Update README.md
Browse files
README.md
CHANGED
@@ -17,12 +17,12 @@ This is a Lora trained on llama2 7B Chat, with its dataset consisting of a large
|
|
17 |
|
18 |
## Training Dataset
|
19 |
|
20 |
-
微調用的資料集由少量個人撰寫與以此為基礎生成的大量AI生成對話內容組成,使用alpaca-format,約
|
21 |
|
22 |
The fine-tuning dataset used consists of a small number of personally written conversations and a large amount of AI-generated dialogue content based on these, utilizing the alpaca-format. It comprises approximately 7,000 instructions in total and has a size of 9MB.
|
23 |
|
24 |
## Training
|
25 |
-
使用text-generation-webui中的Training工具,在google colab上調用V100,以4bit模式讀取Llama2 7B
|
26 |
|
27 |
Using the Training tool in the text-generation-webui, calling a V100 on Google Colab, and reading Llama2 7B in 4-bit mode, training was performed with default parameters. The total training time took approximately 2 hours
|
28 |
|
@@ -51,4 +51,8 @@ Prompt需要撰寫角色人格、與個性;如有必要則需要另寫對話
|
|
51 |
## License
|
52 |
這個lora以CC BY-SA 4.0作為分享
|
53 |
|
54 |
-
整體使用請遵照Meta的社群許可,不要將其作於非法用途或生成不適當的內容
|
|
|
|
|
|
|
|
|
|
17 |
|
18 |
## Training Dataset
|
19 |
|
20 |
+
微調用的資料集由少量個人撰寫與以此為基礎生成的大量AI生成對話內容組成,使用alpaca-format,約9千條instruction、共12.6MB的大小
|
21 |
|
22 |
The fine-tuning dataset used consists of a small number of personally written conversations and a large amount of AI-generated dialogue content based on these, utilizing the alpaca-format. It comprises approximately 7,000 instructions in total and has a size of 9MB.
|
23 |
|
24 |
## Training
|
25 |
+
使用text-generation-webui中的Training工具,在google colab上調用V100,以4bit模式讀取Llama2 7B後以Lora Rank64,Lora Alpha128,Epochs5,其餘用預設參數訓練,訓練總計花費約5小時
|
26 |
|
27 |
Using the Training tool in the text-generation-webui, calling a V100 on Google Colab, and reading Llama2 7B in 4-bit mode, training was performed with default parameters. The total training time took approximately 2 hours
|
28 |
|
|
|
51 |
## License
|
52 |
這個lora以CC BY-SA 4.0作為分享
|
53 |
|
54 |
+
整體使用請遵照Meta的社群許可,不要將其作於非法用途或生成不適當的內容
|
55 |
+
|
56 |
+
|
57 |
+
##Update note
|
58 |
+
2023/09/13以現有的資料集改為用更高的Lora Rank,Lora alpha與epoch再次訓練
|