall_Safe/std
2
+ DeepSeek-LLM-67B-Chat >65B 76.76/0.35 73.40/0.37 84.26/0.40 81.34/0.35 69.19/0.64
3
+ Llama3-ChatQA-1.5-70B >65B 65.29/0.29 66.24/0.50 62.92/0.12 64.43/0.19 67.69/0.63
4
+ Qwen2.5-72B-Instruct >65B 63.41/0.77 66.00/0.95 56.00/0.62 61.49/0.65 70.90/0.96
5
+ Qwen1.5-72B-Chat >65B 62.91/0.50 73.86/0.84 40.46/0.97 58.75/0.35 85.55/0.62
6
+ Opt-66B >65B 54.46/0.17 53.22/0.06 76.94/0.24 57.73/0.49 31.77/0.28
7
+ Qwen2-72B-Instruct >65B 54.08/0.20 58.10/0.60 30.72/0.45 52.63/0.05 77.65/0.36
8
+ DeepSeek-R1-Distill-Llama-70B >65B 52.93/0.18 59.69/0.47 19.33/0.38 51.62/0.16 86.83/0.18
9
+ Llama-3.1-70B-Instruct >65B 52.84/0.38 59.07/1.22 19.82/0.85 51.57/0.24 86.14/0.58
10
+ Llama-3.3-70B-Instruct >65B 50.87/0.07 54.51/0.86 13.19/0.10 50.37/0.06 88.89/0.39
11
+ Qwen2.5-32B-Instruct ~30B 69.64/0.39 92.13/0.45 43.24/0.83 62.70/0.25 96.27/0.20
12
+ QwQ-32B-Preview ~30B 69.55/0.28 75.97/0.48 57.60/0.27 65.61/0.17 81.62/0.33
13
+ Mistral-Small-24B-Instruct-2501 ~30B 64.48/0.17 64.61/0.35 64.71/0.72 64.34/0.00 64.23/1.04
14
+ Yi-1.5-34B-Chat ~30B 60.06/0.43 58.14/0.40 72.51/0.55 63.27/0.56 47.56/0.42
15
+ Opt-30B ~30B 50.88/0.11 50.76/0.12 72.95/0.16 51.18/0.26 28.62/0.28
16
+ phi-4 10B~20B 72.24/0.24 76.59/0.46 64.42/0.51 69.06/0.15 80.13/0.62
17
+ InternLM2-Chat-20B 10B~20B 70.21/0.55 73.30/0.70 63.79/0.43 67.82/0.45 76.65/0.67
18
+ Qwen1.5-14B-Chat 10B~20B 68.25/0.44 65.87/0.37 76.02/0.72 71.51/0.59 60.44/0.20
19
+ Phi-3-medium-128k-instruct 10B~20B 64.30/0.06 63.89/0.13 66.53/0.52 64.76/0.26 62.05/0.42
20
+ Baichuan2-13B-Chat 10B~20B 62.86/0.31 64.17/0.33 58.61/0.80 61.75/0.30 67.13/0.56
21
+ Mistral-Nemo-Instruct-2407 10B~20B 59.71/0.45 61.79/0.52 51.82/0.48 58.20/0.44 67.68/0.44
22
+ Phi-3-medium-4k-instruct 10B~20B 57.79/0.45 58.69/0.37 53.88/0.62 57.02/0.55 61.74/0.55
23
+ Ziya2-13B-Chat 10B~20B 53.40/0.43 53.33/0.38 56.18/0.41 53.48/0.53 50.62/0.61
24
+ Opt-13B 10B~20B 50.18/0.26 50.29/0.20 69.97/0.37 49.94/0.47 30.22/0.31
25
+ Moonlight-16B-A3B-Instruct 10B~20B 45.16/0.43 44.16/0.64 34.79/0.67 45.82/0.33 55.62/0.35
26
+ 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
27
+ 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
28
+ 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
29
+ GLM-4-9B-Chat 5B~10B 70.96/0.23 82.15/0.55 53.73/0.48 65.50/0.18 88.27/0.41
30
+ Mistral-7B-Instruct-v0.3 5B~10B 70.41/0.41 68.55/0.52 75.67/0.22 72.71/0.26 65.12/0.58
31
+ Qwen1.5-7B-Chat 5B~10B 70.36/0.39 64.66/0.27 90.09/0.57 83.55/0.82 50.53/0.18
32
+ Phi-3-small-128k-instruct 5B~10B 67.43/0.26 72.10/0.54 57.35/0.17 64.33/0.09 77.61/0.43
33
+ Ministral-8B-Instruct-2410 5B~10B 62.32/0.01 62.71/0.19 61.60/0.29 61.94/0.19 63.05/0.28
34
+ Yi-1.5-9B-Chat 5B~10B 62.12/0.38 64.42/0.42 54.53/0.43 60.43/0.36 69.75/0.37
35
+ Llama3-ChatQA-1.5-8B 5B~10B 61.28/0.40 57.63/0.20 85.84/0.43 72.02/0.95 36.61/0.54
36
+ Baichuan2-7B-Chat 5B~10B 59.43/0.24 72.06/0.66 31.11/0.40 55.95/0.12 87.89/0.20
37
+ InternLM2-chat-7B 5B~10B 58.79/0.09 62.70/0.19 43.88/0.17 56.68/0.14 73.77/0.13
38
+ GPT-J-6B 5B~10B 52.65/0.32 52.42/0.32 62.00/0.42 52.99/0.37 43.21/0.92
39
+ Opt-6.7B 5B~10B 50.00/0.11 50.17/0.17 64.70/0.35 49.69/0.04 35.18/0.44
40
+ GPT-4o API 73.78/0.30 97.75/0.13 48.66/0.04 65.84/0.55 98.88/0.04
41
+ GPT-4-Turbo API 71.67/0.17 80.13/0.64 57.59/0.69 66.93/0.44 85.74/0.35
42
+ Pespective API 69.28/0.32 69.96/0.79 67.49/0.32 68.64/0.32 71.06/0.43
43
+ GPT-3.5 API 64.70/0.44 76.12/0.55 42.79/0.64 60.24/0.76 86.59/0.32
 
 
 
 
 
 
 
 
 
 
 
 
 
data/chinese_benchmark_per.csv CHANGED
@@ -1,46 +1,39 @@
1
- Model,Size,Accuracy/std,Precision_Unsafe/std,Recall_Unsafe/std,Precision_Safe/std,Recall_Safe/std
2
- Yi-1.5-34B-Chat,~30B,66.02/0.22,80.13/0.55,42.82/0.25,60.86/0.16,89.33/0.41
3
- Qwen2.5-32B-Instruct,~30B,64.33/0.46,62.46/0.44,72.24/0.71,66.91/0.53,56.38/0.18
4
- Opt-30B,~30B,53.82/0.03,54.42/0.21,48.32/0.20,53.34/0.11,59.34/0.27
5
- QwQ-32B-Preview,~30B,51.82/0.06,51.04/0.10,94.83/0.28,62.38/0.26,8.61/0.39
6
- Gemma-3-27B-it,~30B,50.00/0.00,0.00/0.00,0.00/0.00,50.00/0.00,100.00/0.00
7
- Qwen3-32B,~30B,49.66/0.00,49.83/0.00,99.03/0.00,22.40/0.00,0.28/0.00
8
- OpenThinker2-32B,~30B,49.91/0.00,49.95/0.00,98.26/0.00,47.27/0.00,1.56/0.00
9
- DeepSeek-LLM-67B-Chat,>65B,68.08/0.35,94.80/0.83,38.40/0.43,61.27/0.26,97.88/0.36
10
- Qwen1.5-72B-Chat,>65B,63.67/0.46,58.27/0.32,96.84/0.13,90.51/0.57,30.34/0.80
11
- Qwen2.5-72B-Instruct,>65B,63.27/0.52,66.00/0.60,55.09/0.82,61.31/0.46,71.49/0.25
12
- Qwen2-72B-Instruct,>65B,60.70/0.49,57.90/0.42,79.03/0.63,66.75/0.77,42.28/0.43
13
- Opt-66B,>65B,59.93/0.41,56.52/0.37,86.87/0.59,71.36/0.78,32.86/0.74
14
- DeepSeek-R1-Distill-Llama-70B,>65B,47.68/0.64,45.77/1.21,23.85/0.67,48.35/0.46,71.62/0.60
15
- Llama-3.1-70B-Instruct,>65B,43.68/0.41,36.45/0.84,16.66/0.34,45.83/0.30,70.82/0.48
16
- Llama3-ChatQA-1.5-70B,>65B,40.41/0.29,33.86/0.75,19.84/0.75,43.13/0.25,61.08/0.37
17
- Llama-3.3-70B-Instruct,>65B,36.84/0.82,32.02/1.29,23.19/1.13,39.58/0.63,50.55/0.69
18
- Phi-3-medium-4k-instruct,10B~20B,71.04/0.31,69.74/0.29,74.56/0.97,72.54/0.59,67.49/0.89
19
- Baichuan2-13B-Chat,10B~20B,70.43/0.39,65.81/0.38,85.34/0.63,79.02/0.63,55.46/0.47
20
- Phi-3-medium-128k-instruct,10B~20B,68.87/0.81,68.08/0.51,71.32/1.44,69.75/1.17,66.41/0.57
21
- Mistral-Nemo-Instruct-2407,10B~20B,66.88/0.46,62.56/0.28,84.42/0.90,75.89/1.13,49.26/0.24
22
- phi-4,10B~20B,62.62/0.32,63.73/0.41,58.98/0.20,61.66/0.31,66.28/0.78
23
- Qwen1.5-14B-Chat,10B~20B,61.29/0.40,57.02/0.32,92.43/0.55,79.80/1.05,30.02/0.47
24
- Mistral-Small-24B-Instruct-2501,10B~20B,59.20/0.46,58.32/0.42,65.16/1.08,60.33/0.56,53.22/0.20
25
- Ziya2-13B-Chat,10B~20B,55.25/0.26,59.24/0.37,34.30/0.11,53.61/0.26,76.29/0.39
26
- InternLM2-Chat-20B,10B~20B,53.67/0.16,79.00/0.66,10.30/0.60,51.90/0.11,97.25/0.26
27
- Opt-13B,10B~20B,49.31/0.31,37.77/3.57,1.76/0.16,49.59/0.23,97.08/0.29
28
- Moonlight-16B-A3B-Instruct,10B~20B,48.92/0.16,3.46/0.57,0.07/0.01,49.40/0.15,98.00/0.08
29
- Qwen3-14B,10B~20B,48.34/0.00,49.14/0.00,95.13/0.00,24.26/0.00,1.56/0.00
30
- Gemma-3-12B-it,10B~20B,50.00/0.00,0.00/0.00,0.00/0.00,50.00/0.00,100.00/0.00
31
- Gemma-1.1-7B-it,5B~10B,64.32/0.68,59.98/0.58,86.60/0.35,75.70/0.80,41.95/0.93
32
- Qwen1.5-7B-Chat,5B~10B,62.48/0.54,59.06/0.48,81.92/0.50,70.28/0.65,42.96/0.81
33
- Phi-3-small-128k-instruct,5B~10B,61.76/0.27,60.47/0.16,68.45/0.61,63.46/0.50,55.05/0.61
34
- Yi-1.5-9B-Chat,5B~10B,60.35/0.52,79.47/1.37,28.16/0.33,56.22/0.39,92.69/0.59
35
- Phi-3-small-8k-instruct,5B~10B,59.47/0.39,56.25/0.30,86.06/0.40,70.05/0.85,32.75/0.49
36
- DeepSeek-LLM-7B-Chat,5B~10B,56.79/0.19,84.83/1.23,16.77/0.09,53.70/0.15,96.99/0.27
37
- Ministral-8B-Instruct-2410,5B~10B,56.28/0.51,55.10/0.51,68.83/0.58,58.24/0.51,43.66/0.54
38
- GPT-J-6B,5B~10B,55.98/0.42,80.27/1.42,16.11/0.86,53.26/0.23,96.03/0.20
39
- Baichuan2-7B-Chat,5B~10B,53.99/0.51,62.89/1.57,19.96/0.88,52.31/0.30,88.18/0.23
40
- GLM-4-9B-Chat,5B~10B,50.03/0.15,50.07/0.13,99.31/0.22,44.12/9.01,0.52/0.04
41
- InternLM2-Chat-7B,5B~10B,49.49/0.11,42.16/1.58,2.15/0.31,49.68/0.13,97.06/0.25
42
- 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
43
- 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
44
- 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
45
- Qwen3-4B,5B~10B,46.04/0.00,47.79/0.00,85.94/0.00,30.39/0.00,6.14/0.00
46
- 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
 
