Skip to content

Commit

Permalink
Update README.md
Browse files Browse the repository at this point in the history
  • Loading branch information
@ceodspspectrum
ceodspspectrum authored Aug 22, 2023
1 parent b437813 commit efb9915
Showing 1 changed file with 13 additions and 13 deletions.
26 changes: 13 additions & 13 deletions README.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -85,19 +85,19 @@ from scipy.io import loadmat


def dist_nmfk_1d_nnsvd_init_wtsi(nGPUs=None,nccl_comm=None, topology=None ):
args = parse() args.use_gpu = True args.use_gpu = False
args.gpu_partition = 'row' args.IMAX = 40960 #1024 #2048 #512 #256
args.JMAX = 40960 #1024 #2048 #512 #256 args.A_Is_Larage = True args.A_Is_Larage = False
args.init = 'rand' rank = comm.rank size = comm.size if args.use_gpu: if nGPUs is None: nGPUs = size else: nGPUs = size
args.fpath = '../data/'
args.fname,args.ftype = 'US_11260_15056', 'npy'
if args.use_gpu: try: partition_type = args.gpu_partition.lower() if partition_type in ['row', 'row1', '1row' ,'row_1d', '1d_row']: args.gpu_partition = 'row_1d' elif partition_type in ['col', 'col1', '1col' ,'col_1d', '1d_col', 'column', 'column_1d', '1d_column']: args.gpu_partition = 'col_1d' elif partition_type in ['auto', 'optim', 'optimal']: args.gpu_partition = 'auto' else: raise Exception(f"[!!] GPU grid partition {args.gpu_partition} is not supported. Try 'auto', 'column', 'row'") except: args.gpu_partition = 'auto' if args.gpu_partition in ['row_1d']:
p_r, p_c = 1, nGPUs # Row METHOD = 'Row_1d' elif args.gpu_partition in ['col_1d']: p_r, p_c = nGPUs, 1 # Col METHOD = 'Col_1d' else: #AUTO if rank == 0: print(f'[!!][AUTO GPU PARTITION] Trying to find optimal aprtition...') try: NN= args.fname.split('_') if int(NN[1]) < int(NN[2]): p_r, p_c = 1, nGPUs # Row METHOD = 'Row_1d' else: p_r, p_c = nGPUs, 1 # Col METHOD = 'Col_1d' if rank == 0: print(f"[!!][AUTO GPU PARTITION] {METHOD} GPU partition selected") except:
print(f'[!!] USING DEFAULT METHOD') p_r, p_c, METHOD = nGPUs, 1, 'Col_1d' # Col #p_r, p_c, METHOD = 1, nGPUs, 'Row_1d' # Row
comms = MPI_comm(comm, p_r, p_c) comm1 = comms.comm args.size, args.rank, args.comm, args.p_r, args.p_c = size, rank, comms, p_r, p_c args.row_comm, args.col_comm, args.comm1 = comms.cart_1d_row(), comms.cart_1d_column(), comm1 rank = comms.rank #if rank == 0: print(f"Running {METHOD}") args.start_k = 8
args.end_k = 128 #args.k = 4 args.step = 8 args.sill_thr = 0.6 args.itr = 1000 args.verbose = True args.norm = 'fro' args.method = 'mu' args.precision = np.float32 args.checkpoint = False A_ij = data_read(args).read().astype(args.precision)
args.results_path = '../results/' m,n = A_ij.shape J = min(args.JMAX, n) args.nGPUs = nGPUs args.grid_m = m args.grid_m = n args.grid_m = J
nopt = PyNMFk(A_ij, factors=None, params=args).fit()
args = parse() args.use_gpu = True args.use_gpu = False
args.gpu_partition = 'row' args.IMAX = 40960 #1024 #2048 #512 #256
args.JMAX = 40960 #1024 #2048 #512 #256 args.A_Is_Larage = True args.A_Is_Larage = False
args.init = 'rand' rank = comm.rank size = comm.size if args.use_gpu: if nGPUs is None: nGPUs = size else: nGPUs = size
args.fpath = '../data/'
args.fname,args.ftype = 'US_11260_15056', 'npy'
if args.use_gpu: try: partition_type = args.gpu_partition.lower() if partition_type in ['row', 'row1', '1row' ,'row_1d', '1d_row']: args.gpu_partition = 'row_1d' elif partition_type in ['col', 'col1', '1col' ,'col_1d', '1d_col', 'column', 'column_1d', '1d_column']: args.gpu_partition = 'col_1d' elif partition_type in ['auto', 'optim', 'optimal']: args.gpu_partition = 'auto' else: raise Exception(f"[!!] GPU grid partition {args.gpu_partition} is not supported. Try 'auto', 'column', 'row'") except: args.gpu_partition = 'auto' if args.gpu_partition in ['row_1d']:
p_r, p_c = 1, nGPUs # Row METHOD = 'Row_1d' elif args.gpu_partition in ['col_1d']: p_r, p_c = nGPUs, 1 # Col METHOD = 'Col_1d' else: #AUTO if rank == 0: print(f'[!!][AUTO GPU PARTITION] Trying to find optimal aprtition...') try: NN= args.fname.split('_') if int(NN[1]) < int(NN[2]): p_r, p_c = 1, nGPUs # Row METHOD = 'Row_1d' else: p_r, p_c = nGPUs, 1 # Col METHOD = 'Col_1d' if rank == 0: print(f"[!!][AUTO GPU PARTITION] {METHOD} GPU partition selected") except:
print(f'[!!] USING DEFAULT METHOD') p_r, p_c, METHOD = nGPUs, 1, 'Col_1d' # Col #p_r, p_c, METHOD = 1, nGPUs, 'Row_1d' # Row
comms = MPI_comm(comm, p_r, p_c) comm1 = comms.comm args.size, args.rank, args.comm, args.p_r, args.p_c = size, rank, comms, p_r, p_c args.row_comm, args.col_comm, args.comm1 = comms.cart_1d_row(), comms.cart_1d_column(), comm1 rank = comms.rank #if rank == 0: print(f"Running {METHOD}") args.start_k = 8
args.end_k = 128 #args.k = 4 args.step = 8 args.sill_thr = 0.6 args.itr = 1000 args.verbose = True args.norm = 'fro' args.method = 'mu' args.precision = np.float32 args.checkpoint = False A_ij = data_read(args).read().astype(args.precision)
args.results_path = '../results/' m,n = A_ij.shape J = min(args.JMAX, n) args.nGPUs = nGPUs args.grid_m = m args.grid_m = n args.grid_m = J
nopt = PyNMFk(A_ij, factors=None, params=args).fit()
from mpi4py import MPI
comm = MPI.COMM_WORLD
size = comm.Get_size()
Expand Down

0 comments on commit efb9915

Please sign in to comment.