def test_edge_case():
context.set_context(mode=context.GRAPH_MODE)
inputs = Tensor(np.ones([32, 48]).astype(np.float32))
net = Net()
model = Model(net)
with pytest.raises(RuntimeError):
model.infer_predict_layout(inputs)
context.set_auto_parallel_context(parallel_mode="semi_auto_parallel")
with pytest.raises(RuntimeError):
model.infer_predict_layout(inputs)
context.set_auto_parallel_context(full_batch=True,
enable_parallel_optimizer=True)
with pytest.raises(RuntimeError):
model.predict(inputs)
def test_distribute_predict_auto_parallel():
context.set_context(mode=context.GRAPH_MODE, save_graphs=True)
context.set_auto_parallel_context(parallel_mode="auto_parallel", device_num=8, full_batch=True)
inputs = Tensor(np.ones([32, 64, 128]).astype(np.float32))
net = Net()
model = Model(net)
predict_map = model.infer_predict_layout(inputs)
output = model.predict(inputs)
context.reset_auto_parallel_context()
return predict_map, output
def test_inference():
"""distributed inference after distributed training"""
context.set_context(mode=context.GRAPH_MODE)
init(backend_name="hccl")
context.set_auto_parallel_context(full_batch=True, parallel_mode="semi_auto_parallel",
strategy_ckpt_load_file="./train_strategy.ckpt", device_num=8)
predict_data = create_predict_data()
network = Net(matmul_size=(96, 16))
model = Model(network)
predict_layout = model.infer_predict_layout(Tensor(predict_data))
ckpt_file_list = create_ckpt_file_list()
load_distributed_checkpoint(network, ckpt_file_list, predict_layout)
predict_result = model.predict(predict_data)
print(predict_result)