关于预测阶段出out of memory的问题求助： #9990
Description
用的paddle.infer() 方法，每输入2000条元素执行一次这个方法，主要部分代码如下：
input_data = []
res = []
ids = []
count = 0
out_file = open("./data.infer.user", "w");
with open("./data.user") as file:
for line in file:
data = line.strip().split("\t")
if len(data) == 2:
input_data.append([get_sparse_index_data(data[1]), get_sparse_index_data([])])
ids.append(data[0])
elif len(data) == 3:
input_data.append([get_sparse_index_data(data[1]), get_sparse_index_data(data[2])])
ids.append(data[0])
else:
continue
count += 1
if count == 2000:
res = paddle.infer(output_layer=user_features, parameters=parameters, input=input_data, feeding=feeding)
for index in range(0, count):
list = res[index]
output = ''
for i in range(0, 32) :
output += str(round(list[i], 4)) + " "
out_file.write(ids[index] + "\t" + output.strip() + "\n")
input_data = []
ids = []
count = 0
每次运行一段时间就会报出 Cuda Error: out of memory
请问这个是paddle的问题，还是python的问题?