Download evaluation/index.html from ifmain/Moderation-03: direct link, hf CLI and curl.
- Browser
- Download file 36.6 kB
-
https://huggingface.co/ifmain/Moderation-03/resolve/main/evaluation/index.html
- Command line
-
hf download hf://ifmain/Moderation-03/evaluation/index.html
-
curl -L -o index.html https://huggingface.co/ifmain/Moderation-03/resolve/main/evaluation/index.html
36.6 kB
| <html lang="ru"><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><title>Moderation 03 — калибровка и тест</title><style>body{font:16px system-ui;background:#101723;color:#eef3fc;max-width:1150px;margin:36px auto;padding:0 24px}h1,h2{color:#8cc9ff}h2{margin-top:36px}table{border-collapse:collapse;width:100%;font-variant-numeric:tabular-nums}th,td{text-align:right;padding:10px;border-bottom:1px solid #354457}th:first-child{text-align:left}details{margin:14px 0}summary{cursor:pointer;color:#8cc9ff}.note{padding:20px;background:#302719}.scroll{overflow:auto}a{color:#8cc9ff}p{line-height:1.6}</style><h1>Moderation 03 · эпоха 5</h1><p class="note">Этот отчёт проверяет исходную политику контента. Владелец также предполагает запрет нерабочих тем в корпоративных чатах; отдельный детектор релевантности рабочей теме здесь не реализован и не оценивался. Запрет темы организацией не меняет её сырой тег на hate.</p><p>Пороги подобраны только на 828 внутренних примерах. Отложенная проверка: 412 примеров. Внешний тест: 1 680. Веса модели фиксированы.</p><p class="note">Это калибровка порогов, не подтверждение исправления атаки исследователей. Внутренние метки преимущественно унаследованы от исходного датасета; новые проверки синтетические. Переводы связаны: число строк не равно числу независимых ситуаций. Незаконная деятельность не имеет отдельного выхода модели.</p><h2>Внутренний отложенный тест · после калибровки</h2><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>412</td><td>109</td><td>35</td><td>244</td><td>24</td><td>75.7%</td><td>82.0%</td><td>12.5%</td><td>78.7%</td></tr><tr><th>Средний</th><td>412</td><td>189</td><td>51</td><td>146</td><td>26</td><td>78.8%</td><td>87.9%</td><td>25.9%</td><td>83.1%</td></tr><tr><th>Высокий</th><td>412</td><td>245</td><td>24</td><td>121</td><td>22</td><td>91.1%</td><td>91.8%</td><td>16.6%</td><td>91.4%</td></tr><tr><th>Корпоративный</th><td>412</td><td>252</td><td>22</td><td>115</td><td>23</td><td>92.0%</td><td>91.6%</td><td>16.1%</td><td>91.8%</td></tr></table></div><details><summary>До калибровки</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>412</td><td>20</td><td>0</td><td>279</td><td>113</td><td>100.0%</td><td>15.0%</td><td>0.0%</td><td>26.1%</td></tr><tr><th>Средний</th><td>412</td><td>118</td><td>19</td><td>178</td><td>97</td><td>86.1%</td><td>54.9%</td><td>9.6%</td><td>67.0%</td></tr><tr><th>Высокий</th><td>412</td><td>208</td><td>7</td><td>138</td><td>59</td><td>96.7%</td><td>77.9%</td><td>4.8%</td><td>86.3%</td></tr><tr><th>Корпоративный</th><td>412</td><td>246</td><td>19</td><td>118</td><td>29</td><td>92.8%</td><td>89.5%</td><td>13.9%</td><td>91.1%</td></tr></table></div></details><h2>Теги по уровням · внутренний