如text_encoders是Qwen3-VL就完美了
#5
by
hao3760
- opened
4B模型估计就够用了,多省资源呀😄
Thank you for the suggestion! The selection of the text encoder was finalized quite early in our development phase, while Qwen3-VL was released relatively late. However, we will definitely consider adopting a smaller text encoder in future iterations to optimize resource usage.