INNER CODE UNIT · Python
get_nccl_version
awslabs/awsome-distributed-ai · validation/efa-versions.py:31
def get_nccl_version(container=[]):
try:
version = subprocess.check_output(container + ['locate', 'nccl'])
version = version.decode('utf-8')
print(version)
version = re.search(r'/usr/local/cuda-12.8/lib/libnccl.so.(\d+\.\d+\.\d+)', version).group(1)
return version
except Exception as e:
print(f'Error: {e}')
return None
def get_aws_ofi_nccl_version(container=[]):
try:
lib_path = subprocess.check_output(container + ['cat', '/etc/ld.so.conf.d/100_ofinccl.conf']).decode().strip()
version = subprocess.check_output(container + ['strings', f'{lib_path}/libnccl-net.so'])
version = version.decode('utf-8')
version = re.search(r'NET/OFI Initializing aws-ofi-nccl (\d+\.\d+\.\d+)', version).group(1)
return version