Xiangyu Li
Xiangyu Li
想问一下跑QWEN3_MOE_VL的话有官方的megatron实现吗
> mbridge 卸载了重新安装，但会报新的错 TypeError: Qwen3VLSelfAttention.forward() got an unexpected keyword argument 'yarn_mscale' > > megatron还是0.15版本应该，镜像中是装在/opt/megatron-lm目录下，得用RUN rm -rf /opt/megatron /opt/megatron-lm && \ pip uninstall -y megatron-core megatron-lm || true && \ pip...
> good job! it might be better if we add some description in the script file? Thanks for the reminder! I've added the corresponding comments to the script to clarify...