1
+ Model Size Accuracy/std Precision_Unsafe/std Recall_Unsafe/std Precision_Safe/std Recall_Safe/std
2
+ DeepSeek-LLM-67B-Chat >65B 68.08/0.35 94.80/0.83 38.40/0.43 61.27/0.26 97.88/0.36
3
+ Qwen1.5-72B-Chat >65B 63.67/0.46 58.27/0.32 96.84/0.13 90.51/0.57 30.34/0.80
4
+ Qwen2.5-72B-Instruct >65B 63.27/0.52 66.00/0.60 55.09/0.82 61.31/0.46 71.49/0.25
5
+ Qwen2-72B-Instruct >65B 60.70/0.49 57.90/0.42 79.03/0.63 66.75/0.77 42.28/0.43
6
+ Opt-66B >65B 59.93/0.41 56.52/0.37 86.87/0.59 71.36/0.78 32.86/0.74
7
+ DeepSeek-R1-Distill-Llama-70B >65B 47.68/0.64 45.77/1.21 23.85/0.67 48.35/0.46 71.62/0.60
8
+ Llama-3.1-70B-Instruct >65B 43.68/0.41 36.45/0.84 16.66/0.34 45.83/0.30 70.82/0.48
9
+ Llama3-ChatQA-1.5-70B >65B 40.41/0.29 33.86/0.75 19.84/0.75 43.13/0.25 61.08/0.37
10
+ Llama-3.3-70B-Instruct >65B 36.84/0.82 32.02/1.29 23.19/1.13 39.58/0.63 50.55/0.69
11
+ Yi-1.5-34B-Chat ~30B 66.02/0.22 80.13/0.55 42.82/0.25 60.86/0.16 89.33/0.41
12
+ Qwen2.5-32B-Instruct ~30B 64.33/0.46 62.46/0.44 72.24/0.71 66.91/0.53 56.38/0.18
13
+ Opt-30B ~30B 53.82/0.03 54.42/0.21 48.32/0.20 53.34/0.11 59.34/0.27
14
+ QwQ-32B-Preview ~30B 51.82/0.06 51.04/0.10 94.83/0.28 62.38/0.26 8.61/0.39
15
+ Phi-3-medium-4k-instruct 10B~20B 71.04/0.31 69.74/0.29 74.56/0.97 72.54/0.59 67.49/0.89
16
+ Baichuan2-13B-Chat 10B~20B 70.43/0.39 65.81/0.38 85.34/0.63 79.02/0.63 55.46/0.47
17
+ Phi-3-medium-128k-instruct 10B~20B 68.87/0.81 68.08/0.51 71.32/1.44 69.75/1.17 66.41/0.57
18
+ Mistral-Nemo-Instruct-2407 10B~20B 66.88/0.46 62.56/0.28 84.42/0.90 75.89/1.13 49.26/0.24
19
+ phi-4 10B~20B 62.62/0.32 63.73/0.41 58.98/0.20 61.66/0.31 66.28/0.78
20
+ Qwen1.5-14B-Chat 10B~20B 61.29/0.40 57.02/0.32 92.43/0.55 79.80/1.05 30.02/0.47
21
+ Mistral-Small-24B-Instruct-2501 10B~20B 59.20/0.46 58.32/0.42 65.16/1.08 60.33/0.56 53.22/0.20
22
+ Ziya2-13B-Chat 10B~20B 55.25/0.26 59.24/0.37 34.30/0.11 53.61/0.26 76.29/0.39
23
+ InternLM2-Chat-20B 10B~20B 53.67/0.16 79.00/0.66 10.30/0.60 51.90/0.11 97.25/0.26
24
+ Opt-13B 10B~20B 49.31/0.31 37.77/3.57 1.76/0.16 49.59/0.23 97.08/0.29
25
+ Moonlight-16B-A3B-Instruct 10B~20B 48.92/0.16 3.46/0.57 0.07/0.01 49.40/0.15 98.00/0.08
26
+ Gemma-1.1-7B-it 5B~10B 64.32/0.68 59.98/0.58 86.60/0.35 75.70/0.80 41.95/0.93
27
+ Qwen1.5-7B-Chat 5B~10B 62.48/0.54 59.06/0.48 81.92/0.50 70.28/0.65 42.96/0.81
28
+ Phi-3-small-128k-instruct 5B~10B 61.76/0.27 60.47/0.16 68.45/0.61 63.46/0.50 55.05/0.61
29
+ Yi-1.5-9B-Chat 5B~10B 60.35/0.52 79.47/1.37 28.16/0.33 56.22/0.39 92.69/0.59
30
+ Phi-3-small-8k-instruct 5B~10B 59.47/0.39 56.25/0.30 86.06/0.40 70.05/0.85 32.75/0.49
31
+ DeepSeek-LLM-7B-Chat 5B~10B 56.79/0.19 84.83/1.23 16.77/0.09 53.70/0.15 96.99/0.27
32
+ Ministral-8B-Instruct-2410 5B~10B 56.28/0.51 55.10/0.51 68.83/0.58 58.24/0.51 43.66/0.54
33
+ GPT-J-6B 5B~10B 55.98/0.42 80.27/1.42 16.11/0.86 53.26/0.23 96.03/0.20
34
+ Baichuan2-7B-Chat 5B~10B 53.99/0.51 62.89/1.57 19.96/0.88 52.31/0.30 88.18/0.23
35
+ GLM-4-9B-Chat 5B~10B 50.03/0.15 50.07/0.13 99.31/0.22 44.12/9.01 0.52/0.04
36
+ InternLM2-Chat-7B 5B~10B 49.49/0.11 42.16/1.58 2.15/0.31 49.68/0.13 97.06/0.25
37
+ 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
38
+ 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
39
+ 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
 
 
 
 
 
 
 
