As Megatron-Deepspeed has merged lots of commits from Megatron-LM recently, the patch for Megatron-Deepspeed is incompatible with the latest version. For example, the files under the path mpu/ seem to be different. Shall I use the patch for Megatron-LM directly?
As Megatron-Deepspeed has merged lots of commits from Megatron-LM recently, the patch for Megatron-Deepspeed is incompatible with the latest version. For example, the files under the path
mpu/seem to be different. Shall I use the patch for Megatron-LM directly?