тест</h2><details><summary>Лёгкий</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>harassment_threatening</th><td>412</td><td>17</td><td>22</td><td>351</td><td>22</td><td>43.6%</td><td>43.6%</td><td>5.9%</td><td>43.6%</td></tr><tr><th>hate</th><td>412</td><td>39</td><td>7</td><td>350</td><td>16</td><td>84.8%</td><td>70.9%</td><td>2.0%</td><td>77.2%</td></tr><tr><th>hate_threatening</th><td>412</td><td>26</td><td>10</td><td>365</td><td>11</td><td>72.2%</td><td>70.3%</td><td>2.7%</td><td>71.2%</td></tr><tr><th>self_harm_instructions</th><td>412</td><td>4</td><td>1</td><td>388</td><td>19</td><td>80.0%</td><td>17.4%</td><td>0.3%</td><td>28.6%</td></tr><tr><th>sexual_minors</th><td>412</td><td>23</td><td>18</td><td>369</td><td>2</td><td>56.1%</td><td>92.0%</td><td>4.7%</td><td>69.7%</td></tr><tr><th>violence_graphic</th><td>412</td><td>22</td><td>46</td><td>340</td><td>4</td><td>32.4%</td><td>84.6%</td><td>11.9%</td><td>46.8%</td></tr></table></div></details><details><summary>Средний</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>harassment</th><td>412</td><td>82</td><td>37</td><td>265</td><td>28</td><td>68.9%</td><td>74.5%</td><td>12.3%</td><td>71.6%</td></tr><tr><th>harassment_threatening</th><td>412</td><td>53</td><td>59</td><td>294</td><td>6</td><td>47.3%</td><td>89.8%</td><td>16.7%</td><td>62.0%</td></tr><tr><th>hate</th><td>412</td><td>55</td><td>40</td><td>315</td><td>2</td><td>57.9%</td><td>96.5%</td><td>11.3%</td><td>72.4%</td></tr><tr><th>hate_threatening</th><td>412</td><td>26</td><td>10</td><td>365</td><td>11</td><td>72.2%</td><td>70.3%</td><td>2.7%</td><td>71.2%</td></tr><tr><th>self_harm</th><td>412</td><td>18</td><td>5</td><td>370</td><td>19</td><td>78.3%</td><td>48.6%</td><td>1.3%</td><td>60.0%</td></tr><tr><th>self_harm_instructions</th><td>412</td><td>18</td><td>24</td><td>363</td><td>7</td><td>42.9%</td><td>72.0%</td><td>6.2%</td><td>53.7%</td></tr><tr><th>self_harm_intent</th><td>412</td><td>27</td><td>12</td><td>364</td><td>9</td><td>69.2%</td><td>75.0%</td><td>3.2%</td><td>72.0%</td></tr><tr><th>sexual</th><td>412</td><td>21</td><td>6</td><td>378</td><td>7</td><td>77.8%</td><td>75.0%</td><td>1.6%</td><td>76.4%</td></tr><tr><th>sexual_minors</th><td>412</td><td>23</td><td>26</td><td>361</td><td>2</td><td>46.9%</td><td>92.0%</td><td>6.7%</td><td>62.2%</td></tr><tr><th>violence</th><td>412</td><td>63</td><td>66</td><td>279</td><td>4</td><td>48.8%</td><td>94.0%</td><td>19.1%</td><td>64.3%</td></tr><tr><th>violence_graphic</th><td>412</td><td>23</td><td>45</td><td>339</td><td>5</td><td>33.8%</td><td>82.1%</td><td>11.7%</td><td>47.9%</td></tr></table></div></details><details><summary>Высокий</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>harassment</th><td>412</td><td>130</td><td>43</td><td>215</td><td>24</td><td>75.1%</td><td>84.4%</td><td>16.7%</td><td>79.5%</td></tr><tr><th>harassment_threatening</th><td>412</td><td>56</td><td>56</td><td>293</td><td>7</td><td>50.0%</td><td>88.9%</td><td>16.0%</td><td>64.0%</td></tr><tr><th>hate</th><td>412</td><td>62</td><td>35</td><td>311</td><td>4</td><td>63.9%</td><td>93.9%</td><td>10.1%</td><td>76.1%</td></tr><tr><th>hate_threatening</th><td>412</td><td>26</td><td>10</td><td>365</td><td>11</td><td>72.2%</td><td>70.3%</td><td>2.7%</td><td>71.2%</td></tr><tr><th>self_harm</th><td>412</td><td>31</td><td>15</td><td>354</td><td>12</td><td>67.4%</td><td>72.1%</td><td>4.1%</td><td>69.7%</td></tr><tr><th>self_harm_instructions</th><td>412</td><td>18</td><td>24</td><td>363</td><td>7</td><td>42.9%</td><td>72.0%</td><td>6.2%</td><td>53.7%</td></tr><tr><th>self_harm_intent</th><td>412</td><td>29</td><td>10</td><td>363</td><td>10</td><td>74.4%</td><td>74.4%</td><td>2.7%</td><td>74.4%</td></tr><tr><th>sexual</th><td>412</td><td>28</td><td>11</td><td>342</td><td>31</td><td>71.8%</td><td>47.5%</td><td>3.1%</td><td>57.1%</td></tr><tr><th>sexual_minors</th><td>412</td><td>23</td><td>26</td><td>361</td><td>2</td><td>46.9%</td><td>92.0%</td><td>6.7%</td><td>62.2%</td></tr><tr><th>violence</th><td>412</td><td>68</td><td>61</td><td>274</td><td>9</td><td>52.7%</td><td>88.3%</td><td>18.2%</td><td>66.0%</td></tr><tr><th>violence_graphic</th><td>412</td><td>24</td><td>44</td><