jklj077
commented
3 months ago Hi,
It is generally considered unreliable to ask a large language model its knowledge cutoff, as there is no way for the models to learn in pretraining and the date is injected in post-training.
The training data (including pretraining and posttraining) of the model contains materials as late as the end of 2023.
In addition, Qwen models are causal language models and do not make use of knowledge base tenchinques.
当前看21年的，很多事情多不知道。谢谢开源这么棒的模型。