Created by: xiegegege

使用paddle develop分支，运行Bert 的xnli任务，对于fp16=false时会有tensor类型的报错，fp16=true时运行正常。 I0530 03:41:07.354573 44814 parallel_executor.cc:339] The Executor would execute the graph by ParallelGraph Execution which can get better performance,you can force it off by env FLAGS_enable_parallel_graph=0 Traceback (most recent call last): File "run_classifier.py", line 383, in <module> main(args) File "run_classifier.py", line 288, in main outputs = train_exe.run(fetch_list=fetch_list) File "/opt/_internal/cpython-3.6.0/lib/python3.6/site-packages/paddle/fluid/parallel_executor.py", line 286, in run return_numpy=return_numpy) File "/opt/_internal/cpython-3.6.0/lib/python3.6/site-packages/paddle/fluid/executor.py", line 651, in run return_numpy=return_numpy) File "/opt/_internal/cpython-3.6.0/lib/python3.6/site-packages/paddle/fluid/executor.py", line 513, in _run_parallel exe.run(fetch_var_names, fetch_var_name) paddle.fluid.core.EnforceNotMet: Invoke operator adam error. Python Callstacks: File "/opt/_internal/cpython-3.6.0/lib/python3.6/site-packages/paddle/fluid/framework.py", line 1703, in append_op attrs=kwargs.get("attrs", None)) File "/opt/_internal/cpython-3.6.0/lib/python3.6/site-packages/paddle/fluid/optimizer.py", line 1291, in _append_optimize_op stop_gradient=True) File "/opt/_internal/cpython-3.6.0/lib/python3.6/site-packages/paddle/fluid/optimizer.py", line 297, in _create_optimization_pass param_and_grad) File "/opt/_internal/cpython-3.6.0/lib/python3.6/site-packages/paddle/fluid/optimizer.py", line 442, in apply_gradients optimize_ops = self._create_optimization_pass(params_grads) File "/ssd1/xiege/model_5.29/models/PaddleNLP/language_representations_kit/LARK_test/LARK/BERT/optimization.py", line 107, in optimization optimizer.apply_gradients(master_param_grads) File "run_classifier.py", line 187, in main loss_scaling=args.loss_scaling) File "run_classifier.py", line 383, in <module> main(args) C++ Callstacks: Tensor holds the wrong type, it holds ::paddle::platform::float16, but desires to be float at [/ssd1/xiege/paddle_wheel/Paddle/paddle/fluid/framework/tensor_impl.h:30]