diff --git a/package/fftw/3.3.8/install.sh b/package/fftw/3.3.8/install.sh
index 1e21de5e73378e2b65a9d041eecc0c8f9f7f0959..5ca0422fd770eb622353806d53ffb28b2d2860f6 100644
--- a/package/fftw/3.3.8/install.sh
+++ b/package/fftw/3.3.8/install.sh
@@ -5,5 +5,5 @@ set -e
 cd ${JARVIS_TMP}
 tar -xvf ${JARVIS_DOWNLOAD}/fftw-3.3.8.tar.gz
 cd fftw-3.3.8
-./configure --prefix=$1 --enable-shared --enable-threads --enable-openmp --enable-mpi #--enable-float --enable-fma --enable-neon 
+./configure --prefix=$1 --enable-shared --enable-threads --enable-openmp --enable-mpi --enable-float --enable-fma --enable-neon 
 make -j install
diff --git a/package/glibc/2.28/install.sh b/package/glibc/2.28/install.sh
new file mode 100644
index 0000000000000000000000000000000000000000..5fc3464cfa2650c4a08ae02b907073bad10db048
--- /dev/null
+++ b/package/glibc/2.28/install.sh
@@ -0,0 +1,15 @@
+#!/bin/bash
+set -x
+set -e
+glibc_ver='2.28'
+. ${DOWNLOAD_TOOL} -u https://mirrors.huaweicloud.com/gnu/glibc/glibc-${glibc_ver}.tar.xz
+cd ${JARVIS_TMP}
+rm -rf glibc-${glibc_ver}
+tar -xvf ${JARVIS_DOWNLOAD}/glibc-${glibc_ver}.tar.xz
+cd glibc-${glibc_ver}
+mkdir build
+cd build
+unset LD_LIBRARY_PATH
+../configure --prefix=$1 --disable-profile --enable-add-ons --with-headers=/usr/include --disable-sanity-checks --disable-werror
+make -j
+make install
diff --git a/package/kfft/1.0.0/install.sh b/package/kfft/1.0.0/install.sh
index 5d44996fe4e689a62abd6c4c57797180bca6c051..2a6496a761ba3f4029d652f6e506769e8e124d83 100644
--- a/package/kfft/1.0.0/install.sh
+++ b/package/kfft/1.0.0/install.sh
@@ -5,11 +5,13 @@ lib_name='kml_adapter'
 . ${DOWNLOAD_TOOL} -u https://gitee.com/openeuler/kml_adapter.git -t git
 cd $JARVIS_DOWNLOAD
 cd ./$lib_name/kml_fft_adapter
-export C_INCLUDE_PATH=/usr/local/kml/include/
-export CPLUS_INCLUDE_PATH=/usr/local/kml/include/
+sed -i "s/\/usr\/local\/kml/\$ENV{KML_GCC_PATH}/g" CMakeLists.txt
+export C_INCLUDE_PATH=$KML_GCC_PATH/include/
+export CPLUS_INCLUDE_PATH=$KML_GCC_PATH/include/
 chmod +x build.sh
 dos2unix build.sh
 rm -rf build
 ./build.sh
 mkdir -p $1/lib
-cp -rf build/* $1/lib
\ No newline at end of file
+cp -rf build/* $1/lib
+cp -rf include $1/
\ No newline at end of file
diff --git a/package/osu/1.0.0/install.sh b/package/osu/1.0.0/install.sh
index eb56895bc0de5345b0d24b2155500573f9940d45..5540f6c22757ca282d1b18747b45af7e01a502cb 100644
--- a/package/osu/1.0.0/install.sh
+++ b/package/osu/1.0.0/install.sh
@@ -3,6 +3,7 @@ set -x
 set -e
 . ${DOWNLOAD_TOOL} -u https://github.com/forresti/osu-micro-benchmarks/archive/refs/heads/master.zip -f osu-micro-benchmarks.zip
 cd ${JARVIS_TMP}
+rm -rf osu-micro-benchmarks-master
 unzip ${JARVIS_DOWNLOAD}/osu-micro-benchmarks.zip
 cd osu-micro-benchmarks-master
 ./configure --prefix=$1 CC=mpicc CXX=mpicxx
diff --git a/src/dataService.py b/src/dataService.py
index ef7c27d69a5338b7694adbd4d0bd9d2c5b194052..0e8765f70c28ba0f81ca77beefb6e2c0911ac7a6 100644
--- a/src/dataService.py
+++ b/src/dataService.py
