软件环境

paddle-bfloat             0.1.7
paddle2onnx               1.2.6
paddlefsl                 1.1.0
paddlemix                 0.1.0                
paddlenlp                 3.0.0b0.post20240728
paddlepaddle-gpu          2.6.1.post112

重复问题

• 我已搜索了现有的问题

错误描述

按照https://github.com/PaddlePaddle/PaddleNLP/blob/develop/csrc/README.md
教程安装编译 Cuda 算子报错：no type named ‘hash_policy’
'''
!!                                                                                                                                                                            
  self.initialize_options()                                                                                                                                                   
[2024-07-30 22:04:02,198] [    INFO] dist.py:985 - running bdist_egg                                                                                                          
[2024-07-30 22:04:03,177] [    INFO] dist.py:985 - running egg_info                                                                                                           
[2024-07-30 22:04:03,618] [    INFO] egg_info.py:668 - writing paddlenlp_ops.egg-info/PKG-INFO                                                                                
[2024-07-30 22:04:03,623] [    INFO] egg_info.py:292 - writing dependency_links to paddlenlp_ops.egg-info/dependency_links.txt                                                
[2024-07-30 22:04:03,626] [    INFO] egg_info.py:292 - writing top-level names to paddlenlp_ops.egg-info/top_level.txt                                                        
[2024-07-30 22:04:03,918] [    INFO] sdist.py:190 - reading manifest file 'paddlenlp_ops.egg-info/SOURCES.txt'                                                                
[2024-07-30 22:04:03,928] [    INFO] util.py:328 - writing manifest file 'paddlenlp_ops.egg-info/SOURCES.txt'                                                                 
[2024-07-30 22:04:03,929] [    INFO] bdist_egg.py:161 - installing library code to build/paddlenlp_ops/bdist.linux-x86_64/egg                                                 
[2024-07-30 22:04:03,930] [    INFO] dist.py:985 - running install_lib                                                                                                        
[2024-07-30 22:04:03,930] [    INFO] dist.py:985 - running build_ext                   
Compiling user custom op, it will cost a few seconds.....                              
[2024-07-30 22:04:04,628] [    INFO] build_ext.py:533 - building 'paddlenlp_ops' extension                                                                                    
[2024-07-30 22:04:04,628] [    INFO] spawn.py:38 - /mnt/petrelfs/share/cuda-11.2/bin/nvcc -I/mnt/petrelfs/taitan/miniconda3/envs/paddle/lib/python3.9/site-packages/paddle/in
clude -I/mnt/petrelfs/taitan/miniconda3/envs/paddle/lib/python3.9/site-packages/paddle/include/third_party -I/mnt/petrelfs/share/cuda-11.2/include -I/mnt/petrelfs/taitan/mi
niconda3/envs/paddle/include/python3.9 -I/mnt/petrelfs/taitan/miniconda3/envs/paddle/include/python3.9 -c /mnt/petrelfs/taitan/projects/PaddleMIX/PaddleNLP/csrc/generation/
dequant_int8.cu -o /mnt/petrelfs/taitan/projects/PaddleMIX/PaddleNLP/csrc/build/paddlenlp_ops/lib.linux-x86_64-cpython-39/dequant_int8.cu.o -DPADDLE_WITH_CUDA -DEIGEN_USE_GP
U -ccbin cc -Xcompiler -fPIC --expt-relaxed-constexpr -DNVCC -O3 -std=c++11 -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_BFLOAT16_OPERATORS__ -U__C
UDA_NO_BFLOAT16_CONVERSIONS__ -U__CUDA_NO_BFLOAT162_OPERATORS__ -U__CUDA_NO_BFLOAT162_CONVERSIONS__ -gencode arch=compute_80,code=sm_80 -w -DPADDLE_WITH_CUSTOM_KERNEL -D_GLIB
CXX_USE_CXX11_ABI=1                        
/mnt/petrelfs/taitan/miniconda3/envs/paddle/lib/python3.9/site-packages/paddle/include/paddle/utils/flat_hash_map.h: In instantiation of ‘struct paddle::detailv3::HashPolicy
Selector<phi::Place::Hash, void>’:         
/mnt/petrelfs/taitan/miniconda3/envs/paddle/lib/python3.9/site-packages/paddle/include/paddle/utils/flat_hash_map.h:664:51:   required from ‘class paddle::detailv3::sherwood
_v3_table<std::pair<phi::Place, const phi::DeviceContext*>, phi::Place, phi::Place::Hash, paddle::detailv3::KeyOrValueHasher<phi::Place, std::pair<phi::Place, const phi::Devi
ceContext*>, phi::Place::Hash>, std::equal_to<phi::Place>, paddle::detailv3::KeyOrValueEquality<phi::Place, std::pair<phi::Place, const phi::DeviceContext*>, std::equal_to<ph
i::Place> >, std::allocator<std::pair<phi::Place, const phi::DeviceContext*> >, std::allocator<paddle::detailv3::sherwood_v3_entry<std::pair<phi::Place, const phi::DeviceCont
ext*> > > >’                               
/mnt/petrelfs/taitan/miniconda3/envs/paddle/lib/python3.9/site-packages/paddle/include/paddle/utils/flat_hash_map.h:1578:7:   required from ‘class paddle::flat_hash_map<phi:
:Place, const phi::DeviceContext*, phi::Place::Hash>’                                  
/mnt/petrelfs/taitan/miniconda3/envs/paddle/lib/python3.9/site-packages/paddle/include/paddle/phi/api/include/context_pool.h:86:75:   required from here
/mnt/petrelfs/taitan/miniconda3/envs/paddle/lib/python3.9/site-packages/paddle/include/paddle/utils/flat_hash_map.h:198:33: error: no type named ‘hash_policy’ in ‘struct phi
::Place::Hash’                             
   typedef typename T::hash_policy type;                                               
                                 ^                                                     
error: command '/mnt/petrelfs/share/cuda-11.2/bin/nvcc' failed with exit code 1

稳定复现步骤 & 代码

A100 80G
CDUA 11.2, V11.2.152
gcc version 11.2.0 (GCC)
Python 3.9.19