required=False,
default=[100, 128, 12800, 25, 100, 2500, 5, 25, 125])
parser.add_argument('--vectors',
required=False,
type=int,
help='number of vectors',
default=300)
parser.add_argument('--mtx',
required=False,
help='path to mtx file',
nargs="+",
default='none')
args = parser.parse_args()
xclbin_opt = gemx.parse_cfg(args.cfg)
if args.engine == 'spmv':
gemx.createUSPMVHandle(args, xclbin_opt)
else:
gemx.createFCNHandle(args, xclbin_opt)
A_buf = []
B_buf = []
C_buf = []
bias_buf = []
number_runs = args.vectors
stage_size = 1
if args.mtx == 'none':
num_matrix = len(args.matrix) / 3
else:
num_matrix = len(args.mtx)
if args.engine == 'spmv':
min_row = int(xclbin_opt["GEMX_uspmvInterleaves"]) * int(
xclbin_opt["GEMX_ddrWidth"])
help='file path to GEMX host code shared library')
parser.add_argument('--engine',
default='fcn',
choices=['fcn', 'uspmv'],
help='choose fcn, uspmv engine')
parser.add_argument('--train',
default=False,
help='set to True if retrain the model')
args = parser.parse_args()
xclbin_prop = gemx.parse_cfg(args.cfg)
#load xclbin
if args.engine == 'fcn':
gemx.createFCNHandle(args, xclbin_prop)
else:
gemx.createUSPMVHandle(args, xclbin_prop)
(x_train, y_train), (x_test, y_test) = reuters.load_data(num_words=1000,
test_split=0.2)
tokenizer = Tokenizer(num_words=1000)
num_classes = np.max(y_train) + 1
model = create_keras_model(num_classes)
model.load_weights(args.model)
if args.train:
train(model, x_train, y_train, x_test, y_test)
x_test = tokenizer.sequences_to_matrix(x_test, mode='binary')
cpu_out = mlp_common.predict_cpu(model, x_test)