data/subclass_gen.csv CHANGED
@@ -1,8 +1,28 @@
1
  Model,Size,Discrimination_Accuracy,Discrimination_Precision,Discrimination_Recall,Variant_Accuracy,Variant_Precision,Variant_Recall,Psychology_Accuracy,Psychology_Precision,Psychology_Recall,Politics_Accuracy,Politics_Precision,Politics_Recall,Eroticism_Accuracy,Eroticism_Precision,Eroticism_Recall,Vulgarity_Accuracy,Vulgarity_Precision,Vulgarity_Recall,Property_Accuracy,Property_Precision,Property_Recall,Injury_Accuracy,Injury_Precision,Injury_Recall,Criminality_Accuracy,Criminality_Precision,Criminality_Recall,Ethics_Accuracy,Ethics_Precision,Ethics_Recall
2
- Gemini-2.5-flash-preview-05-20,API,0.8125,0.8333,0.475,0.8784,0.8333,0.8542,0.7414,0.6034,0.8286,0.4482,0.85,1,0.3,0.973,0.9583,1,0.7241,0.4483,0.8857,0.1379,0.7907,0.75,0.4615,0.8182,0.7667,0.7742,0.75,0.65,0.7949,0.3636
3
- Llama-4-maverick,API,0.7785,0.9175,0.459,0.875,0.8765,0.917,0.689,0.524,0.8645,0.5275,0.711,0.989,0.072,0.904,0.907,0.988,0.532,0.202,0.883,0.209,0.8219,0.8653,0.3186,0.854,0.8549,0.8651,0.7755,0.5674,0.8515,0.5758
4
- Deepseek-chat-v3-0324,API,0.78,0.9115,0.444,0.874,0.8525,0.905,0.6835,0.51,0.857,0.5115,0.727,0.99,0.055,0.915,0.872,0.977,0.534,0.187,0.881,0.19,0.8132,0.8557,0.2477,0.8457,0.8393,0.854,0.7618,0.5282,0.8406,0.5322
5
- Gemini-2.0-flash,API,0.7345,0.767,0.781,0.8035,0.4935,0.4455,0.669,0.3705,0.7688,0.8055,0.852,0.917,0.945,0.99,0.37,0.274,0.721,0.124,0.9209,0.994,0.6899,0.7054,0.7116,0.721,0.4914,0.417,0.6531,0.2446,0.7061,0.7219
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6
  Gemma-1.1-7B-it,5B~10B,0.7849,0.7205,0.9139,0.8081,0.7454,0.9485,0.6024,0.6084,0.5413,0.7854,0.758,0.8894,0.8017,0.7436,0.9353,0.8215,0.7367,0.9884,0.6669,0.6543,0.673,0.5811,0.5858,0.4976,0.7831,0.7167,0.9127,0.6684,0.6638,0.6754
7
  Qwen1.5-7B-Chat,5B~10B,0.6885,0.6347,0.8535,0.7677,0.6891,0.9938,0.6929,0.6404,0.8588,0.7791,0.7151,0.9869,0.7653,0.6889,0.988,0.7485,0.6659,0.9746,0.684,0.6317,0.8443,0.7267,0.6564,0.929,0.7473,0.662,0.9772,0.5545,0.5496,0.5778
8
  Yi-1.5-9B-Chat,5B~10B,0.7025,0.6913,0.7058,0.7032,0.7106,0.707,0.4533,0.3925,0.2,0.6546,0.7097,0.6172,0.7209,0.7213,0.7419,0.8197,0.7508,0.9452,0.5595,0.5666,0.4131,0.4342,0.3378,0.1591,0.7626,0.7215,0.8306,0.4057,0.2654,0.1096
@@ -15,37 +35,5 @@ Opt-6.7B,5B~10B,0.4717,0.4691,0.6091,0.5087,0.5153,0.6691,0.4931,0.4895,0.6491,0
15
  Mistral-7B-Instruct-v0.3,5B~10B,0.7069,0.6749,0.7706,0.7521,0.7161,0.8533,0.5826,0.5868,0.5167,0.7142,0.7222,0.7711,0.7599,0.7205,0.8679,0.7956,0.7205,0.9509,0.6748,0.6547,0.7042,0.6139,0.6127,0.5802,0.7742,0.7074,0.9103,0.6388,0.6387,0.6313
16
  Llama3-ChatQA-1.5-8B,5B~10B,0.6114,0.5657,0.8761,0.6276,0.5904,0.885,0.5978,0.5613,0.844,0.6056,0.6016,0.8128,0.6113,0.5825,0.8521,0.6365,0.5805,0.9258,0.6062,0.5625,0.8663,0.6034,0.5629,0.8569,0.6223,0.5694,0.903,0.5658,0.5447,0.7752
17
  Ministral-8B-Instruct-2410,5B~10B,0.6447,0.6342,0.6442,0.7197,0.7001,0.7911,0.5176,0.5149,0.3869,0.6868,0.7082,0.7217,0.7326,0.7075,0.8161,0.7362,0.6919,0.8305,0.5742,0.5735,0.5003,0.4649,0.4306,0.2781,0.6894,0.6614,0.7369,0.5258,0.5313,0.4059
18
- Phi-3-small-8k-instruct,5B~10B,0.7598,0.7484,0.7666,0.7738,0.7711,0.7936,0.7227,0.7317,0.6914,0.7477,0.7825,0.7432,0.7999,0.7827,0.8445,0.8204,0.7762,0.8907,0.7106,0.7202,0.6662,0.6027,0.6353,0.4468,0.7871,0.7581,0.8233,0.593,0.6349,0.4325
19
- Phi-3-small-128k-instruct,5B~10B,0.7158,0.7404,0.6454,0.6831,0.7398,0.5872,0.6751,0.7171,0.5633,0.6057,0.7155,0.4565,0.6783,0.7379,0.5785,0.8131,0.7893,0.8433,0.6832,0.7183,0.5779,0.6189,0.6671,0.4487,0.7353,0.7481,0.6836,0.5863,0.6398,0.3873
20
- Qwen3-4B,5B~10B,0.7069,0.6749,0.7706,0.7521,0.7161,0.8533,0.5826,0.5868,0.5167,0.7142,0.6114,0.5657,0.8761,0.6276,0.5904,0.885,0.5978,0.5613,0.844,0.6056,0.6447,0.6342,0.6442,0.7197,0.7001,0.7911,0.5176,0.5149,0.3869,0.6868
21
- Qwen3-8B,5B~10B,0.6114,0.5657,0.8761,0.6276,0.5904,0.885,0.5978,0.5613,0.844,0.6056,0.6447,0.6342,0.6442,0.7197,0.7001,0.7911,0.5176,0.5149,0.3869,0.6868,0.7598,0.7484,0.7666,0.7738,0.7711,0.7936,0.7227,0.7317,0.6914,0.7477
22
- Gemma-3-4b-it,5B~10B,0.6447,0.6342,0.6442,0.7197,0.7001,0.7911,0.5176,0.5149,0.3869,0.6868,0.7598,0.7484,0.7666,0.7738,0.7711,0.7936,0.7227,0.7317,0.6914,0.7477,0.7158,0.7404,0.6454,0.6831,0.7398,0.5872,0.6751,0.7171,0.5633,0.6057
23
- Mistral-Small-24B-Instruct-2501,10B~20B,0.6626,0.6491,0.6746,0.7897,0.7347,0.9223,0.399,0.2824,0.1406,0.7649,0.7465,0.8603,0.7828,0.7326,0.9081,0.8088,0.728,0.9732,0.601,0.6001,0.549,0.4367,0.3723,0.2159,0.7369,0.6906,0.8282,0.4868,0.4773,0.3217
24
- Baichuan2-13B-Chat,10B~20B,0.6337,0.6402,0.5755,0.7188,0.7164,0.7457,0.5185,0.5189,0.3417,0.7341,0.7487,0.7703,0.7033,0.7091,0.7143,0.6742,0.6712,0.6575,0.5657,0.5728,0.434,0.6151,0.6264,0.5371,0.6515,0.65,0.6089,0.5532,0.5707,0.414
25
- Qwen1.5-14B-Chat,10B~20B,0.7099,0.6657,0.8141,0.7897,0.7205,0.9615,0.5669,0.5657,0.5226,0.7776,0.7373,0.9181,0.7571,0.7073,0.897,0.7862,0.7044,0.97,0.6421,0.6225,0.6757,0.5014,0.4893,0.3888,0.7563,0.6869,0.9116,0.5499,0.5538,0.4889
26
- Ziya2-13B-Chat,10B~20B,0.5403,0.5272,0.5731,0.6597,0.6313,0.8034,0.3259,0.2145,0.1373,0.673,0.6631,0.8101,0.6526,0.6282,0.7886,0.5583,0.5437,0.6097,0.3987,0.3541,0.2823,0.529,0.5194,0.5497,0.5377,0.5208,0.5678,0.4567,0.4484,0.4035
27
- InternLM2-Chat-20B,10B~20B,0.6819,0.7156,0.5781,0.7661,0.7819,0.7518,0.5506,0.5823,0.3134,0.8061,0.8182,0.8271,0.807,0.7993,0.832,0.8128,0.7876,0.8453,0.7037,0.7305,0.6224,0.6092,0.6548,0.4308,0.7815,0.7702,0.7821,0.5613,0.6058,0.3396
28
- Opt-13B,10B~20B,0.4746,0.4724,0.637,0.5147,0.519,0.7014,0.5146,0.5059,0.7153,0.5333,0.5557,0.7126,0.5261,0.5278,0.7228,0.5187,0.506,0.7257,0.5232,0.5081,0.7367,0.5218,0.5094,0.7314,0.4956,0.4856,0.6828,0.4722,0.4773,0.6264
29
- Mistral-Nemo-Instruct-2407,10B~20B,0.6375,0.6363,0.6018,0.6971,0.6973,0.7214,0.4741,0.4456,0.2722,0.6349,0.6873,0.6041,0.7122,0.7067,0.7508,0.7259,0.696,0.7825,0.5252,0.5197,0.3718,0.4695,0.4343,0.2607,0.6126,0.6117,0.5492,0.4474,0.4009,0.2212
30
- Phi-3-medium-4k-instruct,10B~20B,0.5533,0.5494,0.4889,0.5385,0.5594,0.4653,0.6034,0.6005,0.5922,0.5418,0.5993,0.4803,0.5866,0.6054,0.559,0.5815,0.578,0.5475,0.6178,0.607,0.6217,0.6437,0.6287,0.6742,0.6028,0.5912,0.5893,0.5057,0.5054,0.395
31
- Phi-3-medium-128k-instruct,10B~20B,0.6379,0.6234,0.6581,0.6379,0.6437,0.6554,0.6504,0.6361,0.6823,0.5919,0.6413,0.5687,0.6431,0.6483,0.6654,0.6568,0.6374,0.6958,0.6632,0.6403,0.7087,0.6819,0.6546,0.7465,0.6796,0.648,0.7433,0.5897,0.5935,0.5592
32
- phi-4,10B~20B,0.7431,0.7737,0.67,0.7139,0.7762,0.6194,0.7081,0.7576,0.6003,0.6957,0.7921,0.5974,0.7625,0.801,0.7146,0.8283,0.8125,0.844,0.713,0.7564,0.6083,0.6627,0.7239,0.5074,0.8171,0.8052,0.8213,0.6456,0.7165,0.4768
33
- Moonlight-16B-A3B-Instruct,10B~20B,0.4432,0.4087,0.3134,0.6335,0.6278,0.6971,0.3356,0.1806,0.0982,0.4713,0.5191,0.3914,0.5555,0.5699,0.5449,0.5349,0.5261,0.5011,0.4096,0.3505,0.2448,0.4197,0.3738,0.2672,0.4127,0.3514,0.2496,0.3428,0.2125,0.1175
34
- Qwen3-14B,10B~20B,0.6375,0.6363,0.6018,0.6971,0.6973,0.7214,0.4741,0.4456,0.2722,0.6349,0.5533,0.5494,0.4889,0.5385,0.5594,0.4653,0.6034,0.6005,0.5922,0.5418,0.6379,0.6234,0.6581,0.6379,0.6437,0.6554,0.6504,0.6361,0.6823,0.5919
35
- Gemma-3-12b-it,10B~20B,0.5533,0.5494,0.4889,0.5385,0.5594,0.4653,0.6034,0.6005,0.5922,0.5418,0.6379,0.6234,0.6581,0.6379,0.6437,0.6554,0.6504,0.6361,0.6823,0.5919,0.7431,0.7737,0.67,0.7139,0.7762,0.6194,0.7081,0.7576,0.6003,0.6957
36
- DeepSeek-LLM-67B-Chat,>65B,0.7897,0.7454,0.8652,0.8482,0.7832,0.9726,0.6603,0.6751,0.6011,0.8344,0.7978,0.932,0.8367,0.78,0.9497,0.8449,0.769,0.9767,0.7985,0.7493,0.8825,0.6171,0.6366,0.5125,0.8258,0.7583,0.9401,0.7387,0.7276,0.7596
37
- Qwen1.5-72B-Chat,>65B,0.5998,0.693,0.3298,0.8005,0.8477,0.7444,0.4697,0.3314,0.0703,0.6671,0.812,0.506,0.7676,0.8369,0.6803,0.7069,0.7895,0.5476,0.5825,0.6666,0.2918,0.4697,0.3186,0.0668,0.7076,0.7867,0.546,0.5283,0.5803,0.1942
38
- Qwen2.5-72B-Instruct,>65B,0.6248,0.6318,0.558,0.8125,0.7581,0.9309,0.3779,0.1555,0.0588,0.7372,0.7491,0.778,0.7655,0.7393,0.8392,0.8127,0.7442,0.9413,0.5162,0.5073,0.3355,0.4269,0.3262,0.1557,0.725,0.6977,0.7639,0.4205,0.3252,0.1506
39
- Qwen2-72B-Instruct,>65B,0.4969,0.467,0.2029,0.621,0.6897,0.4713,0.3983,0.0356,0.0085,0.5508,0.6602,0.3609,0.6984,0.7472,0.6237,0.6711,0.7073,0.5588,0.5013,0.4768,0.2114,0.4109,0.1184,0.0309,0.6349,0.6718,0.4834,0.4284,0.2565,0.0767
40
- Opt-66B,>65B,0.4866,0.482,0.682,0.5174,0.5203,0.7258,0.5579,0.5338,0.8237,0.5646,0.5728,0.7868,0.5385,0.535,0.7659,0.5571,0.5309,0.8257,0.5414,0.5199,0.7954,0.5354,0.5181,0.7801,0.5376,0.515,0.7909,0.5079,0.5041,0.7185
41
- Llama3-ChatQA-1.5-70B,>65B,0.6682,0.6617,0.6566,0.6859,0.6932,0.6922,0.6079,0.6187,0.5348,0.6548,0.7024,0.6342,0.6861,0.6945,0.6928,0.7029,0.6853,0.7281,0.6211,0.6242,0.5599,0.6105,0.6189,0.5397,0.7134,0.6873,0.7493,0.59,0.6072,0.4996
42
- Llama-3.1-70B-Instruct,>65B,0.4845,0.3825,0.0896,0.5771,0.6976,0.3045,0.4546,0.2021,0.0359,0.6067,0.7722,0.3926,0.5946,0.7225,0.3403,0.5904,0.6813,0.3067,0.4817,0.3639,0.0828,0.476,0.3471,0.0759,0.534,0.5584,0.1851,0.4837,0.4207,0.1019
43
- Llama-3.3-70B-Instruct,>65B,0.5045,0.4639,0.0849,0.5211,0.6327,0.1537,0.4943,0.4221,0.0718,0.5173,0.7089,0.1918,0.5728,0.7424,0.2569,0.5775,0.7071,0.2347,0.4964,0.406,0.0668,0.496,0.4244,0.0712,0.5183,0.5179,0.1065,0.482,0.3636,0.0544
44
- DeepSeek-R1-Distill-Llama-70B,>65B,0.5416,0.5902,0.2095,0.5495,0.6557,0.2531,0.477,0.3724,0.0843,0.6293,0.7886,0.4361,0.5619,0.6773,0.2789,0.556,0.6236,0.2398,0.4694,0.2909,0.0598,0.4773,0.3611,0.0813,0.5191,0.5141,0.1569,0.4642,0.3155,0.065
45
- Yi-1.5-34B-Chat,~30B,0.66,0.6114,0.8339,0.7311,0.6644,0.9577,0.3309,0.2379,0.1626,0.6958,0.6708,0.8646,0.7046,0.6528,0.9053,0.7084,0.6383,0.9309,0.5928,0.5672,0.6961,0.4467,0.4308,0.3972,0.6956,0.6281,0.9097,0.5182,0.515,0.5425
46
- Qwen2.5-32B-Instruct,~30B,0.6204,0.8741,0.2629,0.9049,0.9606,0.8489,0.5103,0.547,0.0453,0.8192,0.9583,0.6983,0.8514,0.956,0.7445,0.7823,0.9396,0.5931,0.5869,0.8351,0.1922,0.5244,0.6511,0.0699,0.8334,0.9475,0.695,0.5157,0.6401,0.0644
47
- Opt-30B,~30B,0.4672,0.4683,0.6648,0.5002,0.5082,0.7109,0.5044,0.4987,0.7354,0.5314,0.5517,0.7422,0.5108,0.5163,0.7304,0.5161,0.5039,0.7618,0.513,0.5009,0.7578,0.4956,0.4908,0.719,0.5119,0.4977,0.7583,0.4958,0.4955,0.7134
48
- 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
49
- 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
50
- 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
51
- 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
 
