in optimum/amd/cli.py [0:0]
def get_amd_rocm_env():
"""
Returns a dictionary of environment variables that are optimized for AMD's ROCm platform.
The target environment variables are:
- `ROCR_VISIBLE_DEVICES`: The list of devices to use (maximizing the average bandwidth between them).
"""
from .topology_utils import extract_max_avg_bandwidth_cluster, get_bandwidth_matrix
# extract the number of devices to use
if "--nproc_per_node" in sys.argv:
# torchrun style
nproc_per_node_index = sys.argv.index("--nproc_per_node")
num_devices = int(sys.argv[nproc_per_node_index + 1])
elif "--ngpus" in sys.argv:
# accelerate/deepspeed style
ngpus_index = sys.argv.index("--ngpus")
num_devices = int(sys.argv[ngpus_index + 1])
else:
# early exit if we can't find the number of devices
return {}
bandwidth_matrix = get_bandwidth_matrix()
max_avg_bandwidth_cluster, max_avg_bandwidth = extract_max_avg_bandwidth_cluster(bandwidth_matrix, num_devices)
# lowest level isolation env var on AMD GPUs
ROCR_VISIBLE_DEVICES = ",".join(list(map(str, max_avg_bandwidth_cluster)))
logger.info("AMD ROCm environment variables:")
logger.info(f"- ROCR_VISIBLE_DEVICES: {ROCR_VISIBLE_DEVICES}")
return {"ROCR_VISIBLE_DEVICES": ROCR_VISIBLE_DEVICES}