td>339</td><td>5</td><td>35.3%</td><td>82.8%</td><td>11.5%</td><td>49.5%</td></tr></table></div></details><details><summary>Корпоративный</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>harassment</th><td>412</td><td>136</td><td>37</td><td>208</td><td>31</td><td>78.6%</td><td>81.4%</td><td>15.1%</td><td>80.0%</td></tr><tr><th>harassment_threatening</th><td>412</td><td>62</td><td>97</td><td>250</td><td>3</td><td>39.0%</td><td>95.4%</td><td>28.0%</td><td>55.4%</td></tr><tr><th>hate</th><td>412</td><td>70</td><td>27</td><td>310</td><td>5</td><td>72.2%</td><td>93.3%</td><td>8.0%</td><td>81.4%</td></tr><tr><th>hate_threatening</th><td>412</td><td>26</td><td>10</td><td>365</td><td>11</td><td>72.2%</td><td>70.3%</td><td>2.7%</td><td>71.2%</td></tr><tr><th>self_harm</th><td>412</td><td>35</td><td>19</td><td>348</td><td>10</td><td>64.8%</td><td>77.8%</td><td>5.2%</td><td>70.7%</td></tr><tr><th>self_harm_instructions</th><td>412</td><td>18</td><td>24</td><td>363</td><td>7</td><td>42.9%</td><td>72.0%</td><td>6.2%</td><td>53.7%</td></tr><tr><th>self_harm_intent</th><td>412</td><td>37</td><td>16</td><td>352</td><td>7</td><td>69.8%</td><td>84.1%</td><td>4.3%</td><td>76.3%</td></tr><tr><th>sexual</th><td>412</td><td>32</td><td>10</td><td>336</td><td>34</td><td>76.2%</td><td>48.5%</td><td>2.9%</td><td>59.3%</td></tr><tr><th>sexual_minors</th><td>412</td><td>24</td><td>25</td><td>361</td><td>2</td><td>49.0%</td><td>92.3%</td><td>6.5%</td><td>64.0%</td></tr><tr><th>violence</th><td>412</td><td>90</td><td>46</td><td>264</td><td>12</td><td>66.2%</td><td>88.2%</td><td>14.8%</td><td>75.6%</td></tr><tr><th>violence_graphic</th><td>412</td><td>24</td><td>44</td><td>339</td><td>5</td><td>35.3%</td><td>82.8%</td><td>11.5%</td><td>49.5%</td></tr></table></div></details><h2>Пять направлений из письма · дополнительные проверки</h2><p>Положительные запросы, безопасные контрпримеры, изменения написания; 10 переформулировок EN/RU. Это небольшой диагностический набор, не внешний бенчмарк. Эротика взрослых разрешена только на лёгком уровне; остальные четыре направления блокируются на всех уровнях.</p><h3>disturbing_content</h3><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>53</td><td>23</td><td>0</td><td>17</td><td>13</td><td>100.0%</td><td>63.9%</td><td>0.0%</td><td>78.0%</td></tr><tr><th>Средний</th><td>53</td><td>24</td><td>0</td><td>17</td><td>12</td><td>100.0%</td><td>66.7%</td><td>0.0%</td><td>80.0%</td></tr><tr><th>Высокий</th><td>53</td><td>27</td><td>1</td><td>16</td><td>9</td><td>96.4%</td><td>75.0%</td><td>5.9%</td><td>84.4%</td></tr><tr><th>Корпоративный</th><td>53</td><td>27</td><td>1</td><td>16</td><td>9</td><td>96.4%</td><td>75.0%</td><td>5.9%</td><td>84.4%</td></tr></table></div><h3>hate</h3><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>53</td><td>22</td><td>4</td><td>13</td><td>14</td><td>84.6%</td><td>61.1%</td><td>23.5%</td><td>71.0%</td></tr><tr><th>Средний</th><td>53</td><td>36</td><td>16</td><td>1</td><td>0</td><td>69.2%</td><td>100.0%</td><td>94.1%</td><td>81.8%</td></tr><tr><th>Высокий</th><td>53</td><td>36</td><td>16</td><td>1</td><td>0</td><td>69.2%</td><td>100.0%</td><td>94.1%</td><td>81.8%</td></tr><tr><th>Корпоративный</th><td>53</td><td>36</td><td>16</td><td>1</td><td>0</td><td>69.2%</td><td>100.0%</td><td>94.1%</td><td>81.8%</td></tr></table></div><h3>illegal_activity</h3><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>53</td><td>0</td><td>0</td><td>17</td><td>36</td><td>—</td><td>0.0%</td><td>0.0%</td><td>0.0%</td></tr><tr><th>Средний</th><td>53</td><td>0</td><td>0</td><td>17</td><td>36</td><td>—</td><td>0.0%</td><td>0.0%</td><td>0.0%</td></tr><tr><th>Высокий</