Created by: dingsiyu

为使您的问题得到快速解决，在建立Issues前，请您先通过如下方式搜索是否有相似问题:【搜索issue关键字】【使用labels筛选】【官方文档】

如果您没有查询到相似问题，为快速解决您的提问，建立issue时请提供如下细节信息：

• 标题：简洁、精准概括您的问题，例如“Insufficient Memory xxx" ”

• 版本、环境信息：    1）PaddlePaddle版本：1.8    2）CPU：    3）GPU：v100， cuda 10, cudnn 7.2    4）系统环境：python2 注：您可以通过执行summary_env.py获取以上信息。

• 训练信息    1）多机多卡

• 复现信息：如为报错，请给出复现环境、复现步骤

• 问题描述：请详细描述您的问题，同步贴出报错信息、日志、可复现的代码片段

Python Call Stacks (More useful to users):

File "/root/paddlejob/workspace/env_run/one_stage_adv_atten_recompute/python/lib/python2.7/site-packages/paddle/fluid/framework.py", line 2610, in append_op attrs=kwargs.get("attrs", None)) File "/root/paddlejob/workspace/env_run/one_stage_adv_atten_recompute/python/lib/python2.7/site-packages/paddle/fluid/layer_helper.py", line 43, in append_op return self.main_program.current_block().append_op(*args, **kwargs) File "/root/paddlejob/workspace/env_run/one_stage_adv_atten_recompute/python/lib/python2.7/site-packages/paddle/fluid/layers/nn.py", line 1719, in fc "y_num_col_dims": 1}) File "/root/paddlejob/workspace/env_run/one_stage_adv_atten_recompute/model/transformer_encoder.py", line 84, in __compute_qkv bias_attr=name + '_query_fc.b_0') File "/root/paddlejob/workspace/env_run/one_stage_adv_atten_recompute/model/transformer_encoder.py", line 169, in multi_head_attention q, k, v = __compute_qkv(queries, keys, values, n_head, d_key, d_value) File "/root/paddlejob/workspace/env_run/one_stage_adv_atten_recompute/model/transformer_encoder.py", line 320, in encoder_layer name=name + '_multi_head_att') File "/root/paddlejob/workspace/env_run/one_stage_adv_atten_recompute/model/transformer_encoder.py", line 395, in encoder name=name + 'layer' + str(i)) File "/root/paddlejob/workspace/env_run/one_stage_adv_atten_recompute/model/ernie.py", line 189, in _build_model name='encoder') File "/root/paddlejob/workspace/env_run/one_stage_adv_atten_recompute/model/adversarial.py", line 85, in vat_loss ernie._build_model(None, None, None, None, None, emb_out=embedded, noise=d, ind=index) File "./train.py", line 159, in create_model adv_loss = vat_loss(ernie, logits_vars, task_fc_fn, args.epsilon, args.small_constant_for_finite_diff, batch_size=args.batch_size) File "./train.py", line 340, in train pyreader_name='train_reader', ernie_config=ernie_config, task_group=task_group) File "./train.py", line 500, in train(args)

Error Message Summary:

NotFoundError: No Input(DOut) found for mul operator. [Hint: Expected ctx->HasInput("DOut") == true, but received ctx->HasInput("DOut"):0 != true:1.] at (/paddle/paddle/fluid/operators/mul_op.cc:258) [operator < mul_grad_grad > error]