Use softlink instead of copying the llama2-70b model to save disk space

[anandhu-eng]

ref:

mlperf-automations/script/get-ml-model-llama2/run-nvidia.sh

Line 9 in 619734d

cp -r ${LLAMA2_CHECKPOINT_PATH}/* ${MLC_NVIDIA_MLPERF_SCRATCH_PATH}/models/Llama2/Llama-2-70b-chat-hf