Thank you for your research contributions and for open-sourcing the code in this field.
I have a little question when I'm trying to use the modeling_liger_qwen3.py.
Considering the structure of Qwen3 Model, do we need to add the q_norm, k_norm on the q,k before we send it into the gla_func? If so, when should we turn the k into the gate? Before the k_norm or after it?
Thank you for your research contributions and for open-sourcing the code in this field.
I have a little question when I'm trying to use the modeling_liger_qwen3.py.
Considering the structure of Qwen3 Model, do we need to add the q_norm, k_norm on the q,k before we send it into the gla_func? If so, when should we turn the k into the gate? Before the k_norm or after it?