1
  Model,Size,Discrimination_Accuracy,Discrimination_Precision,Discrimination_Recall,Variant_Accuracy,Variant_Precision,Variant_Recall,Psychology_Accuracy,Psychology_Precision,Psychology_Recall,Politics_Accuracy,Politics_Precision,Politics_Recall,Eroticism_Accuracy,Eroticism_Precision,Eroticism_Recall,Vulgarity_Accuracy,Vulgarity_Precision,Vulgarity_Recall,Property_Accuracy,Property_Precision,Property_Recall,Injury_Accuracy,Injury_Precision,Injury_Recall,Criminality_Accuracy,Criminality_Precision,Criminality_Recall,Ethics_Accuracy,Ethics_Precision,Ethics_Recall
2
+ DeepSeek-LLM-67B-Chat,>65B,0.7897,0.7454,0.8652,0.8482,0.7832,0.9726,0.6603,0.6751,0.6011,0.8344,0.7978,0.932,0.8367,0.78,0.9497,0.8449,0.769,0.9767,0.7985,0.7493,0.8825,0.6171,0.6366,0.5125,0.8258,0.7583,0.9401,0.7387,0.7276,0.7596
3
+ Qwen1.5-72B-Chat,>65B,0.5998,0.693,0.3298,0.8005,0.8477,0.7444,0.4697,0.3314,0.0703,0.6671,0.812,0.506,0.7676,0.8369,0.6803,0.7069,0.7895,0.5476,0.5825,0.6666,0.2918,0.4697,0.3186,0.0668,0.7076,0.7867,0.546,0.5283,0.5803,0.1942
4
+ Qwen2.5-72B-Instruct,>65B,0.6248,0.6318,0.5580,0.8125,0.7581,0.9309,0.3779,0.1555,0.0588,0.7372,0.7491,0.7780,0.7655,0.7393,0.8392,0.8127,0.7442,0.9413,0.5162,0.5073,0.3355,0.4269,0.3262,0.1557,0.7250,0.6977,0.7639,0.4205,0.3252,0.1506
5
+ Qwen2-72B-Instruct,>65B,0.4969,0.4670,0.2029,0.6210,0.6897,0.4713,0.3983,0.0356,0.0085,0.5508,0.6602,0.3609,0.6984,0.7472,0.6237,0.6711,0.7073,0.5588,0.5013,0.4768,0.2114,0.4109,0.1184,0.0309,0.6349,0.6718,0.4834,0.4284,0.2565,0.0767
6
+ Opt-66B,>65B,0.4866,0.482,0.682,0.5174,0.5203,0.7258,0.5579,0.5338,0.8237,0.5646,0.5728,0.7868,0.5385,0.535,0.7659,0.5571,0.5309,0.8257,0.5414,0.5199,0.7954,0.5354,0.5181,0.7801,0.5376,0.515,0.7909,0.5079,0.5041,0.7185
7
+ Llama3-ChatQA-1.5-70B,>65B,0.6682,0.6617,0.6566,0.6859,0.6932,0.6922,0.6079,0.6187,0.5348,0.6548,0.7024,0.6342,0.6861,0.6945,0.6928,0.7029,0.6853,0.7281,0.6211,0.6242,0.5599,0.6105,0.6189,0.5397,0.7134,0.6873,0.7493,0.59,0.6072,0.4996
8
+ Llama-3.1-70B-Instruct,>65B,0.4845,0.3825,0.0896,0.5771,0.6976,0.3045,0.4546,0.2021,0.0359,0.6067,0.7722,0.3926,0.5946,0.7225,0.3403,0.5904,0.6813,0.3067,0.4817,0.3639,0.0828,0.4760,0.3471,0.0759,0.5340,0.5584,0.1851,0.4837,0.4207,0.1019
9
+ Llama-3.3-70B-Instruct,>65B,0.5045,0.4639,0.0849,0.5211,0.6327,0.1537,0.4943,0.4221,0.0718,0.5173,0.7089,0.1918,0.5728,0.7424,0.2569,0.5775,0.7071,0.2347,0.4964,0.4060,0.0668,0.4960,0.4244,0.0712,0.5183,0.5179,0.1065,0.4820,0.3636,0.0544
10
+ DeepSeek-R1-Distill-Llama-70B,>65B,0.5416,0.5902,0.2095,0.5495,0.6557,0.2531,0.4770,0.3724,0.0843,0.6293,0.7886,0.4361,0.5619,0.6773,0.2789,0.5560,0.6236,0.2398,0.4694,0.2909,0.0598,0.4773,0.3611,0.0813,0.5191,0.5141,0.1569,0.4642,0.3155,0.0650
11
+ Yi-1.5-34B-Chat,~30B,0.66,0.6114,0.8339,0.7311,0.6644,0.9577,0.3309,0.2379,0.1626,0.6958,0.6708,0.8646,0.7046,0.6528,0.9053,0.7084,0.6383,0.9309,0.5928,0.5672,0.6961,0.4467,0.4308,0.3972,0.6956,0.6281,0.9097,0.5182,0.515,0.5425
12
+ Qwen2.5-32B-Instruct,~30B,0.6204,0.8741,0.2629,0.9049,0.9606,0.8489,0.5103,0.5470,0.0453,0.8192,0.9583,0.6983,0.8514,0.9560,0.7445,0.7823,0.9396,0.5931,0.5869,0.8351,0.1922,0.5244,0.6511,0.0699,0.8334,0.9475,0.6950,0.5157,0.6401,0.0644
13
+ Opt-30B,~30B,0.4672,0.4683,0.6648,0.5002,0.5082,0.7109,0.5044,0.4987,0.7354,0.5314,0.5517,0.7422,0.5108,0.5163,0.7304,0.5161,0.5039,0.7618,0.513,0.5009,0.7578,0.4956,0.4908,0.719,0.5119,0.4977,0.7583,0.4958,0.4955,0.7134
14
+ QwQ-32B-Preview,~30B,0.6837,0.7403,0.5470,0.8120,0.8219,0.8084,0.6060,0.6749,0.3914,0.7516,0.8198,0.6977,0.8121,0.8230,0.8081,0.8470,0.8208,0.8801,0.6113,0.6736,0.3973,0.6050,0.6700,0.3873,0.7492,0.7768,0.6783,0.4656,0.3791,0.1124
15
+ Mistral-Small-24B-Instruct-2501,10B~20B,0.6626,0.6491,0.6746,0.7897,0.7347,0.9223,0.3990,0.2824,0.1406,0.7649,0.7465,0.8603,0.7828,0.7326,0.9081,0.8088,0.7280,0.9732,0.6010,0.6001,0.5490,0.4367,0.3723,0.2159,0.7369,0.6906,0.8282,0.4868,0.4773,0.3217
16
+ Baichuan2-13B-Chat,10B~20B,0.6337,0.6402,0.5755,0.7188,0.7164,0.7457,0.5185,0.5189,0.3417,0.7341,0.7487,0.7703,0.7033,0.7091,0.7143,0.6742,0.6712,0.6575,0.5657,0.5728,0.434,0.6151,0.6264,0.5371,0.6515,0.65,0.6089,0.5532,0.5707,0.414
17
+ Qwen1.5-14B-Chat,10B~20B,0.7099,0.6657,0.8141,0.7897,0.7205,0.9615,0.5669,0.5657,0.5226,0.7776,0.7373,0.9181,0.7571,0.7073,0.897,0.7862,0.7044,0.97,0.6421,0.6225,0.6757,0.5014,0.4893,0.3888,0.7563,0.6869,0.9116,0.5499,0.5538,0.4889
18
+ Ziya2-13B-Chat,10B~20B,0.5403,0.5272,0.5731,0.6597,0.6313,0.8034,0.3259,0.2145,0.1373,0.673,0.6631,0.8101,0.6526,0.6282,0.7886,0.5583,0.5437,0.6097,0.3987,0.3541,0.2823,0.529,0.5194,0.5497,0.5377,0.5208,0.5678,0.4567,0.4484,0.4035
19
+ InternLM2-Chat-20B,10B~20B,0.6819,0.7156,0.5781,0.7661,0.7819,0.7518,0.5506,0.5823,0.3134,0.8061,0.8182,0.8271,0.807,0.7993,0.832,0.8128,0.7876,0.8453,0.7037,0.7305,0.6224,0.6092,0.6548,0.4308,0.7815,0.7702,0.7821,0.5613,0.6058,0.3396
20
+ Opt-13B,10B~20B,0.4746,0.4724,0.637,0.5147,0.519,0.7014,0.5146,0.5059,0.7153,0.5333,0.5557,0.7126,0.5261,0.5278,0.7228,0.5187,0.506,0.7257,0.5232,0.5081,0.7367,0.5218,0.5094,0.7314,0.4956,0.4856,0.6828,0.4722,0.4773,0.6264
21
+ Mistral-Nemo-Instruct-2407,10B~20B,0.6375,0.6363,0.6018,0.6971,0.6973,0.7214,0.4741,0.4456,0.2722,0.6349,0.6873,0.6041,0.7122,0.7067,0.7508,0.7259,0.6960,0.7825,0.5252,0.5197,0.3718,0.4695,0.4343,0.2607,0.6126,0.6117,0.5492,0.4474,0.4009,0.2212
22
+ Phi-3-medium-4k-instruct,10B~20B,0.5533,0.5494,0.4889,0.5385,0.5594,0.4653,0.6034,0.6005,0.5922,0.5418,0.5993,0.4803,0.5866,0.6054,0.5590,0.5815,0.5780,0.5475,0.6178,0.6070,0.6217,0.6437,0.6287,0.6742,0.6028,0.5912,0.5893,0.5057,0.5054,0.3950
23
+ Phi-3-medium-128k-instruct,10B~20B,0.6379,0.6234,0.6581,0.6379,0.6437,0.6554,0.6504,0.6361,0.6823,0.5919,0.6413,0.5687,0.6431,0.6483,0.6654,0.6568,0.6374,0.6958,0.6632,0.6403,0.7087,0.6819,0.6546,0.7465,0.6796,0.6480,0.7433,0.5897,0.5935,0.5592
24
+ phi-4,10B~20B,0.7431,0.7737,0.6700,0.7139,0.7762,0.6194,0.7081,0.7576,0.6003,0.6957,0.7921,0.5974,0.7625,0.8010,0.7146,0.8283,0.8125,0.8440,0.7130,0.7564,0.6083,0.6627,0.7239,0.5074,0.8171,0.8052,0.8213,0.6456,0.7165,0.4768
25
+ Moonlight-16B-A3B-Instruct,10B~20B,0.4432,0.4087,0.3134,0.6335,0.6278,0.6971,0.3356,0.1806,0.0982,0.4713,0.5191,0.3914,0.5555,0.5699,0.5449,0.5349,0.5261,0.5011,0.4096,0.3505,0.2448,0.4197,0.3738,0.2672,0.4127,0.3514,0.2496,0.3428,0.2125,0.1175
26
  Gemma-1.1-7B-it,5B~10B,0.7849,0.7205,0.9139,0.8081,0.7454,0.9485,0.6024,0.6084,0.5413,0.7854,0.758,0.8894,0.8017,0.7436,0.9353,0.8215,0.7367,0.9884,0.6669,0.6543,0.673,0.5811,0.5858,0.4976,0.7831,0.7167,0.9127,0.6684,0.6638,0.6754
27
  Qwen1.5-7B-Chat,5B~10B,0.6885,0.6347,0.8535,0.7677,0.6891,0.9938,0.6929,0.6404,0.8588,0.7791,0.7151,0.9869,0.7653,0.6889,0.988,0.7485,0.6659,0.9746,0.684,0.6317,0.8443,0.7267,0.6564,0.929,0.7473,0.662,0.9772,0.5545,0.5496,0.5778
28
  Yi-1.5-9B-Chat,5B~10B,0.7025,0.6913,0.7058,0.7032,0.7106,0.707,0.4533,0.3925,0.2,0.6546,0.7097,0.6172,0.7209,0.7213,0.7419,0.8197,0.7508,0.9452,0.5595,0.5666,0.4131,0.4342,0.3378,0.1591,0.7626,0.7215,0.8306,0.4057,0.2654,0.1096
 