@@ -23,6 +23,7 @@ class DataService(object,metaclass=Singleton):
     dependency = ''
     module_content=''
     env_file = 'env.sh'
+
     # Application Info
     app_name = ''
     build_dir = ''
@@ -37,6 +38,7 @@ class DataService(object,metaclass=Singleton):
     batch_cmd = ''
     loop_cmd = ''
     #Other Info
+    env_config_file = 'JARVIS_CONFIG'
     config_file = 'data.config'
     meta_file = '.meta'
     root_path = os.getcwd()
@@ -55,8 +57,14 @@ class DataService(object,metaclass=Singleton):
         self.isARM = platform.machine() == 'aarch64'
         self.tool = ToolService()
         self.data_process()
-
+    # 优先读取环境变量的JARVIS_CONFIG配置,方便多人同时操控
+    # 然后读取.meta文件中的值
+    # 最后读取data.config中的值
     def get_config_file_name(self):
+        CONFIG_ENV = os.getenv(DataService.env_config_file)
+        if CONFIG_ENV is not None:
+            print("LOAD Config file from ENV:", CONFIG_ENV)
+            return CONFIG_ENV
         if not os.path.exists(DataService.meta_file):
             return DataService.config_file
         return self.tool.read_file(DataService.meta_file)
diff --git a/templates/amber/22/amber.arm.gpu.kgcc.config b/templates/amber/22/amber.arm.gpu.kgcc.config
new file mode 100644
index 0000000000000000000000000000000000000000..e9617f97c67e08e5427a3487babea97264b78e7e
--- /dev/null
+++ b/templates/amber/22/amber.arm.gpu.kgcc.config
@@ -0,0 +1,55 @@
+[SERVER]
+11.11.11.11
+
+[DOWNLOAD]
+
+[DEPENDENCY]
+module use ./software/modulefiles
+module purge
+./jarvis -install openblas/0.3.18 gcc
+./jarvis -install kgcc/9.3.1 com
+module load kgcc/9.3.1
+export CC=gcc CXX=g++ FC=gfortran
+if [ ! -d "./amber22_src_kgcc" ];then
+    mkdir amber22_src_kgcc
+    tar -xvf $JARVIS_DOWNLOAD/Amber22.tar.bz2 -C amber22_src_kgcc
+    tar -xvf $JARVIS_DOWNLOAD/AmberTools22.tar.bz2 -C amber22_src_kgcc
+fi
+
+[ENV]
+module use ./software/modulefiles
+module use ./software/moduledeps
+module purge
+module load kgcc/9.3.1
+#module load gcc7.3.0/openblas/0.3.18
+module load kml-gcc/1.6.0
+
+[APP]
+app_name = Amber
+build_dir = $JARVIS_ROOT/amber22_src_kgcc/amber22_src/
+binary_dir = $JARVIS_ROOT/amber22_kgcc/bin
+case_dir = $JARVIS_ROOT/workloads/amber/PME/STMV_production_NPT_4fs
+#binary_dir = 
+#case_dir = $JARVIS_ROOT/workloads/amber/PME/JAC_production_NPT_4fs
+
+[BUILD]
+content='cmake $AMBER_PREFIX/amber22_src  -DCMAKE_INSTALL_PREFIX='$JARVIS_ROOT'/amber22_kgcc -DCOMPILER=GNU -DBUILD_PYTHON=FALSE -DMPI=FALSE -DCUDA=TRUE -DINSTALL_TESTS=FALSE -DDOWNLOAD_MINICONDA=FALSE -DMINICONDA_USE_PY3=FLASE -DBLAS_LIBRARIES='$KML_GCC_PATH'/lib/kblas/omp/libkblas.so -DLAPACK_LIBRARIES='$KML_GCC_PATH'/lib/libklapack_full.so -DCMAKE_VERBOSE_MAKEFILE=ON -DCMAKE_C_FLAGS="-O3 -march=armv8.2-a -mcpu=tsv110" -DCMAKE_CXX_FLAGS="-O3 -march=armv8.2-a -mcpu=tsv110" -DCMAKE_Fortran_FLAGS="-O3 -march=armv8.2-a -mcpu=tsv110"  2>&1 | tee  cmake.log' 
+cd build
+sed -i "40d" run_cmake 
+sed -i "39a $content" run_cmake
+# may stuck at Running updater..., just retry...
