PaddleSeg
PaddleSeg copied to clipboard
PaddlePaddle
[General Issue]使用paddle inference推理时报错（python）
在进行推理这一步时，输入 python deploy/python/infer.py --emo --enable_auto_tune=True --use_trt=True 报错 实测不使用tensorrt 或者改为cpu都不报错。 以下为报错信息
/home/speed/Segmentation/paddleseg/paddleseg/models/losses/decoupledsegnet_relax_boundary_loss.py:19: DeprecationWarning: Please use shift from the scipy.ndimage namespace, the scipy.ndimage.interpolation namespace is deprecated.
from scipy.ndimage.interpolation import shift
/home/speed/Segmentation/paddleseg/paddleseg/transforms/functional.py:18: DeprecationWarning: Please use distance_transform_edt from the scipy.ndimage namespace, the scipy.ndimage.morphology namespace is deprecated.
from scipy.ndimage.morphology import distance_transform_edt
2022-08-08 11:01:30 [INFO] Auto tune the dynamic shape for GPU TRT.
I0808 11:01:30.041559 4054828 analysis_config.cc:1044] In CollectShapeInfo mode, we will disable optimizations and collect the shape information of all intermediate tensors in the compute graph and calculate the min_shape, max_shape and opt_shape.
W0808 11:01:30.776191 4054828 analysis_predictor.cc:1118] The one-time configuration of analysis predictor failed, which may be due to native predictor called first and its configurations taken effect.
I0808 11:01:30.794692 4054828 analysis_predictor.cc:985] ir_optim is turned off, no IR pass will be executed
--- Running analysis [ir_graph_build_pass]
--- Running analysis [ir_graph_clean_pass]
--- Running analysis [ir_analysis_pass]
--- Running analysis [ir_params_sync_among_devices_pass]
I0808 11:01:30.828686 4054828 ir_params_sync_among_devices_pass.cc:100] Sync params from CPU to GPU
--- Running analysis [adjust_cudnn_workspace_size_pass]
--- Running analysis [inference_op_replace_pass]
--- Running analysis [ir_graph_to_program_pass]
I0808 11:01:30.910813 4054828 analysis_predictor.cc:1035] ======= optimize end =======
I0808 11:01:30.915236 4054828 naive_executor.cc:102] --- skip [feed], feed -> x
I0808 11:01:30.917994 4054828 naive_executor.cc:102] --- skip [argmax_0.tmp_0], fetch -> fetch
W0808 11:01:30.993405 4054828 gpu_resources.cc:61] Please NOTE: device: 0, GPU Compute Capability: 8.6, Driver API Version: 11.7, Runtime API Version: 11.2
W0808 11:01:30.994926 4054828 gpu_resources.cc:91] device: 0, cuDNN Version: 8.1.
2022-08-08 11:01:33 [INFO] Auto tune success.
2022-08-08 11:01:33 [INFO] Use GPU 2022-08-08 11:01:33 [INFO] Use TRT 2022-08-08 11:01:33 [INFO] Use auto tuned dynamic shape I0808 11:01:33.587178 4054828 analysis_predictor.cc:881] TensorRT subgraph engine is enabled --- Running analysis [ir_graph_build_pass] --- Running analysis [ir_graph_clean_pass] --- Running analysis [ir_analysis_pass] 2022-08-08 11:01:33 [INFO] (InvalidArgument) Pass preln_embedding_eltwise_layernorm_fuse_pass has not been registered. [Hint: Expected Has(pass_type) == true, but received Has(pass_type):0 != true:1.] (at /paddle/paddle/fluid/framework/ir/pass.h:242)
2022-08-08 11:01:33 [INFO] If the above error is '(InvalidArgument) some trt inputs dynamic shape info not set, ..., Expected all_dynamic_shape_set == true, ...', please set --enable_auto_tune=True to use auto_tune.
欢迎您反馈PaddleSeg使用问题，辛苦您提供以下信息，方便我们快速定位和解决问题：
- PaddleSeg版本：2.6.0
- PaddlePaddle版本：2.3.1
- 操作系统信息：linux
- Python版本号：3.9
- CUDA/cuDNN版本：11.6/8.4.1
- tensorRT版本：8.4.2.4
看起来是输入了动态shape导致的报错，使用的是paddleseg的哪个模型呢，如果是自己的模型，可以检查下是否有xx.shape 的操作，改成paddle.shape(xx)
看起来是输入了动态shape导致的报错，使用的是paddleseg的哪个模型呢，如果是自己的模型，可以检查下是否有xx.shape 的操作，改成paddle.shape(xx)
用动态和静态的shape都试了 都有这个错误。而且不启用trt错误消失
对 你需要安装编译支持trt的版本。
刚刚验证了一下 paddle安装没有问题，如果报错Pass trt_squeeze2_matmul_fuse_pass has not been registered.可能是paddle版本问题。但是我报的错是Pass preln_embedding_eltwise_layernorm_fuse_pass has not been registered.
这个问题是没有正确安装联编trt的paddle whl包。具体参考安装文档：https://github.com/PaddlePaddle/PaddleSeg/blob/release/2.6/docs/deployment/inference/python_inference_cn.md