th><td>53</td><td>2</td><td>0</td><td>17</td><td>34</td><td>100.0%</td><td>5.6%</td><td>0.0%</td><td>10.5%</td></tr><tr><th>Корпоративный</th><td>53</td><td>7</td><td>0</td><td>17</td><td>29</td><td>100.0%</td><td>19.4%</td><td>0.0%</td><td>32.6%</td></tr></table></div><h3>pornography</h3><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>53</td><td>0</td><td>32</td><td>21</td><td>0</td><td>0.0%</td><td>—</td><td>60.4%</td><td>0.0%</td></tr><tr><th>Средний</th><td>53</td><td>35</td><td>6</td><td>11</td><td>1</td><td>85.4%</td><td>97.2%</td><td>35.3%</td><td>90.9%</td></tr><tr><th>Высокий</th><td>53</td><td>36</td><td>7</td><td>10</td><td>0</td><td>83.7%</td><td>100.0%</td><td>41.2%</td><td>91.1%</td></tr><tr><th>Корпоративный</th><td>53</td><td>36</td><td>7</td><td>10</td><td>0</td><td>83.7%</td><td>100.0%</td><td>41.2%</td><td>91.1%</td></tr></table></div><h3>violence</h3><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>53</td><td>23</td><td>7</td><td>10</td><td>13</td><td>76.7%</td><td>63.9%</td><td>41.2%</td><td>69.7%</td></tr><tr><th>Средний</th><td>53</td><td>32</td><td>11</td><td>6</td><td>4</td><td>74.4%</td><td>88.9%</td><td>64.7%</td><td>81.0%</td></tr><tr><th>Высокий</th><td>53</td><td>32</td><td>12</td><td>5</td><td>4</td><td>72.7%</td><td>88.9%</td><td>70.6%</td><td>80.0%</td></tr><tr><th>Корпоративный</th><td>53</td><td>32</td><td>17</td><td>0</td><td>4</td><td>65.3%</td><td>88.9%</td><td>100.0%</td><td>75.3%</td></tr></table></div><h2>Внешний тест · восемь категорий</h2><p>Неизвестные метки исключены. Стандартные пороги категорий также подобраны только на внутренней части.</p><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>sexual</th><td>984</td><td>228</td><td>77</td><td>670</td><td>9</td><td>74.8%</td><td>96.2%</td><td>10.3%</td><td>84.1%</td></tr><tr><th>hate</th><td>771</td><td>158</td><td>196</td><td>413</td><td>4</td><td>44.6%</td><td>97.5%</td><td>32.2%</td><td>61.2%</td></tr><tr><th>violence</th><td>1450</td><td>86</td><td>189</td><td>1167</td><td>8</td><td>31.3%</td><td>91.5%</td><td>13.9%</td><td>46.6%</td></tr><tr><th>harassment</th><td>1444</td><td>67</td><td>443</td><td>925</td><td>9</td><td>13.1%</td><td>88.2%</td><td>32.4%</td><td>22.9%</td></tr><tr><th>self_harm</th><td>1447</td><td>43</td><td>106</td><td>1290</td><td>8</td><td>28.9%</td><td>84.3%</td><td>7.6%</td><td>43.0%</td></tr><tr><th>sexual_minors</th><td>994</td><td>85</td><td>212</td><td>697</td><td>0</td><td>28.6%</td><td>100.0%</td><td>23.3%</td><td>44.5%</td></tr><tr><th>hate_threatening</th><td>761</td><td>36</td><td>69</td><td>651</td><td>5</td><td>34.3%</td><td>87.8%</td><td>9.6%</td><td>49.3%</td></tr><tr><th>violence_graphic</th><td>1447</td><td>21</td><td>147</td><td>1276</td><td>3</td><td>12.5%</td><td>87.5%</td><td>10.3%</td><td>21.9%</td></tr></table></div><p>Лимит модели: 512 токенов. Обрезано 76 из 1680 внешних текстов. Они остаются в основном тесте, отражая реальный режим работы.</p><h2>Внешний тест · проекция четырёх уровней</h2><p>Внешний набор не размечен по нашим четырём политикам. Здесь сравнивается объединение доступных бинарных категорий. Правило мата и три отсутствующие категории исключены; это не полная точность уровня модерации.