+#./run_cmake
+#This may have  error at 49%, just retry...
+make -j install
+
+[CLEAN]
+cd build
+./clean_build
+
+[RUN]
+#run = dsub -s run.sh
+run = numactl -C 0,1,2,3
+binary = pmemd.cuda -O -i mdinOPT.GPU -o mdout-kgcc -p ../Topologies/STMV.prmtop -c ../Coordinates/STMV.inpcrd && cat mdout-kgcc
+#binary = pmemd.cuda -O -i mdinOPT.GPU -o mdout-kgcc -p ../Topologies/JAC.prmtop -c ../Coordinates/JAC.inpcrd && cat mdout-kgcc
+#binary =
+nodes = 1
diff --git a/templates/amber/22/data.amber.arm.gpu.config b/templates/amber/22/data.amber.arm.gpu.config
new file mode 100644
index 0000000000000000000000000000000000000000..55bf68112262aaafad81d7557767029ac9e3d836
--- /dev/null
+++ b/templates/amber/22/data.amber.arm.gpu.config
@@ -0,0 +1,70 @@
+[SERVER]
+11.11.11.11
+
+[DOWNLOAD]
+Amber-benchmark/20 https://ambermd.org/Amber20_Benchmark_Suite.tar.gz
+
+[DEPENDENCY]
+module use ./software/modulefiles
+module purge
+./jarvis -install openblas/0.3.18 gcc
+./jarvis -install bisheng/2.5.0 com
+module load bisheng/2.5.0
+export CC=clang CXX=clang++ FC=flang
+./jarvis -install hmpi/1.2.0 clang
+if [ ! -d "./amber22_src" ];then
+    tar -xvf $JARVIS_DOWNLOAD/Amber22.tar.bz2
+    tar -xvf $JARVIS_DOWNLOAD/AmberTools22.tar.bz2
+    cd amber22_src
+    sed -i '22{h;d};23G' AmberTools/src/reduce/libpdb/pdb++.cpp
+    sed -i '30d' AmberTools/src/reduce/libpdb/pdb++.cpp
+    sed -i '24a }' AmberTools/src/reduce/libpdb/pdb++.cpp
+
+    sed -i '23{h;d};24G' AmberTools/src/reduce/libpdb/pdb_type.cpp
+    sed -i '33d' AmberTools/src/reduce/libpdb/pdb_type.cpp
+    sed -i '26a }' AmberTools/src/reduce/libpdb/pdb_type.cpp
+
+    sed -i "27s/^/!/g" AmberTools/src/sebomd/se_etimer.F
+    sed -i "27a call cpu_time(tnow)" AmberTools/src/sebomd/se_etimer.F
+fi
+#yum install cmake* flex* bison* boost* python2-pip -y
+#pip2 install numpy scipy matplotlib
+
+[ENV]
+module use ./software/modulefiles
+module use ./software/moduledeps
+module purge
+module load bisheng/2.5.0
+module load hmpi/1.2.0
+module load gcc7.3.0/openblas/0.3.18
+export CC=mpicc CXX=mpicxx FC=mpifort
+
+[APP]
+app_name = Amber
+build_dir = $JARVIS_ROOT/amber22_src/build
+#binary_dir = $JARVIS_ROOT/amber22/bin
+case_dir = $JARVIS_ROOT/workloads/amber/PME/STMV_production_NPT_4fs
+binary_dir = 
+#case_dir = $JARVIS_ROOT/workloads/amber/PME/JAC_production_NPT_4fs
+
+[BUILD]
+sed -i "137s/gfortran/flang/g" ../cmake/AmberCompilerConfig.cmake
+mpi_path=`which mpicc`
+mpi_path=${mpi_path%/*/*}
+