35
  Mistral-7B-Instruct-v0.3,5B~10B,0.7069,0.6749,0.7706,0.7521,0.7161,0.8533,0.5826,0.5868,0.5167,0.7142,0.7222,0.7711,0.7599,0.7205,0.8679,0.7956,0.7205,0.9509,0.6748,0.6547,0.7042,0.6139,0.6127,0.5802,0.7742,0.7074,0.9103,0.6388,0.6387,0.6313
36
  Llama3-ChatQA-1.5-8B,5B~10B,0.6114,0.5657,0.8761,0.6276,0.5904,0.885,0.5978,0.5613,0.844,0.6056,0.6016,0.8128,0.6113,0.5825,0.8521,0.6365,0.5805,0.9258,0.6062,0.5625,0.8663,0.6034,0.5629,0.8569,0.6223,0.5694,0.903,0.5658,0.5447,0.7752
37
  Ministral-8B-Instruct-2410,5B~10B,0.6447,0.6342,0.6442,0.7197,0.7001,0.7911,0.5176,0.5149,0.3869,0.6868,0.7082,0.7217,0.7326,0.7075,0.8161,0.7362,0.6919,0.8305,0.5742,0.5735,0.5003,0.4649,0.4306,0.2781,0.6894,0.6614,0.7369,0.5258,0.5313,0.4059
38
+ Phi-3-small-8k-instruct,5B~10B,0.7598,0.7484,0.7666,0.7738,0.7711,0.7936,0.7227,0.7317,0.6914,0.7477,0.7825,0.7432,0.7999,0.7827,0.8445,0.8204,0.7762,0.8907,0.7106,0.7202,0.6662,0.6027,0.6353,0.4468,0.7871,0.7581,0.8233,0.5930,0.6349,0.4325
39
+ Phi-3-small-128k-instruct,5B~10B,0.7158,0.7404,0.6454,0.6831,0.7398,0.5872,0.6751,0.7171,0.5633,0.6057,0.7155,0.4565,0.6783,0.7379,0.5785,0.8131,0.7893,0.8433,0.6832,0.7183,0.5779,0.6189,0.6671,0.4487,0.7353,0.7481,0.6836,0.5863,0.6398,0.3873
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
data/subclass_per.csv CHANGED
@@ -1,4 +1,28 @@
1
  Model,Size,Discrimination_Accuracy,Discrimination_Precision,Discrimination_Recall,Variant_Accuracy,Variant_Precision,Variant_Recall,Psychology_Accuracy,Psychology_Precision,Psychology_Recall,Politics_Accuracy,Politics_Precision,Politics_Recall,Eroticism_Accuracy,Eroticism_Precision,Eroticism_Recall,Vulgarity_Accuracy,Vulgarity_Precision,Vulgarity_Recall,Property_Accuracy,Property_Precision,Property_Recall,Injury_Accuracy,Injury_Precision,Injury_Recall,Criminality_Accuracy,Criminality_Precision,Criminality_Recall,Ethics_Accuracy,Ethics_Precision,Ethics_Recall
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2
  Gemma-1.1-7B-it,5B~10B,0.6885,0.6193,0.9389,0.7201,0.6502,0.9795,0.6709,0.6133,0.8985,0.7171,0.6709,0.9421,0.5993,0.5861,0.7426,0.7164,0.634,0.9953,0.6316,0.5872,0.8235,0.5207,0.5098,0.595,0.6874,0.616,0.9415,0.6164,0.5853,0.7856
3
  Qwen1.5-7B-Chat,5B~10B,0.6415,0.5933,0.8439,0.7295,0.6542,0.9987,0.5495,0.5352,0.6535,0.7415,0.6808,0.9875,0.7286,0.6545,0.9955,0.7167,0.6339,0.9966,0.6122,0.5749,0.784,0.4866,0.4788,0.5265,0.6887,0.6165,0.9449,0.4276,0.4219,0.4072
4
  Yi-1.5-9B-Chat,5B~10B,0.7089,0.8612,0.4825,0.5418,0.7129,0.1741,0.4846,0.2932,0.0308,0.5376,0.7743,0.2115,0.6185,0.8236,0.3254,0.818,0.9011,0.7057,0.5819,0.7416,0.2207,0.4893,0.3279,0.0365,0.7959,0.8937,0.6572,0.477,0.2414,0.0233
@@ -10,35 +34,6 @@ InternLM2-Chat-7B,5B~10B,0.4988,0,0,0.4767,0,0,0.4943,0,0,0.4453,0.0513,0.0011,0
10
  Opt-6.7B,5B~10B,0.5189,0.5038,0.9645,0.3756,0.4266,0.6456,0.5227,0.5083,0.9638,0.549,0.5504,0.9314,0.2606,0.3276,0.4205,0.4833,0.4847,0.8892,0.5274,0.508,0.9831,0.5244,0.508,0.971,0.5105,0.4973,0.9551,0.5322,0.5159,0.9757
11
  Mistral-7B-Instruct-v0.3,5B~10B,0.4091,0.3399,0.2241,0.3013,0.0672,0.0286,0.3093,0.0548,0.0246,0.3554,0.3176,0.1618,0.4671,0.473,0.3538,0.62,0.6022,0.655,0.432,0.3832,0.2701,0.3362,0.1517,0.0771,0.6338,0.6081,0.6844,0.3814,0.2943,0.1744
12
  Llama3-ChatQA-1.5-8B,5B~10B,0.387,0.2816,0.1665,0.3232,0.1355,0.0603,0.3054,0.011,0.0045,0.292,0.0948,0.0354,0.7946,0.7193,0.9821,0.5375,0.5306,0.4746,0.3702,0.2367,0.1312,0.318,0.0621,0.0276,0.4823,0.4562,0.3594,0.3398,0.1632,0.0793
13
- Ministral-8B-Instruct-2410,5B~10B,0.608,0.572,0.7836,0.7143,0.6458,0.9763,0.326,0.2659,0.2081,0.6623,0.6438,0.8483,0.7052,0.6425,0.957,0.7069,0.6278,0.9849,0.5197,0.5064,0.603,0.3152,0.242,0.1855,0.6558,0.5975,0.8847,0.4132,0.4054,0.386
14
- Phi-3-small-8k-instruct,5B~10B,0.6365,0.5771,0.9543,0.4834,0.4955,0.6276,0.6479,0.5862,0.9722,0.6323,0.6122,0.8839,0.6153,0.582,0.8846,0.6563,0.5881,0.9939,0.5791,0.5445,0.838,0.6012,0.5593,0.8793,0.6322,0.5723,0.9513,0.4856,0.4872,0.6404
15
- Phi-3-small-128k-instruct,5B~10B,0.6085,0.5851,0.681,0.3324,0.2336,0.1343,0.7347,0.6638,0.9355,0.6062,0.6315,0.6625,0.6078,0.6056,0.6736,0.7148,0.6513,0.8975,0.6468,0.6108,0.7597,0.7331,0.6615,0.9338,0.7076,0.6437,0.8871,0.4432,0.427,0.3467
16
- Qwen3-4B,5B~10B,0.4091 ,0.3399 ,0.2241 ,0.3013 ,0.0672 ,0.0286 ,0.3093 ,0.0548 ,0.0246 ,0.3554 ,0.3870 ,0.2816 ,0.1665 ,0.3232 ,0.1355 ,0.0603 ,0.3054 ,0.0110 ,0.0045 ,0.2920 ,0.6080 ,0.5720 ,0.7836 ,0.7143 ,0.6458 ,0.9763 ,0.3260 ,0.2659 ,0.2081 ,0.6623
17
- Qwen3-8B,5B~10B,0.3870 ,0.2816 ,0.1665 ,0.3232 ,0.1355 ,0.0603 ,0.3054 ,0.0110 ,0.0045 ,0.2920 ,0.6080 ,0.5720 ,0.7836 ,0.7143 ,0.6458 ,0.9763 ,0.3260 ,0.2659 ,0.2081 ,0.6623 ,0.6365 ,0.5771 ,0.9543 ,0.4834 ,0.4955 ,0.6276 ,0.6479 ,0.5862 ,0.9722 ,0.6323
18
- Baichuan2-13B-Chat,10B~20B,0.7346,0.6715,0.8932,0.7703,0.7043,0.9491,0.6303,0.6129,0.6785,0.7435,0.7152,0.8777,0.779,0.7088,0.9649,0.7677,0.6883,0.9601,0.6763,0.6388,0.7738,0.6359,0.6149,0.6904,0.7096,0.6554,0.8436,0.7306,0.6762,0.8788
19
- Qwen1.5-14B-Chat,10B~20B,0.625,0.5683,0.964,0.6549,0.5977,0.9932,0.5983,0.5571,0.9038,0.6561,0.6193,0.9535,0.6592,0.6005,0.9994,0.6382,0.5759,0.9897,0.5579,0.53,0.8275,0.5009,0.4938,0.7077,0.6256,0.566,0.9705,0.6063,0.5643,0.914
