0.9.0版本的label构建方式

[zhangzhili1112]

新版本的单论对话label构建前后都加<|endoftext|>，0.8.3版本单论是只在后面加<|endoftext|>

Reminder

• I have read the README and searched the existing issues.

System Info

training example:
input_ids:
[151331, 151333, 151335, 198, 98406, 107794, 99941, 106867, 113255, 3837, 98406, 99569, 2078, 38, 2127, 55, 1773, 103318, 98323, 99833, 99770, 99546, 106867, 5373, 103461, 5373, 101183, 101156, 99766, 98622, 3837, 119257, 104714, 100315, 5373, 98516, 100033, 5373, 101367, 107175, 103461, 3837, 103078, 114702, 99089, 110932, 100774, 1773, 151336, 198, 6023, 151337, 198, 9703, 0, 358, 1079, 5867, 606, 37953, 458, 15223, 17821, 7881, 553, 5867, 3094, 3417, 13, 2585, 646, 358, 7789, 498, 3351, 30, 151329]
inputs:
[gMASK] <|system|>
你是一位智能编程助手，你叫CodeGeeX。你会为用户回答关于编程、代码、计算机方面的任何问题，并提供格式规范、可以执行、准确安全的代码，并在必要时提供详细的解释。 <|user|>
hi <|assistant|>
Hello! I am {{name}}, an AI assistant developed by {{author}}. How can I assist you today? <|endoftext|>
label_ids:
[151329, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, 9703, 0, 358, 1079, 5867, 606, 37953, 458, 15223, 17821, 7881, 553, 5867, 3094, 3417, 13, 2585, 646, 358, 7789, 498, 3351, 30, 151329]
labels:
<|endoftext|>Hello! I am {{name}}, an AI assistant developed by {{author}}. How can I assist you today? <|endoftext|>

Reproduction

无错误信息，只是label的构建方式改变想确认一下

Expected behavior

No response

Others

No response