+content='cmake $AMBER_PREFIX/amber22_src  -DCMAKE_INSTALL_PREFIX='$JARVIS_ROOT'/amber22 -DCOMPILER=CLANG -DBUILD_PYTHON=FALSE -DMPI=TRUE -DCUDA=TRUE -DINSTALL_TESTS=TRUE -DDOWNLOAD_MINICONDA=FALSE -DMINICONDA_USE_PY3=FLASE -DCMAKE_VERBOSE_MAKEFILE=ON -DCMAKE_C_FLAGS="-O3 -mllvm -force-customized-pipeline -march=armv8.2-a -mcpu=tsv110 -Wl,-z,muldefs -fuse-ld=lld" -DCMAKE_CXX_FLAGS="-O3 -mllvm -force-customized-pipeline -march=armv8.2-a -mcpu=tsv110 -Wno-c++11-narrowing -Wl,-z,muldefs -fuse-ld=lld" -DCMAKE_Fortran_FLAGS="-O3 -mllvm -force-customized-pipeline -march=armv8.2-a -mcpu=tsv110 -I'${mpi_path}'/include -L'${mpi_path}'/lib -fuse-ld=lld"  2>&1 | tee  cmake.log' 
+sed -i "40d" run_cmake 
+sed -i "39a $content" run_cmake
+#./run_cmake
+make -j install
+
+[CLEAN]
+./clean_build
+
+[RUN]
+run = dsub -s run.sh
+#run = numactl -C 0,1,2,3
+#binary = pmemd.cuda -O -i mdinOPT.GPU -o mdout -p ../Topologies/STMV.prmtop -c ../Coordinates/STMV.inpcrd && cat mdout
+#binary = pmemd.cuda -O -i mdinOPT.GPU -o mdout -p ../Topologies/JAC.prmtop -c ../Coordinates/JAC.inpcrd && cat mdout
+binary =
+nodes = 1
diff --git a/templates/amber/22/run.sh b/templates/amber/22/run.sh
new file mode 100644
index 0000000000000000000000000000000000000000..20c4c744d3f75e3fa01149833a8ab5e153a9d369
--- /dev/null
+++ b/templates/amber/22/run.sh
@@ -0,0 +1,11 @@
+#!/bin/sh
+#DSUB -n Amber22_JAC_NPT_4fs
+#DSUB -A default
+#DSUB -N 1
+#DSUB -R cpu=128;gpu=1
+#DSUB --job_type cosched
+#DSUB -oo out_%J.log
+#DSUB -eo err_%J.log
+
+time -p $JARVIS_ROOT/amber22/bin/pmemd.cuda -O -i mdinOPT.GPU -o mdout -p ../Topologies/JAC.prmtop -c ../Coordinates/JAC.inpcrd
+cat mdout
diff --git a/templates/gromacs/2022.5/data.gromacs.arm.gpu.config b/templates/gromacs/2022.5/data.gromacs.arm.gpu.config
new file mode 100644
index 0000000000000000000000000000000000000000..95f0d4b9c6ec0912c61b647ec521914418ad8bc9
--- /dev/null
+++ b/templates/gromacs/2022.5/data.gromacs.arm.gpu.config
@@ -0,0 +1,71 @@
+[SERVER]
+11.11.11.11
+
+[DOWNLOAD]
+gromacs/2022.3 http://ftp.gromacs.org/pub/gromacs/gromacs-2022.5.tar.gz
+
+[DEPENDENCY]
+#set -e
+#set -x
+module purge
+module use ./software/modulefiles
+module use ./software/moduledeps
+./jarvis -install bisheng/2.5.0 com
+./jarvis -install openblas/0.3.18 gcc
+module load kgcc/9.3.1
+module load kgcc9.3.1/kml-gcc/1.6.0
+./jarvis -install kfft/1.0.0 gcc
+module load bisheng/2.5.0
+./jarvis -install hmpi/1.2.0 clang
+module load hmpi/1.2.0
+export CC=mpicc CXX=mpicxx FC=mpifort
+./jarvis -install fftw/3.3.8 clang+mpi
+#tar -xvf ${JARVIS_DOWNLOAD}/gromacs-2022.5.tar.gz
