Datasets:
The performance of Qwen models degrades after training on COIG subset?
#25
by
jiangxg
- opened
The performance of Qwen shown in Table 5 is about 3-5% lower than that claimed by Qwen group.
See https://github.com/QwenLM/Qwen for details
Is it reasonable or caused by using different testing scripts?