--- license: mit datasets: - kxdw2580/catgirl-dataset language: - zho - eng - fra - spa - por - deu - ita - rus - jpn - kor - vie - tha - ara base_model: - Qwen/Qwen2.5-0.5B-Instruct --- # kxdw2580/Qwen2.5-0.5B-Catgirl-test0426 This model is a test model, designed for phased lightweight testing of the dataset. The test dataset has fixed this issue: - Model's outputs "~" causing rendering errors. After testing, the objectives of the dataset fixes have been achieved. ## Other As a 0.5b model, its performance is very poor, especially in the missing English part of the dataset. We do not recommend using this model unless there is a specific need. We are working hard to improve the training results on smaller models, but it is obviously unlikely for the 0.5b model. Specific training results can be seen at [swanlab](https://swanlab.cn/@shadow01a/qwen-catgirl/runs/q58lh8yf6itgoamcoq4q8/chart) Additionally, I have observed that with models of this size, a smaller training loss does not always indicate better model performance, and sometimes even leads to a decline in performance. [This swanlab record](https://swanlab.cn/@shadow01a/qwen-catgirl/runs/d29hi6y9d7g772ib5vbtx/chart) is the result of further training for this model. After testing, I found that its performance is even worse than the original model. This model has not been publicly released and has been deleted. I would be very happy to communicate if you wish to!