+
+[ENV]
+module purge
+module use ./software/modulefiles
+module use ./software/moduledeps
+module load bisheng/2.5.0
+export PATH=/usr/local/cuda/bin:$PATH
+module load hmpi/1.2.0
+module load kgcc/9.3.1
+module load fftw/3.3.8
+#module load kgcc9.3.1/kml-gcc/1.6.0
+#module load kgcc9.3.1/kfft/1.0.0
+module load gcc7.3.0/openblas/0.3.18
+export CC=mpicc CXX=mpicxx FC=mpifort
+export GMX_FORCE_UPDATE_DEFAULT_GPU=true
+export GMX_GPU_PME_PP_COMMS=true
+export GMX_GPU_DD_COMMS=true
+
+[APP]
+app_name = gromacs
+build_dir = $JARVIS_ROOT/gromacs-2022.5
+binary_dir = 
+case_dir = $JARVIS_ROOT/workloads/gromacs/water-cut1.0_GMX50_bare/0768/
+
+[BUILD]
+sed -i '204s/return fftw_version;/return 0;/g' src/gromacs/utility/binaryinformation.cpp
+sed -i '206s/return fftwf_version;/return 0;/g' src/gromacs/utility/binaryinformation.cpp
+sed -i '505s/int fftwflags = FFTW_DESTROY_INPUT;/int fftwflags = 0;/g' src/gromacs/fft/fft5d.cpp
+rm -rf build
+mkdir -p build
+cd build
+FLAGS="-mcpu=tsv110 -O3"; CFLAGS=$FLAGS CXXFLAGS="-mcpu=tsv110 -O3" CC=mpicc CXX=mpicxx FC=mpifort cmake -DCMAKE_INSTALL_PREFIX=$JARVIS_ROOT/gromacs-kml -DBUILD_SHARED_LIBS=on -DBUILD_TESTING=on -DREGRESSIONTEST_DOWNLOAD=off -DGMX_BUILD_OWN_FFTW=off -DGMX_SIMD=ARM_NEON_ASIMD -DGMX_DOUBLE=off -DGMX_EXTERNAL_BLAS=on -DGMX_EXTERNAL_LAPACK=on -DGMX_FFT_LIBRARY=fftw3 -DGMX_BLAS_USER=${OPEBLAS_PATH}/lib/libopenblas.a -DGMX_LAPACK_USER=${OPEBLAS_PATH}/lib/libopenblas.a -DFFTWF_LIBRARY=$FFTW_PATH/lib/libfftw3f.so -DFFTWF_INCLUDE_DIR=$FFTW_PATH_PATH/include -DGMX_GPU=CUDA -DGMX_MPI=off -DGMX_OPENMP=on -DGMX_X11=off -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda ../
+
+make -j40 V=1
+make install
+
+#-DGMX_BLAS_USER=$openblas_path/lib/libopenblas.a #-DGMX_LAPACK_USER=$openblas_path/lib/libopenblas.a 
+#-DGMX_BLAS_USER=${KML_GCC_PATH}/lib/kblas/omp/libkblas.so -DGMX_LAPACK_USER=${KML_GCC_PATH}/lib/libklapack_full.so -DFFTWF_LIBRARY=$KFFT_PATH/lib/libfftw3f.so -DFFTWF_INCLUDE_DIR=$KFFT_PATH/include
+
+[CLEAN]
+./clean_build
+
+[RUN]
+run = chmod +x run.sh && dsub -s run.sh
+# binary = gmx grompp -f pme.mdp
+# binary = gmx mdrun -dlb yes -pin on -pinoffset 0 -pinstride 1  -ntmpi 1  -ntomp 48 -v -nsteps 100000  -resetstep 80000 -noconfout -nb gpu -bonded cpu -pme gpu -nstlist 400 -s ./topol.tpr
+binary =
+nodes = 1
diff --git a/templates/gromacs/2022.5/run.sh b/templates/gromacs/2022.5/run.sh
new file mode 100644
index 0000000000000000000000000000000000000000..5a3701361137cd9e42e002f213088cb408408b21
--- /dev/null
