修复wordsize=8

colossalai.egg-info/PKG-INFO

 Metadata-Version: 2.1
 Name: colossalai
-Version: 0.0.2
+Version: 0.0.2-gitc789114680-dtk2210
 Summary: An integrated large-scale model training system with efficient parallelization techniques
 Home-page: UNKNOWN
 License: UNKNOWN

compile.sh

@@ -4,9 +4,11 @@ export CPLUS_INCLUDE_PATH=${ROCM_PATH}/hiprand/include/:${ROCM_PATH}/rocrand/inc
 export LD_LIBRARY_PATH=${ROCM_PATH}/hipblas/lib/:$LD_LIBRARY_PATH
 export LD_LIBRARY_PATH=${ROCM_PATH}/llvm/lib:$LD_LIBRARY_PATH
 
-NAME="colossalai-0.0.2-cp37-cp37m-linux_x86_64"
-rm dist/${NAME}.whl
+#NAME="colossalai-0.0.2-cp37-cp37m-linux_x86_64"
+rm dist/*.whl -f
 
 MAX_JOBS=32 python3 setup.py -v  bdist_wheel
 
-pip3 install dist/${NAME}.whl  -i http://mirrors.aliyun.com/pypi/simple/  --trusted-host mirrors.aliyun.com --no-deps --force-reinstall
+pip3 install dist/*.whl  -i http://mirrors.aliyun.com/pypi/simple/  --trusted-host mirrors.aliyun.com --no-deps --force-reinstall
+
+./test_gpu.sh

setup.py

@@ -281,7 +281,7 @@ if build_cuda_ext:
 
 setup(
     name='colossalai',
-    version='0.0.2',
+    version='0.0.2-gitc789114680-dtk2210',
     packages=find_packages(exclude=('benchmark',
                                     'docker',
                                     'tests',

test_gpu.sh

@@ -3,4 +3,9 @@ export LD_LIBRARY_PATH=/usr/local/lib/python3.7/site-packages/torch/lib/:$LD_LIB
 
 #export HIP_VISIBLE_DEVICES=0,1,2,3 
 DATA=./cifar_dataset pytest  tests
-#DATA=./cifar_dataset pytest -v tests/test_zero_tensor_parallel/
+
+#DATA=./cifar_dataset pytest -v tests/test_engine/test_engine/test_engine_naive_amp.py
+#DATA=./cifar_dataset pytest -v tests/test_zero_data_parallel/test_zero_level_3.py
+#DATA=./cifar_dataset pytest -v tests/test_zero_data_parallel/test_zero_level_2.py
+#DATA=./cifar_dataset pytest -v tests/test_zero_data_parallel/test_zero_level_3.py
+#DATA=./cifar_dataset pytest -v tests/test_zero_tensor_parallel/test_vit_2d_level_2.py