</p><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>800</td><td>202</td><td>210</td><td>323</td><td>65</td><td>49.0%</td><td>75.7%</td><td>39.4%</td><td>59.5%</td></tr><tr><th>Средний</th><td>859</td><td>493</td><td>196</td><td>141</td><td>29</td><td>71.6%</td><td>94.4%</td><td>58.2%</td><td>81.4%</td></tr><tr><th>Высокий</th><td>859</td><td>504</td><td>220</td><td>117</td><td>18</td><td>69.6%</td><td>96.6%</td><td>65.3%</td><td>80.9%</td></tr><tr><th>Корпоративный</th><td>859</td><td>505</td><td>222</td><td>115</td><td>17</td><td>69.5%</td><td>96.7%</td><td>65.9%</td><td>80.9%</td></tr></table></div><h2>Внутренний тест по языкам</h2><details><summary>be</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>24</td><td>6</td><td>2</td><td>15</td><td>1</td><td>75.0%</td><td>85.7%</td><td>11.8%</td><td>80.0%</td></tr><tr><th>Средний</th><td>24</td><td>9</td><td>5</td><td>7</td><td>3</td><td>64.3%</td><td>75.0%</td><td>41.7%</td><td>69.2%</td></tr><tr><th>Высокий</th><td>24</td><td>14</td><td>3</td><td>6</td><td>1</td><td>82.4%</td><td>93.3%</td><td>33.3%</td><td>87.5%</td></tr><tr><th>Корпоративный</th><td>24</td><td>15</td><td>2</td><td>6</td><td>1</td><td>88.2%</td><td>93.8%</td><td>25.0%</td><td>90.9%</td></tr></table></div></details><details><summary>cs</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>23</td><td>7</td><td>0</td><td>16</td><td>0</td><td>100.0%</td><td>100.0%</td><td>0.0%</td><td>100.0%</td></tr><tr><th>Средний</th><td>23</td><td>11</td><td>3</td><td>8</td><td>1</td><td>78.6%</td><td>91.7%</td><td>27.3%</td><td>84.6%</td></tr><tr><th>Высокий</th><td>23</td><td>15</td><td>2</td><td>6</td><td>0</td><td>88.2%</td><td>100.0%</td><td>25.0%</td><td>93.8%</td></tr><tr><th>Корпоративный</th><td>23</td><td>15</td><td>2</td><td>6</td><td>0</td><td>88.2%</td><td>100.0%</td><td>25.0%</td><td>93.8%</td></tr></table></div></details><details><summary>de</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>24</td><td>7</td><td>1</td><td>14</td><td>2</td><td>87.5%</td><td>77.8%</td><td>6.7%</td><td>82.4%</td></tr><tr><th>Средний</th><td>24</td><td>13</td><td>1</td><td>9</td><td>1</td><td>92.9%</td><td>92.9%</td><td>10.0%</td><td>92.9%</td></tr><tr><th>Высокий</th><td>24</td><td>15</td><td>0</td><td>8</td><td>1</td><td>100.0%</td><td>93.8%</td><td>0.0%</td><td>96.8%</td></tr><tr><th>Корпоративный</th><td>24</td><td>15</td><td>0</td><td>8</td><td>1</td><td>100.0%</td><td>93.8%</td><td>0.0%</td><td>96.8%</td></tr></table></div></details><details><summary>en</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>25</td><td>8</td><td>2</td><td>15</td><td>0</td><td>80.0%</td><td>100.0%</td><td>11.8%</td><td>88.9%</td></tr><tr><th>Средний</th><td>25</td><td>12</td><td>0</td><td>13</td><td>0</td><td>100.0%</td><td>100.0%</td><td>0.0%</td><td>100.0%</td></tr><tr><th>Высокий</th><td>25</td><td>14</td><td>0</td><td>11</td><td>0</td><td>100.0%</td><td>100.0%</td><td>0.0%</td><td>100.0%</td></tr><tr><th>Корпоративный</th><td>25</td><td>14</td><td>0</td><td>10</td><td>1</td><td>100.0%</td><td>93.3%</td><td>0.0%</td><td>96.6%</td></tr></table></div></details><details><summary>es</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>25</td><td>6</td><td>2</td><td>15</td><td>2</td><td>75.0%</td><td>75.0%</td><td>11.8%</td><td>75.0%</td></tr><tr><th>Средний</th><td>25</td><td>13</td><td>2</td><td>9</td><td>1</td><td>86.7%</td><td>92.9%</td><td>18.2%</td><td>89.7%</td></tr><tr><th>Высокий</th><td>25</td><td>16</td><td>1</td><td>6</td><td>2</td><td>94.1%</td><td>88.9%</td><td>14.3%</td><td>91.4%</td></tr><tr><th>Корпоративный</th><td>25</td><td>17</td><td>0</td><td>6</td><td>2</td><td>100.0%</td><td>89.5%</td><td>0.0%</td><td>94.4%</td></tr></table></div></details><details><summary>fi</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>24</td><td>5</td><td>3</td><td>13</td><td>3</td><td>62.5%</td><td>62.5%</td><td>18.8%</td><td>62.5%</td></tr><tr><th>Средний</th><td>24</td><td>11</td><td>4</td><td>8</td><td>1</td><td>73.3%</td><td>91.7%</td><td>33.3%</td><td>81.5%</td></tr><tr><th>Высокий</th><td>24</td><td>13</td><td>3</