+++ b/templates/gromacs/2022.5/run.sh
@@ -0,0 +1,14 @@
+#!/bin/sh
+#DSUB -n gromacs_job
+#DSUB -A default
+#DSUB -N 1
+#DSUB -R cpu=128;gpu=1
+#DSUB --job_type cosched
+#DSUB -oo out_%J.log
+#DSUB -eo err_%J.log
+
+ddd=`date "+%Y-%m-%d-%H:%M:%S"`
+
+cat $CCSCHEDULER_ALLOC_FILE |awk '{print $1}'|awk NF > hostfile
+export CUDA_VISIBLE_DEVICES=0
+$JARVIS_ROOT/gromacs/bin/gmx mdrun -dlb yes -pin on -pinoffset 0 -pinstride 1  -ntmpi 1 -ntomp 64 -v -nsteps 100000  -resetstep 80000 -noconfout -nb gpu -nstlist 400 -s ./topol.tpr
diff --git a/templates/pytorch/1.11.0/cublas-cutlass.py b/templates/pytorch/1.11.0/cublas-cutlass.py
new file mode 100644
index 0000000000000000000000000000000000000000..e03457c8be2f3f2ea616688cd29e7113bffe6e40
--- /dev/null
+++ b/templates/pytorch/1.11.0/cublas-cutlass.py
@@ -0,0 +1,30 @@
+import torch
+
+# 指定使用GPU
+device = torch.device("cuda")
+
+# 创建两个随机矩阵
+A = torch.randn(1000, 1000, device=device)
+B = torch.randn(1000, 1000, device=device)
+
+# 进行矩阵乘法运算
+C = torch.matmul(A, B)
+
+# 将结果转移到CPU上
+C = C.cpu()
+
+# 打印结果
+print(C)
+
+#添加export CUBLASLT_LOG_LEVEL=5 可判别是否调用了cutlass
+#输出：使能cutlass 算法是21号，切分Block Size为128x128x32
+[cublasLt][3794797][Trace][cublasLtSSSMatmul] A=0X400380DD0A00 Adesc=[type=R_32F rows=1000 cols=1000 ld=1000] B=0X400380A00000 Bdesc=[type=R_32F rows=1000 cols=1000 ld=1000] C=0X4003811A1400 Cdesc=[type=R_32F rows=1000 cols=1000 ld=1000] D=0X4003811A1400 Ddesc=[type=R_32F rows=1000 cols=1000 ld=1000] computeDesc=[computeType=COMPUTE_32F_FAST_TF32 scaleType=R_32F smCountTarget=108] algo=[algoId=21 tile=MATMUL_TILE_128x128 stages=MATMUL_STAGES_32x5] workSpace=0X0 workSpaceSizeInBytes=0 beta=0 outOfPlace=0 stream=0X0
+#输出：未使能cutlass 算法是0号或者21号，切分Block Size为64x64x32
+[cublasLt][3837588][Trace][cublasLtSSSMatmul] A=0X400484FD0A00 Adesc=[type=R_32F rows=1000 cols=1000 ld=1000] B=0X400484C00000 Bdesc=[type=R_32F rows=1000 cols=1000 ld=1000] C=0X4004853A1400 Cdesc=[type=R_32F rows=1000 cols=1000 ld=1000] D=0X4004853A1400 Ddesc=[type=R_32F rows=1000 cols=1000 ld=1000] computeDesc=[computeType=COMPUTE_32F_FAST_TF32 scaleType=R_32F] algo=[algoId=21 tile=MATMUL_TILE_64x64 stages=MATMUL_STAGES_32x6] workSpace=0X0 workSpaceSizeInBytes=0 beta=0 outOfPlace=0 stream=0X0
+tensor([[-35.9243, -17.1637,  43.2401,  ...,  -5.8674,  27.3440,   5.0968],
+        [ 67.7127,   0.9203,  -7.6000,  ...,  37.2051, -80.3450,  23.1073],
+        [ -6.7195,   4.0443,  24.7202,  ..., -36.4448,  19.3249, -31.6387],
+        ...,
+        [-13.0232, -41.7477, -19.7421,  ...,  40.6747,  -0.6182, -56.0985],
