|
|
--- |
|
|
license: mit |
|
|
--- |
|
|
|
|
|
Text data related to the subject of Mechanical Engineering. |
|
|
|
|
|
----------------------------------------------------------------------------------------------------------- |
|
|
|
|
|
Compressed file "*.tar.gz" is a collection of json files, which are filtered from Fineweb dataset. |
|
|
|
|
|
The classification was done by fasttext (https://github.com/ZJLab-ZBZX/Data-Processing-Toolkit-for-LLMs/tree/main/Subject_Classifier). |
|
|
|
|
|
And we randomly selected 12.8% classified data, and checked them by Qwen-2-7B LLM through customized prompt. |
|
|
|
|
|
The accuracy of the classified data is about 94.9926%. The total data volume is 4.9B token. |
|
|
|
|
|
----------------------------------------------------------------------------------------------------------- |
|
|
|
|
|
File "wiki_Mechanical_yes.jsonl" is a json file filtered from Wikipedia dataset. |
|
|
|
|
|
File "sft-Mechanical-en.jsonl" is a json file for supervised fine-tuning (SFT). |