td><td>6</td><td>2</td><td>81.2%</td><td>86.7%</td><td>33.3%</td><td>83.9%</td></tr><tr><th>Корпоративный</th><td>24</td><td>13</td><td>3</td><td>6</td><td>2</td><td>81.2%</td><td>86.7%</td><td>33.3%</td><td>83.9%</td></tr></table></div></details><details><summary>fr</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>23</td><td>7</td><td>3</td><td>12</td><td>1</td><td>70.0%</td><td>87.5%</td><td>20.0%</td><td>77.8%</td></tr><tr><th>Средний</th><td>23</td><td>12</td><td>4</td><td>7</td><td>0</td><td>75.0%</td><td>100.0%</td><td>36.4%</td><td>85.7%</td></tr><tr><th>Высокий</th><td>23</td><td>16</td><td>0</td><td>7</td><td>0</td><td>100.0%</td><td>100.0%</td><td>0.0%</td><td>100.0%</td></tr><tr><th>Корпоративный</th><td>23</td><td>16</td><td>1</td><td>6</td><td>0</td><td>94.1%</td><td>100.0%</td><td>14.3%</td><td>97.0%</td></tr></table></div></details><details><summary>it</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>25</td><td>7</td><td>4</td><td>13</td><td>1</td><td>63.6%</td><td>87.5%</td><td>23.5%</td><td>73.7%</td></tr><tr><th>Средний</th><td>25</td><td>13</td><td>4</td><td>8</td><td>0</td><td>76.5%</td><td>100.0%</td><td>33.3%</td><td>86.7%</td></tr><tr><th>Высокий</th><td>25</td><td>16</td><td>2</td><td>7</td><td>0</td><td>88.9%</td><td>100.0%</td><td>22.2%</td><td>94.1%</td></tr><tr><th>Корпоративный</th><td>25</td><td>16</td><td>2</td><td>7</td><td>0</td><td>88.9%</td><td>100.0%</td><td>22.2%</td><td>94.1%</td></tr></table></div></details><details><summary>ja</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>24</td><td>6</td><td>1</td><td>14</td><td>3</td><td>85.7%</td><td>66.7%</td><td>6.7%</td><td>75.0%</td></tr><tr><th>Средний</th><td>24</td><td>8</td><td>4</td><td>8</td><td>4</td><td>66.7%</td><td>66.7%</td><td>33.3%</td><td>66.7%</td></tr><tr><th>Высокий</th><td>24</td><td>12</td><td>3</td><td>7</td><td>2</td><td>80.0%</td><td>85.7%</td><td>30.0%</td><td>82.8%</td></tr><tr><th>Корпоративный</th><td>24</td><td>13</td><td>2</td><td>7</td><td>2</td><td>86.7%</td><td>86.7%</td><td>22.2%</td><td>86.7%</td></tr></table></div></details><details><summary>kk</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>25</td><td>6</td><td>3</td><td>13</td><td>3</td><td>66.7%</td><td>66.7%</td><td>18.8%</td><td>66.7%</td></tr><tr><th>Средний</th><td>25</td><td>12</td><td>5</td><td>7</td><td>1</td><td>70.6%</td><td>92.3%</td><td>41.7%</td><td>80.0%</td></tr><tr><th>Высокий</th><td>25</td><td>17</td><td>2</td><td>6</td><td>0</td><td>89.5%</td><td>100.0%</td><td>25.0%</td><td>94.4%</td></tr><tr><th>Корпоративный</th><td>25</td><td>17</td><td>3</td><td>5</td><td>0</td><td>85.0%</td><td>100.0%</td><td>37.5%</td><td>91.9%</td></tr></table></div></details><details><summary>ko</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>24</td><td>6</td><td>2</td><td>15</td><td>1</td><td>75.0%</td><td>85.7%</td><td>11.8%</td><td>80.0%</td></tr><tr><th>Средний</th><td>24</td><td>10</td><td>3</td><td>10</td><td>1</td><td>76.9%</td><td>90.9%</td><td>23.1%</td><td>83.3%</td></tr><tr><th>Высокий</th><td>24</td><td>14</td><td>2</td><td>7</td><td>1</td><td>87.5%</td><td>93.3%</td><td>22.2%</td><td>90.3%</td></tr><tr><th>Корпоративный</th><td>24</td><td>15</td><td>1</td><td>7</td><td>1</td><td>93.8%</td><td>93.8%</td><td>12.5%</td><td>93.8%</td></tr></table></div></details><details><summary>lv</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>24</td><td>5</td><td>1</td><td>15</td><td>3</td><td>83.3%</td><td>62.5%</td><td>6.2%</td><td>71.4%</td></tr><tr><th>Средний</th><td>24</td><td>9</td><td>1</td><td>11</td><td>3</td><td>90.0%</td><td>75.0%</td><td>8.3%</td><td>81.8%</td></tr><tr><th>Высокий</th><td>24</td><td>13</td><td>0</td><td>8</td><td>3</td><td>100.0%</td><td>81.2%</td><td>0.0%</td><td>89.7%</td></tr><tr><th>Корпоративный</th><td>24</td><td>13</td><td>0</td><td>8</td><td>3</td><td>100.0%</td><td>81.2%</td><td>0.0%</td><td>89.7%</td></tr></table></div></details><details><summary>pl</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>23</td><td>5</td><td>4</td><td>12</td><td>2</td><td>55.6%</td><td>71.4%</td><td>25.0%</td><td>62.5%</td></tr><tr><th>Средний</th><td>23</td><td>11</td><td>3</td><td>7</td><td>2</td><td>78.6%</td><td>84.6%</td><td>30.0%</td><td>81.5%</td></tr><tr><th>Высокий</th><td>23</td><td>12</td><td>2</td><td>6</td><td>3</td><td>85.7%</td><td>80.0%</td><td>25.0%</td><td>82.8%</td></tr><tr><th>Корпоративный</th><td>23</td><td>12</td><td>2</td><td>5</td><td>4</td><td>85.7%</td><td>75.0%</td><td>28.6%</td><td>80.0%</td></tr></table></div></details><details><summary>ru</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>24</td><td>7</td><td>3</td><td>14</td><td>0</td><td>70.0%</td><td>100.0%</td><td>17.6%</td><td>82.4%</td></tr><tr><th>Средний</th><td>24</td><td>11</td><td>3</td><td>8</td><td>2</td><td>78.6%</td><td>84.6%</td><td>27.3%</td><td>81.5%</td></tr><tr><th>Высокий</th><td>24</td><td>14</td><td>0</td><td>8</td><td>2</td><td>100.0%</td><td>87.5%</td><td>0.0%</td><td>93.3%</td></tr><tr><th>Корпоративный</th><td>24</td><td>14</td><td>0</td><td>8</td><td>2</td><td>100.0%</td><td>87.5%</td><td>0.0%</td><td>93.3%</td></tr></table></div></details><details><summary>sv</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>25</td><td>7</td><td>0</td><td>17</td><td>1</td><td>100.0%</td><td>87.5%</td><td>0.0%</td><td>93.3%</td></tr><tr><th>Средний</th><td>25</td><td>12</td><td>2</td><td>9</td><td>2</td><td>85.7%</td><td>85.7%</td><td>18.2%</td><td>85.7%</td></tr><tr><th>Высокий</th><td>25</td><td>16</td><td>0</td><td>8</td><td>1</td><td>100.0%</td><td>94.1%</td><td>0.0%</td><td>97.0%</td></tr><tr><th>Корпоративный</th><td>25</td><td>17</td><td>1</td><td>7</td><td>0</td><td>94.4%</td><td>100.0%</td><td>12.5%</td><td>97.1%</td></tr></table></div></details><details><summary>uk</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>25</td><td>7</td><td>2</td><td>16</td><td>0</td><td>77.8%</td><td>100.0%</td><td>11.1%</td><td>87.5%</td></tr><tr><th>Средний</th><td>25</td><td>12</td><td>3</td><td>9</td><td>1</td><td>80.0%</td><td>92.3%</td><td>25.0%</td><td>85.7%</td></tr><tr><th>Высокий</th><td>25</td><td>14</td><td>2</td><td>7</td><td>2</td><td>87.5%</td><td>87.5%</td><td>22.2%</td><td>87.5%</td></tr><tr><th>Корпоративный</th><td>25</td><td>15</td><td>2</td><td>6</td><td>2</td><td>88.2%</td><td>88.2%</td><td>25.0%</td><td>88.2%</td></tr></table></div></details><details><summary>zh</summary><div class="scroll"><table><tr><th>Уровень / категория</th><th>n</th><th>TP</th><th>FP</th><th>TN</th><th>FN</th><th>precision</th><th>recall</th><th>false_positive_rate</th><th>f1</th></tr><tr><th>Лёгкий</th><td>25</td><td>7</td><td>2</td><td>15</td><td>1</td><td>77.8%</td><td>87.5%</td><td>11.8%</td><td>82.4%</td></tr><tr><th>Средний</th><td>25</td><td>10</td><td>4</td><td>8</td><td>3</td><td>71.4%</td><td>76.9%</td><td>33.3%</td><td>74.1%</td></tr><tr><th>Высокий</th><td>25</td><td>14</td><td>2</td><td>7</td><td>2</td><td>87.5%</td><td>87.5%</td><td>22.2%</td><td>87.5%</td></tr><tr><th>Корпоративный</th><td>25</td><td>15</td><td>1</td><td>7</td><td>2</td><td>93.8%</td><td>88.2%</td><td>12.5%</td><td>90.9%</td></tr></table></div></details><h2>Карта покрытия 11 × 17</h2><p>Положительные примеры по исходной оценке ≥ 0,5. В ячейках: калибровка / отложенная проверка. Переводы связаны общими группами.