import deep500 as d5
from deep500.frameworks import tensorflow as d5tf
# Uses the ONNX Test Parser to run a test
if __name__ == '__main__':
if len(sys.argv) > 2:
print('USAGE: onnxtest.py [TESTNAME]')
print('If TESTNAME is not specified, chooses and runs a random test')
sys.exit(1)
parser = d5.OnnxTestParser()
if len(sys.argv) == 1:
testname = random.choice(parser.all_tests['Node'])
print('Chose test "%s"' % testname)
else:
testname = sys.argv[1]
test = parser.get_test(testname)
if test is None:
print('ERROR: Test "%s" not found' % testname)
sys.exit(2)
model = d5.parser.load_and_parse_model(test.model)
executor = d5tf.from_model(model)
d5.test_executor_inference(executor,
inputs=test.data_sets[0].inputs,
reference_outputs=test.data_sets[0].outputs,
metrics=[d5.L2Error()])
# Create Pytorch "model" that only receives two variables
def model(A, B):
if test.a_trans:
Amul = A.transpose(0, 1)
else:
Amul = A
if test.b_trans:
Bmul = B.transpose(0, 1)
else:
Bmul = B
return torch.matmul(Amul, Bmul)
# Run Deep500 test
l2err, maxerr, times = \
d5pt.test_nativeop_forward(model, [var_A, var_B], [Amult @ Bmult],
metrics=[d5.L2Error(), d5.MaxError(),
d5.WallclockTime(RUNS*AVG_OVER, AVG_OVER)])
except Exception as ex:
print('Exception:', ex)
l2err = -1.0
maxerr = -1.0
times = [-1.0]
with open('0_d5pt_gemm_deepbench.log', 'a') as fp:
fp.writelines([
'{test.m},{test.n},{test.k},'
'{test.a_trans},{test.b_trans},'
'{time:.15f},{l2err},{maxerr}\n'.format(test=test,
time=time,
l2err=l2err,
maxerr=maxerr)
def DefaultOptimizerMetrics():
return [d5.L2Error(), d5.WallclockTime()]