micro benchmark, numa node extraction is not compatible with Grace CPU (GH/GB GPUs)

[Fizzbb]

the way extract numa node is incompatible with Grace CPU
https://github.com/bytedance/ByteMLPerf/blob/main/byte_micro_perf/launch.py#L47

lscpu on GH200. To decide how many numa nodes available. Please count the node has cpu cores. the example below, should be 1, instead of 9.

NUMA:
NUMA node(s): 9
NUMA node0 CPU(s): 0-71
NUMA node1 CPU(s):
NUMA node2 CPU(s):
NUMA node3 CPU(s):
NUMA node4 CPU(s):
NUMA node5 CPU(s):
NUMA node6 CPU(s):
NUMA node7 CPU(s):
NUMA node8 CPU(s):