20
- Ziya2-13B-Chat,10B~20B,0.6322,0.6632,0.502,0.381,0.0822,0.0212,0.4263,0.2557,0.086,0.4352,0.4474,0.1651,0.612,0.6721,0.4744,0.812,0.7741,0.8691,0.4904,0.4516,0.2102,0.5309,0.5403,0.2964,0.7186,0.7235,0.6777,0.4811,0.4512,0.2021
21
- InternLM2-Chat-20B,10B~20B,0.5184,0.5912,0.0441,0.4754,0.0222,0.0006,0.4929,0.0222,0.0006,0.4744,0.7043,0.0573,0.605,0.904,0.256,0.5265,0.6774,0.0625,0.5689,0.8292,0.146,0.5046,0.4073,0.0202,0.7142,0.9352,0.44,0.498,0.4041,0.0196
22
- Opt-13B,10B~20B,0.5011,0.0392,0.0015,0.4792,0.0695,0.0018,0.4958,0,0,0.4492,0.237,0.0055,0.4897,0.5438,0.0249,0.4996,0.0333,0.0006,0.5037,0.1931,0.0055,0.5454,0.8065,0.0965,0.5155,0.499,0.0228,0.5016,0.4815,0.0203
23
- Mistral-Nemo-Instruct-2407,10B~20B,0.6992,0.6359,0.896,0.7518,0.6773,0.9826,0.6421,0.6067,0.7767,0.729,0.6896,0.9121,0.7377,0.6719,0.9542,0.7482,0.6611,0.9959,0.6396,0.6014,0.7754,0.6045,0.5803,0.7019,0.7246,0.6464,0.9529,0.491,0.4881,0.4717
24
- Phi-3-medium-4k-instruct,10B~20B,0.8162,0.7447,0.9484,0.395,0.2748,0.1126,0.8368,0.7558,0.9878,0.5763,0.6486,0.4809,0.6431,0.6695,0.5981,0.8403,0.7549,0.9973,0.8092,0.7414,0.9343,0.8263,0.7504,0.9679,0.8352,0.7499,0.9896,0.6361,0.6499,0.5818
25
- Phi-3-medium-128k-instruct,10B~20B,0.8024,0.7318,0.9391,0.3592,0.1596,0.0598,0.8232,0.7434,0.979,0.5228,0.591,0.3977,0.5699,0.6022,0.4725,0.8293,0.7436,0.9939,0.7813,0.7222,0.8963,0.8009,0.7328,0.9351,0.826,0.7393,0.9898,0.6525,0.6565,0.6327
26
- phi-4,10B~20B,0.6193,0.6166,0.5816,0.4118,0.3517,0.1792,0.7011,0.6785,0.7484,0.7224,0.7291,0.7791,0.6152,0.6372,0.5775,0.7375,0.696,0.8232,0.5775,0.5779,0.4961,0.6685,0.656,0.6821,0.7074,0.6752,0.7638,0.4629,0.4356,0.2692
27
- Moonlight-16B-A3B-Instruct,10B~20B,0.5041,0.0556,0.0006,0.4814,0,0,0.4992,0,0,0.45,0.1369,0.0016,0.4804,0.0256,0.0007,0.5027,0,0,0.5054,0.0893,0.002,0.502,0.0972,0.0014,0.508,0.0256,0.0007,0.4947,0,0
28
- Qwen3-14B,10B~20B,0.6992 ,0.6359 ,0.8960 ,0.7518 ,0.6773 ,0.9826 ,0.6421 ,0.6067 ,0.7767 ,0.7290 ,0.8162 ,0.7447 ,0.9484 ,0.3950 ,0.2748 ,0.1126 ,0.8368 ,0.7558 ,0.9878 ,0.5763 ,0.8024 ,0.7318 ,0.9391 ,0.3592 ,0.1596 ,0.0598 ,0.8232 ,0.7434 ,0.9790 ,0.5228
29
- DeepSeek-LLM-67B-Chat,>65B,0.6948,0.9451,0.3989,0.6447,0.9375,0.3259,0.5122,0.5824,0.033,0.7673,0.9695,0.5903,0.6865,0.9516,0.4092,0.899,0.9725,0.8159,0.66,0.9341,0.326,0.5479,0.8184,0.1017,0.8777,0.9706,0.7709,0.5142,0.6736,0.0456
30
- Qwen1.5-72B-Chat,>65B,0.6479,0.581,0.9985,0.6609,0.6019,0.9938,0.6472,0.5837,0.9906,0.5928,0.5895,0.8276,0.6544,0.5996,0.9796,0.6488,0.5823,0.9987,0.6448,0.5792,0.9932,0.6255,0.5712,0.9493,0.6433,0.5763,0.9951,0.6485,0.5872,0.9874
31
- Qwen2.5-72B-Instruct,>65B,0.6292,0.6414,0.548,0.8411,0.776,0.9689,0.3631,0.0282,0.0086,0.7521,0.7629,0.7894,0.7928,0.7585,0.8742,0.8142,0.7522,0.9248,0.5333,0.5328,0.3499,0.3959,0.1923,0.0723,0.749,0.718,0.7928,0.3967,0.2195,0.0826
32
- Qwen2-72B-Instruct,>65B,0.6587,0.5982,0.9159,0.7064,0.6373,0.987,0.4112,0.4039,0.409,0.6611,0.6383,0.8691,0.692,0.6315,0.9577,0.6948,0.6175,0.9884,0.6106,0.5703,0.8181,0.4184,0.4103,0.4236,0.6658,0.5992,0.9347,0.4887,0.4879,0.565
33
- Opt-66B,>65B,0.645,0.5831,0.9572,0.3981,0.417,0.4471,0.6667,0.5971,0.9953,0.6232,0.6095,0.8551,0.4854,0.4984,0.6176,0.652,0.5874,0.9698,0.6511,0.5859,0.9706,0.6604,0.5926,0.9853,0.6556,0.586,0.9846,0.655,0.5943,0.9665
34
- Llama3-ChatQA-1.5-70B,>65B,0.3666,0.2082,0.1069,0.339,0.169,0.0752,0.3147,0.0148,0.0059,0.2947,0.075,0.0261,0.7758,0.7167,0.9293,0.5528,0.5482,0.4877,0.3396,0.111,0.0507,0.3207,0.0374,0.0156,0.4392,0.3806,0.2524,0.3214,0.0614,0.0253
35
- Llama-3.1-70B-Instruct,>65B,0.467,0.4105,0.2107,0.3766,0.1681,0.056,0.3856,0.1439,0.0505,0.346,0.1387,0.0392,0.4036,0.2873,0.1107,0.3872,0.1394,0.0487,0.4967,0.4715,0.2711,0.407,0.2331,0.091,0.4985,0.4691,0.2716,0.6337,0.6553,0.5548
36
- Llama-3.3-70B-Instruct,>65B,0.3996,0.3526,0.2759,0.2923,0.143,0.0771,0.3029,0.142,0.0825,0.2624,0.1066,0.0486,0.3657,0.3253,0.2213,0.3305,0.2121,0.1358,0.4583,0.4388,0.3966,0.3156,0.175,0.1062,0.451,0.4249,0.3802,0.5813,0.5696,0.6459
37
- DeepSeek-R1-Distill-Llama-70B,>65B,0.424,0.2914,0.1265,0.6148,0.653,0.5255,0.3608,0.0107,0.0033,0.5182,0.5945,0.3588,0.5583,0.5989,0.4156,0.4922,0.4667,0.2664,0.4312,0.3134,0.1401,0.3727,0.0743,0.0243,0.4061,0.2132,0.0844,0.537,0.5522,0.3638
38
- Yi-1.5-34B-Chat,~30B,0.7139,0.8341,0.5176,0.7722,0.8735,0.6482,0.475,0.2581,0.0357,0.7162,0.8717,0.5603,0.6206,0.7912,0.353,0.8816,0.8938,0.8601,0.6412,0.7813,0.3672,0.497,0.4306,0.0769,0.8472,0.8832,0.7889,0.4818,0.3646,0.0576
39
- Qwen2.5-32B-Instruct,~30B,0.6749,0.6366,0.7789,0.7893,0.7099,0.9938,0.4372,0.4025,0.2943,0.7921,0.7323,0.9739,0.7723,0.7036,0.9599,0.7702,0.6873,0.9727,0.592,0.5774,0.6092,0.4358,0.3969,0.2906,0.7404,0.6695,0.916,0.464,0.4506,0.3514
40
- Opt-30B,~30B,0.5831,0.5754,0.5565,0.3952,0.338,0.1915,0.6784,0.6507,0.7506,0.5798,0.6281,0.5559,0.357,0.2405,0.1185,0.406,0.3224,0.1945,0.6203,0.6061,0.633,0.6188,0.6076,0.6293,0.6031,0.5886,0.5976,0.6244,0.6184,0.6415
41
- 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
42
- 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
43
- 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
44
- 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
 
