Update README.md
Browse files
README.md
CHANGED
|
@@ -16,10 +16,45 @@ widget:
|
|
| 16 |
<!-- Provide a quick summary of what the model is/does. -->
|
| 17 |
|
| 18 |
本模型基於 [Gemma2:2b](https://huggingface.co/google/gemma-2-2b) 進行微調,目的是讓其依據台灣刑法學中常用的「刑法三階理論」,針對大型語言模型生成的詐欺罪「犯罪事實」段落,依照詐欺罪法條所規定的構成要件進行標註。具備生成詐欺罪「犯罪事實」的模型,可以參考以 BLOOM 560M 為基礎的[BLOOM 560M Fraud](https://huggingface.co/jslin09/bloom-560m-finetuned-fraud)微調模型,或是以 Gemma2 為基礎的[Gemma2:2b Fraud](https://huggingface.co/jslin09/gemma2-2b-fraud)微調模型。
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
目前可以識別出來的標註標籤有以下七種具名實體,無法識別出來的構成要件要素具名實體,則會傳回 None。
|
| 20 |
|
| 21 |
<pre>
|
| 22 |
-
<code>
|
| 23 |
from colorama import Fore, Back, Style
|
| 24 |
|
| 25 |
elements = {'LEO_SOC': ('犯罪主體', 'Subject of Crime'),
|
|
@@ -38,7 +73,7 @@ tag_color = {'LEO_SOC': Fore.BLACK + Back.RED,
|
|
| 38 |
'LEO_ROH': Fore.BLACK + Back.BLUE,
|
| 39 |
'LEO_ATP': Fore.WHITE + Back.BLACK,
|
| 40 |
}
|
| 41 |
-
</code>
|
| 42 |
</pre>
|
| 43 |
|
| 44 |
為了要將本模型標註出來的結果以更明顯的方式識別,可以參考以下的程式碼,將本大型語言模型生成的標註結果以及所標註的標籤,同時送入以下的函數,就可以將結果以 colorama 的方式著色標註。
|
|
@@ -89,42 +124,6 @@ def tag_in_color(response_content, tag):
|
|
| 89 |
</code>
|
| 90 |
</pre>
|
| 91 |
|
| 92 |
-
## Model Details
|
| 93 |
-
|
| 94 |
-
### Model Description
|
| 95 |
-
|
| 96 |
-
<!-- Provide a longer summary of what this model is. -->
|
| 97 |
-
本模型目前在識別出詐欺罪犯罪事實構成要件要素的平均準確率(percision)及召回率(recall)分別為0.98及0.75。從本模型訓練初期的語料資料錄為 979 筆開始,採用強化學習的流程,將生成的標註資料,採用人工對齊的方式修正後再投入語料庫中進行訓練。最終訓練用的語料計有 2577 筆,經過微調 3 個回合,就完成了本模型。以下是訓練過程各代的準確率及召回率的變化。
|
| 98 |
-
|
| 99 |
-
|版次|資料量|準確率|召回率|
|
| 100 |
-
|---|---|---|---|
|
| 101 |
-
|v1|979|0.272727273|0.218623482|
|
| 102 |
-
|v2|1538|0.725888325|0.581300813|
|
| 103 |
-
|v3|1886|0.717277487|0.465986395|
|
| 104 |
-
|v4|2173|0.826086957|0.550724638|
|
| 105 |
-
|v5|2577|0.983606557|0.75|
|
| 106 |
-
|
| 107 |
-
|
| 108 |
-
- **Developed by:** [Chun-Hsien Lin](https://huggingface.co/jslin09)
|
| 109 |
-
- **Funded by [optional]:** [More Information Needed]
|
| 110 |
-
- **Shared by [optional]:** [More Information Needed]
|
| 111 |
-
- **Model type:** [More Information Needed]
|
| 112 |
-
- **Language(s) (NLP):** Traditional Chinese
|
| 113 |
-
- **License:** [More Information Needed]
|
| 114 |
-
- **Finetuned from model [optional]:** [Gemma2-2b](https://huggingface.co/google/gemma-2-2b)
|
| 115 |
-
|
| 116 |
-
### Model Sources [optional]
|
| 117 |
-
|
| 118 |
-
<!-- Provide the basic links for the model. -->
|
| 119 |
-
|
| 120 |
-
- **Repository:** [More Information Needed]
|
| 121 |
-
- **Paper [optional]:** [More Information Needed]
|
| 122 |
-
- **Demo [optional]:** [More Information Needed]
|
| 123 |
-
|
| 124 |
-
## Uses
|
| 125 |
-
|
| 126 |
-
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
|
| 127 |
-
|
| 128 |
### Direct Use
|
| 129 |
|
| 130 |
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
|
|
@@ -140,6 +139,7 @@ def tag_in_color(response_content, tag):
|
|
| 140 |
### Out-of-Scope Use
|
| 141 |
|
| 142 |