+        [-33.5083,  12.9116,  17.2967,  ...,  45.3489,  49.2308, -16.7261],
+        [ 25.1251, -43.3838, -30.2829,  ...,  10.3012,  19.7520, -32.6221]])
\ No newline at end of file
diff --git a/templates/pytorch/1.11.0/pytorch.arm.cuda.config b/templates/pytorch/1.11.0/pytorch.arm.cuda.config
new file mode 100644
index 0000000000000000000000000000000000000000..6c9ce1212eb3ade0e00be26bee91feff21a246ef
--- /dev/null
+++ b/templates/pytorch/1.11.0/pytorch.arm.cuda.config
@@ -0,0 +1,65 @@
+[SERVER]
+11.11.11.11
+
+[DOWNLOAD]
+pytorch/1.11.0 https://github.com/pytorch/pytorch/archive/refs/tags/v1.11.0.zip pytorch-1.11.0.zip
+
+[DEPENDENCY]
+if [ ! -d "./pytorch" ];then
+    #git clone --depth 1 -b release/1.11 https://github.com/pytorch/pytorch.git
+    git clone --depth 1 -b v1.11.0 https://github.com/pytorch/pytorch.git pytorch-1.11
+    bc2c6edaf163b1a1330e37a6e34caf8c553e4755
+    #unzip $JARVIS_DOWNLOAD/pytorch-1.11.0.zip
+    cd pytorch
+    git submodule sync
+    git submodule update --init --recursive --jobs 0 2>&1 | tee git-submodule.log
+fi
+conda create --name pytorch1.11 python=3.8
+conda install -y pyyaml typing_extensions numpy ccache 
+
+[ENV]
+if ! type conda >/dev/null 2>&1;then
+  echo -e "conda command does not exist, please execute following commands:\n"
+  echo "module use /home/share/apps/modulefiles/"
+  echo "module load anaconda/2021.11"
+  echo -e "source activate pytorch1.11\n"
+  exit 1
+fi
+module use /home/share/apps/modulefiles/
+module load compilers/cuda/11.3.0
+module load compilers/gcc/9.3.1
+module load cudnn/8.2.1_cuda11.3
+module load libs/nccl/2.17.1-1_cuda11.3
+export GIT_SSL_NO_VERIFY=1
+export BUILD_SPLIT_CUDA=ON
+export CAFFE2_STATIC_LINK_CUDA=ON
+export ATEN_STATIC_CUDA=ON
+export USE_STATIC_CUDNN=ON
+export LD_LIBRARY_PATH=$CONDA_PREFIX/lib:$LD_LIBRARY_PATH
+
+[APP]
+app_name = Pytorch
+build_dir = $JARVIS_ROOT/pytorch-1.11
+binary_dir = $JARVIS_ROOT/pytorch-1.11
+case_dir = $JARVIS_ROOT/
+
+[BUILD]
+set -e
+set -x
+python3 setup.py install 2>&1 | tee compile.log
+python3 setup.py sdist bdist_wheel 2>&1| tee gen-whl.log
+
+[CLEAN]
+python3 setup.py clean
+
+[RUN]
+run = python3 -c "import torch; print(torch.__version__)"
+binary =  
+nodes = 1
+
+[BATCH]
+cd build/lib
+echo "check cutlass from--- libtorch_cuda_cpp.so --- library"
+strings libtorch_cuda_cpp.so | grep "cutlass"
+echo "check cutlass from--- libtorch_cuda_cu.so --- library"
+strings libtorch_cuda_cu.so | grep "cutlass"
\ No newline at end of file