</p><div class="scroll"><table><tr><th>Категория</th><th>en</th><th>de</th><th>fr</th><th>es</th><th>it</th><th>sv</th><th>fi</th><th>pl</th><th>cs</th><th>lv</th><th>zh</th><th>ja</th><th>ko</th><th>ru</th><th>uk</th><th>be</th><th>kk</th></tr><tr><th>harassment</th><td>10 / 5</td><td>17 / 6</td><td>12 / 9</td><td>17 / 8</td><td>12 / 7</td><td>14 / 9</td><td>11 / 7</td><td>14 / 7</td><td>16 / 6</td><td>17 / 7</td><td>15 / 6</td><td>13 / 6</td><td>12 / 7</td><td>17 / 6</td><td>14 / 7</td><td>14 / 6</td><td>12 / 9</td></tr><tr><th>harassment_threatening</th><td>3 / 4</td><td>5 / 4</td><td>4 / 3</td><td>3 / 4</td><td>4 / 3</td><td>4 / 3</td><td>4 / 3</td><td>4 / 3</td><td>5 / 3</td><td>4 / 3</td><td>4 / 3</td><td>3 / 5</td><td>2 / 4</td><td>3 / 3</td><td>3 / 3</td><td>3 / 4</td><td>3 / 4</td></tr><tr><th>hate</th><td>3 / 4</td><td>4 / 3</td><td>3 / 4</td><td>4 / 3</td><td>4 / 3</td><td>3 / 4</td><td>4 / 3</td><td>3 / 4</td><td>7 / 3</td><td>4 / 4</td><td>3 / 4</td><td>5 / 4</td><td>3 / 3</td><td>4 / 3</td><td>3 / 3</td><td>5 / 3</td><td>2 / 5</td></tr><tr><th>hate_threatening</th><td>1 / 3</td><td>3 / 2</td><td>2 / 2</td><td>2 / 2</td><td>2 / 2</td><td>2 / 2</td><td>2 / 2</td><td>1 / 2</td><td>2 / 2</td><td>2 / 2</td><td>2 / 2</td><td>2 / 2</td><td>1 / 2</td><td>1 / 3</td><td>1 / 2</td><td>2 / 2</td><td>1 / 3</td></tr><tr><th>self_harm</th><td>3 / 3</td><td>2 / 4</td><td>4 / 2</td><td>5 / 2</td><td>4 / 3</td><td>4 / 2</td><td>3 / 4</td><td>4 / 2</td><td>5 / 3</td><td>3 / 2</td><td>5 / 2</td><td>4 / 3</td><td>4 / 2</td><td>3 / 3</td><td>4 / 3</td><td>4 / 2</td><td>3 / 3</td></tr><tr><th>self_harm_instructions</th><td>2 / 1</td><td>1 / 2</td><td>3 / 1</td><td>3 / 1</td><td>2 / 2</td><td>3 / 1</td><td>2 / 2</td><td>2 / 1</td><td>2 / 2</td><td>1 / 2</td><td>3 / 1</td><td>2 / 2</td><td>2 / 1</td><td>2 / 1</td><td>1 / 2</td><td>2 / 1</td><td>1 / 2</td></tr><tr><th>self_harm_intent</th><td>3 / 3</td><td>2 / 4</td><td>3 / 2</td><td>5 / 2</td><td>4 / 3</td><td>4 / 2</td><td>3 / 4</td><td>4 / 2</td><td>4 / 2</td><td>3 / 2</td><td>4 / 2</td><td>4 / 3</td><td>4 / 3</td><td>3 / 3</td><td>4 / 2</td><td>3 / 2</td><td>3 / 3</td></tr><tr><th>sexual</th><td>6 / 3</td><td>7 / 3</td><td>10 / 3</td><td>7 / 4</td><td>10 / 5</td><td>7 / 4</td><td>9 / 4</td><td>8 / 3</td><td>7 / 3</td><td>9 / 5</td><td>7 / 4</td><td>6 / 3</td><td>9 / 5</td><td>8 / 2</td><td>6 / 3</td><td>12 / 2</td><td>10 / 3</td></tr><tr><th>sexual_minors</th><td>4 / 2</td><td>5 / 1</td><td>5 / 2</td><td>5 / 2</td><td>5 / 2</td><td>4 / 2</td><td>5 / 2</td><td>5 / 1</td><td>4 / 1</td><td>4 / 2</td><td>4 / 2</td><td>4 / 1</td><td>5 / 1</td><td>5 / 1</td><td>5 / 1</td><td>5 / 1</td><td>5 / 1</td></tr><tr><th>violence</th><td>11 / 5</td><td>9 / 5</td><td>12 / 3</td><td>8 / 5</td><td>10 / 4</td><td>7 / 4</td><td>10 / 5</td><td>10 / 4</td><td>7 / 4</td><td>8 / 4</td><td>9 / 6</td><td>7 / 6</td><td>8 / 4</td><td>5 / 6</td><td>7 / 4</td><td>9 / 7</td><td>10 / 5</td></tr><tr><th>violence_graphic</th><td>5 / 1</td><td>3 / 3</td><td>5 / 1</td><td>3 / 2</td><td>4 / 1</td><td>3 / 2</td><td>4 / 1</td><td>4 / 1</td><td>3 / 2</td><td>4 / 1</td><td>3 / 2</td><td>3 / 2</td><td>4 / 1</td><td>2 / 2</td><td>3 / 2</td><td>3 / 2</td><td>5 / 1</td></tr></table></div><p>TP — верно заблокировано; FP — лишняя блокировка; TN — верно пропущено; FN — пропуск запрещённого. Precision — доля верных блокировок; recall — доля обнаруженного запрещённого.</p><p><a href="evaluation.json">Полные метрики JSON</a> · <a href="thresholds.json">Зафиксированные пороги</a> · <a href="scores_without_text.jsonl">Оценки по строкам без исходных текстов</a></p></html> |