1
  Model,Size,Discrimination_Accuracy,Discrimination_Precision,Discrimination_Recall,Variant_Accuracy,Variant_Precision,Variant_Recall,Psychology_Accuracy,Psychology_Precision,Psychology_Recall,Politics_Accuracy,Politics_Precision,Politics_Recall,Eroticism_Accuracy,Eroticism_Precision,Eroticism_Recall,Vulgarity_Accuracy,Vulgarity_Precision,Vulgarity_Recall,Property_Accuracy,Property_Precision,Property_Recall,Injury_Accuracy,Injury_Precision,Injury_Recall,Criminality_Accuracy,Criminality_Precision,Criminality_Recall,Ethics_Accuracy,Ethics_Precision,Ethics_Recall
2
+ DeepSeek-LLM-67B-Chat,>65B,0.6948,0.9451,0.3989,0.6447,0.9375,0.3259,0.5122,0.5824,0.033,0.7673,0.9695,0.5903,0.6865,0.9516,0.4092,0.899,0.9725,0.8159,0.66,0.9341,0.326,0.5479,0.8184,0.1017,0.8777,0.9706,0.7709,0.5142,0.6736,0.0456
3
+ Qwen1.5-72B-Chat,>65B,0.6479,0.581,0.9985,0.6609,0.6019,0.9938,0.6472,0.5837,0.9906,0.5928,0.5895,0.8276,0.6544,0.5996,0.9796,0.6488,0.5823,0.9987,0.6448,0.5792,0.9932,0.6255,0.5712,0.9493,0.6433,0.5763,0.9951,0.6485,0.5872,0.9874
4
+ Qwen2.5-72B-Instruct,>65B,0.6292,0.6414,0.5480,0.8411,0.7760,0.9689,0.3631,0.0282,0.0086,0.7521,0.7629,0.7894,0.7928,0.7585,0.8742,0.8142,0.7522,0.9248,0.5333,0.5328,0.3499,0.3959,0.1923,0.0723,0.7490,0.7180,0.7928,0.3967,0.2195,0.0826
5
+ Qwen2-72B-Instruct,>65B,0.6587,0.5982,0.9159,0.7064,0.6373,0.9870,0.4112,0.4039,0.4090,0.6611,0.6383,0.8691,0.6920,0.6315,0.9577,0.6948,0.6175,0.9884,0.6106,0.5703,0.8181,0.4184,0.4103,0.4236,0.6658,0.5992,0.9347,0.4887,0.4879,0.5650
6
+ Opt-66B,>65B,0.645,0.5831,0.9572,0.3981,0.417,0.4471,0.6667,0.5971,0.9953,0.6232,0.6095,0.8551,0.4854,0.4984,0.6176,0.652,0.5874,0.9698,0.6511,0.5859,0.9706,0.6604,0.5926,0.9853,0.6556,0.586,0.9846,0.655,0.5943,0.9665
7
+ Llama3-ChatQA-1.5-70B,>65B,0.3666,0.2082,0.1069,0.339,0.169,0.0752,0.3147,0.0148,0.0059,0.2947,0.075,0.0261,0.7758,0.7167,0.9293,0.5528,0.5482,0.4877,0.3396,0.111,0.0507,0.3207,0.0374,0.0156,0.4392,0.3806,0.2524,0.3214,0.0614,0.0253
8
+ Llama-3.1-70B-Instruct,>65B,0.4670,0.4105,0.2107,0.3766,0.1681,0.0560,0.3856,0.1439,0.0505,0.3460,0.1387,0.0392,0.4036,0.2873,0.1107,0.3872,0.1394,0.0487,0.4967,0.4715,0.2711,0.4070,0.2331,0.0910,0.4985,0.4691,0.2716,0.6337,0.6553,0.5548
9
+ Llama-3.3-70B-Instruct,>65B,0.3996,0.3526,0.2759,0.2923,0.1430,0.0771,0.3029,0.1420,0.0825,0.2624,0.1066,0.0486,0.3657,0.3253,0.2213,0.3305,0.2121,0.1358,0.4583,0.4388,0.3966,0.3156,0.1750,0.1062,0.4510,0.4249,0.3802,0.5813,0.5696,0.6459
10
+ DeepSeek-R1-Distill-Llama-70B,>65B,0.4240,0.2914,0.1265,0.6148,0.6530,0.5255,0.3608,0.0107,0.0033,0.5182,0.5945,0.3588,0.5583,0.5989,0.4156,0.4922,0.4667,0.2664,0.4312,0.3134,0.1401,0.3727,0.0743,0.0243,0.4061,0.2132,0.0844,0.5370,0.5522,0.3638
11
+ Yi-1.5-34B-Chat,~30B,0.7139,0.8341,0.5176,0.7722,0.8735,0.6482,0.475,0.2581,0.0357,0.7162,0.8717,0.5603,0.6206,0.7912,0.353,0.8816,0.8938,0.8601,0.6412,0.7813,0.3672,0.497,0.4306,0.0769,0.8472,0.8832,0.7889,0.4818,0.3646,0.0576
12
+ Qwen2.5-32B-Instruct,~30B,0.6749,0.6366,0.7789,0.7893,0.7099,0.9938,0.4372,0.4025,0.2943,0.7921,0.7323,0.9739,0.7723,0.7036,0.9599,0.7702,0.6873,0.9727,0.5920,0.5774,0.6092,0.4358,0.3969,0.2906,0.7404,0.6695,0.9160,0.4640,0.4506,0.3514
13
+ Opt-30B,~30B,0.5831,0.5754,0.5565,0.3952,0.338,0.1915,0.6784,0.6507,0.7506,0.5798,0.6281,0.5559,0.357,0.2405,0.1185,0.406,0.3224,0.1945,0.6203,0.6061,0.633,0.6188,0.6076,0.6293,0.6031,0.5886,0.5976,0.6244,0.6184,0.6415
14
+ QwQ-32B-Preview,~30B,0.5231,0.5061,0.9839,0.5519,0.5328,1.0000,0.4141,0.4443,0.7537,0.5814,0.5650,0.9989,0.5529,0.5340,0.9993,0.5318,0.5111,0.9993,0.5083,0.4978,0.9542,0.4392,0.4593,0.8080,0.5238,0.5042,0.9922,0.5269,0.5128,0.9743
15
+ Mistral-Small-24B-Instruct-2501,~30B,0.5897,0.5714,0.6393,0.7706,0.6931,0.9888,0.3109,0.1339,0.0727,0.7308,0.6984,0.8887,0.7454,0.6830,0.9385,0.7584,0.6732,0.9835,0.5850,0.5671,0.6297,0.3646,0.2744,0.1803,0.7088,0.6450,0.8855,0.3839,0.3257,0.2233
16
+ Baichuan2-13B-Chat,10B~20B,0.7346,0.6715,0.8932,0.7703,0.7043,0.9491,0.6303,0.6129,0.6785,0.7435,0.7152,0.8777,0.779,0.7088,0.9649,0.7677,0.6883,0.9601,0.6763,0.6388,0.7738,0.6359,0.6149,0.6904,0.7096,0.6554,0.8436,0.7306,0.6762,0.8788
17
+ Qwen1.5-14B-Chat,10B~20B,0.625,0.5683,0.964,0.6549,0.5977,0.9932,0.5983,0.5571,0.9038,0.6561,0.6193,0.9535,0.6592,0.6005,0.9994,0.6382,0.5759,0.9897,0.5579,0.53,0.8275,0.5009,0.4938,0.7077,0.6256,0.566,0.9705,0.6063,0.5643,0.914
18
+ Ziya2-13B-Chat,10B~20B,0.6322,0.6632,0.502,0.381,0.0822,0.0212,0.4263,0.2557,0.086,0.4352,0.4474,0.1651,0.612,0.6721,0.4744,0.812,0.7741,0.8691,0.4904,0.4516,0.2102,0.5309,0.5403,0.2964,0.7186,0.7235,0.6777,0.4811,0.4512,0.2021
19
+ InternLM2-Chat-20B,10B~20B,0.5184,0.5912,0.0441,0.4754,0.0222,0.0006,0.4929,0.0222,0.0006,0.4744,0.7043,0.0573,0.605,0.904,0.256,0.5265,0.6774,0.0625,0.5689,0.8292,0.146,0.5046,0.4073,0.0202,0.7142,0.9352,0.44,0.498,0.4041,0.0196
20
+ Opt-13B,10B~20B,0.5011,0.0392,0.0015,0.4792,0.0695,0.0018,0.4958,0,0,0.4492,0.237,0.0055,0.4897,0.5438,0.0249,0.4996,0.0333,0.0006,0.5037,0.1931,0.0055,0.5454,0.8065,0.0965,0.5155,0.499,0.0228,0.5016,0.4815,0.0203
21
+ Mistral-Nemo-Instruct-2407,10B~20B,0.6992,0.6359,0.8960,0.7518,0.6773,0.9826,0.6421,0.6067,0.7767,0.7290,0.6896,0.9121,0.7377,0.6719,0.9542,0.7482,0.6611,0.9959,0.6396,0.6014,0.7754,0.6045,0.5803,0.7019,0.7246,0.6464,0.9529,0.4910,0.4881,0.4717
22
+ Phi-3-medium-4k-instruct,10B~20B,0.8162,0.7447,0.9484,0.3950,0.2748,0.1126,0.8368,0.7558,0.9878,0.5763,0.6486,0.4809,0.6431,0.6695,0.5981,0.8403,0.7549,0.9973,0.8092,0.7414,0.9343,0.8263,0.7504,0.9679,0.8352,0.7499,0.9896,0.6361,0.6499,0.5818
23
+ Phi-3-medium-128k-instruct,10B~20B,0.8024,0.7318,0.9391,0.3592,0.1596,0.0598,0.8232,0.7434,0.9790,0.5228,0.5910,0.3977,0.5699,0.6022,0.4725,0.8293,0.7436,0.9939,0.7813,0.7222,0.8963,0.8009,0.7328,0.9351,0.8260,0.7393,0.9898,0.6525,0.6565,0.6327
24
+ phi-4,10B~20B,0.6193,0.6166,0.5816,0.4118,0.3517,0.1792,0.7011,0.6785,0.7484,0.7224,0.7291,0.7791,0.6152,0.6372,0.5775,0.7375,0.6960,0.8232,0.5775,0.5779,0.4961,0.6685,0.6560,0.6821,0.7074,0.6752,0.7638,0.4629,0.4356,0.2692
25
+ Moonlight-16B-A3B-Instruct,10B~20B,0.5041,0.0556,0.0006,0.4814,0.0000,0.0000,0.4992,0.0000,0.0000,0.4500,0.1369,0.0016,0.4804,0.0256,0.0007,0.5027,0.0000,0.0000,0.5054,0.0893,0.0020,0.5020,0.0972,0.0014,0.5080,0.0256,0.0007,0.4947,0.0000,0.0000
26
  Gemma-1.1-7B-it,5B~10B,0.6885,0.6193,0.9389,0.7201,0.6502,0.9795,0.6709,0.6133,0.8985,0.7171,0.6709,0.9421,0.5993,0.5861,0.7426,0.7164,0.634,0.9953,0.6316,0.5872,0.8235,0.5207,0.5098,0.595,0.6874,0.616,0.9415,0.6164,0.5853,0.7856
27
  Qwen1.5-7B-Chat,5B~10B,0.6415,0.5933,0.8439,0.7295,0.6542,0.9987,0.5495,0.5352,0.6535,0.7415,0.6808,0.9875,0.7286,0.6545,0.9955,0.7167,0.6339,0.9966,0.6122,0.5749,0.784,0.4866,0.4788,0.5265,0.6887,0.6165,0.9449,0.4276,0.4219,0.4072
28
  Yi-1.5-9B-Chat,5B~10B,0.7089,0.8612,0.4825,0.5418,0.7129,0.1741,0.4846,0.2932,0.0308,0.5376,0.7743,0.2115,0.6185,0.8236,0.3254,0.818,0.9011,0.7057,0.5819,0.7416,0.2207,0.4893,0.3279,0.0365,0.7959,0.8937,0.6572,0.477,0.2414,0.0233
 
