Add support for grouped-query attention

[yasuhisa-nakashima]

Add support for grouped-query attention for Llama 2 70B and Code Llama 34B compatibility.

References:

• https://github.com/NVIDIA/Megatron-LM/blob/main/megatron/model/transformer.py
• https://github.com/NVIDIA/Megatron-LM/blob/main/megatron/core/transformer/attention.py