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
|
|
|
|
| 143 |
|
| 144 |
[More Information Needed]
|
| 145 |
|
|
|
|
| 16 |
<!-- Provide a quick summary of what the model is/does. -->
|
| 17 |
|
| 18 |
本模型基於 [Gemma2:2b](https://huggingface.co/google/gemma-2-2b) 進行微調,目的是讓其依據台灣刑法學中常用的「刑法三階理論」,針對大型語言模型生成的詐欺罪「犯罪事實」段落,依照詐欺罪法條所規定的構成要件進行標註。具備生成詐欺罪「犯罪事實」的模型,可以參考以 BLOOM 560M 為基礎的[BLOOM 560M Fraud](https://huggingface.co/jslin09/bloom-560m-finetuned-fraud)微調模型,或是以 Gemma2 為基礎的[Gemma2:2b Fraud](https://huggingface.co/jslin09/gemma2-2b-fraud)微調模型。
|
| 19 |
+
|
| 20 |
+
## Model Details
|
| 21 |
+
|
| 22 |
+
### Model Description
|
| 23 |
+
|
| 24 |
+
<!-- Provide a longer summary of what this model is. -->
|
| 25 |
+
本模型目前在識別出詐欺罪犯罪事實構成要件要素的平均準確率(percision)及召回率(recall)分別為0.98及0.75。從本模型訓練初期的語料資料錄為 979 筆開始,採用強化學習的流程,將生成的標註資料,採用人工對齊的方式修正後再投入語料庫中進行訓練。最終訓練用的語料計有 2577 筆,經過微調 3 個回合,就完成了本模型。以下是訓練過程各代的準確率及召回率的變化。
|
| 26 |
+
|
| 27 |
+
|版次|資料量|準確率|召回率|
|
| 28 |
+
|---|---|---|---|
|
| 29 |
+
|v1|979|0.272727273|0.218623482|
|
| 30 |
+
|v2|1538|0.725888325|0.581300813|
|
| 31 |
+
|v3|1886|0.717277487|0.465986395|
|
| 32 |
+
|v4|2173|0.826086957|0.550724638|
|
| 33 |
+
|v5|2577|0.983606557|0.75|
|
| 34 |
+
|
| 35 |
+
- **Developed by:** [Chun-Hsien Lin](https://huggingface.co/jslin09)
|
| 36 |
+
- **Funded by [optional]:** [More Information Needed]
|
| 37 |
+
- **Shared by [optional]:** [More Information Needed]
|
| 38 |
+
- **Model type:** [More Information Needed]
|
| 39 |
+
- **Language(s) (NLP):** Traditional Chinese
|
| 40 |
+
- **License:** [More Information Needed]
|
| 41 |
+
- **Finetuned from model [optional]:** [Gemma2-2b](https://huggingface.co/google/gemma-2-2b)
|
| 42 |
+
|
| 43 |
+
### Model Sources [optional]
|
| 44 |
+
|
| 45 |
+
<!-- Provide the basic links for the model. -->
|
| 46 |
+
|
| 47 |
+
- **Repository:** [More Information Needed]
|
| 48 |
+
- **Paper [optional]:** [More Information Needed]
|
| 49 |
+
- **Demo [optional]:** [More Information Needed]
|
| 50 |
+
|
| 51 |
+
## Uses
|
| 52 |
+
|
| 53 |
+
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
|
| 54 |
目前可以識別出來的標註標籤有以下七種具名實體,無法識別出來的構成要件要素具名實體,則會傳回 None。
|
| 55 |
|
| 56 |
<pre>
|
| 57 |
+
<code>
|
| 58 |
from colorama import Fore, Back, Style
|
| 59 |
|
| 60 |
elements = {'LEO_SOC': ('犯罪主體', 'Subject of Crime'),
|
|
|
|
| 73 |
'LEO_ROH': Fore.BLACK + Back.BLUE,
|
| 74 |
'LEO_ATP': Fore.WHITE + Back.BLACK,
|
| 75 |
}
|
| 76 |
+
</code>
|
| 77 |
</pre>
|
| 78 |
|
| 79 |
為了要將本模型標註出來的結果以更明顯的方式識別,可以參考以下的程式碼,將本大型語言模型生成的標註結果以及所標註的標籤,同時送入以下的函數,就可以將結果以 colorama 的方式著色標註。
|
|
|
|
| 124 |
</code>
|
| 125 |
</pre>
|
| 126 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 127 |
### Direct Use
|
| 128 |
|
| 129 |
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
|
|
|
|
| 139 |
### Out-of-Scope Use
|
| 140 |
|
| 141 |
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
|
| 142 |
+
本模型目前僅能標示依據中���民國刑法規定的「詐欺罪」所擬撰(或是語言模型生成)之「犯罪事實」中的構成要件要素,若要具備標註其餘各種不同的犯罪構成要件要素之標註能力,則是後續可以發展以及擴增語料庫的方向。
|
| 143 |
|
| 144 |
[More Information Needed]
|
| 145 |
|