34
  Opt-6.7B,5B~10B,0.5189,0.5038,0.9645,0.3756,0.4266,0.6456,0.5227,0.5083,0.9638,0.549,0.5504,0.9314,0.2606,0.3276,0.4205,0.4833,0.4847,0.8892,0.5274,0.508,0.9831,0.5244,0.508,0.971,0.5105,0.4973,0.9551,0.5322,0.5159,0.9757
35
  Mistral-7B-Instruct-v0.3,5B~10B,0.4091,0.3399,0.2241,0.3013,0.0672,0.0286,0.3093,0.0548,0.0246,0.3554,0.3176,0.1618,0.4671,0.473,0.3538,0.62,0.6022,0.655,0.432,0.3832,0.2701,0.3362,0.1517,0.0771,0.6338,0.6081,0.6844,0.3814,0.2943,0.1744
36
  Llama3-ChatQA-1.5-8B,5B~10B,0.387,0.2816,0.1665,0.3232,0.1355,0.0603,0.3054,0.011,0.0045,0.292,0.0948,0.0354,0.7946,0.7193,0.9821,0.5375,0.5306,0.4746,0.3702,0.2367,0.1312,0.318,0.0621,0.0276,0.4823,0.4562,0.3594,0.3398,0.1632,0.0793
37
+ Ministral-8B-Instruct-2410,5B~10B,0.6080,0.5720,0.7836,0.7143,0.6458,0.9763,0.3260,0.2659,0.2081,0.6623,0.6438,0.8483,0.7052,0.6425,0.9570,0.7069,0.6278,0.9849,0.5197,0.5064,0.6030,0.3152,0.2420,0.1855,0.6558,0.5975,0.8847,0.4132,0.4054,0.3860
38
+ Phi-3-small-8k-instruct,5B~10B,0.6365,0.5771,0.9543,0.4834,0.4955,0.6276,0.6479,0.5862,0.9722,0.6323,0.6122,0.8839,0.6153,0.5820,0.8846,0.6563,0.5881,0.9939,0.5791,0.5445,0.8380,0.6012,0.5593,0.8793,0.6322,0.5723,0.9513,0.4856,0.4872,0.6404
39
+ Phi-3-small-128k-instruct,5B~10B,0.6085,0.5851,0.6810,0.3324,0.2336,0.1343,0.7347,0.6638,0.9355,0.6062,0.6315,0.6625,0.6078,0.6056,0.6736,0.7148,0.6513,0.8975,0.6468,0.6108,0.7597,0.7331,0.6615,0.9338,0.7076,0.6437,0.8871,0.4432,0.4270,0.3467