cuda version about vl-bert HOT 15 CLOSED

2333

from vl-bert.

Please make sure the environment is same with README, including system, cuda and python. Good luck for you. Thank you.

from vl-bert.

If you don't provide detailed problems you met, I would close this issue after several days. Thanks!

from vl-bert.

我在运行pretrain脚本时， ./scripts/dist_run_single.sh 4 pretrain/train_end2end.py cfgs/pretrain/base_prec_4x16G_fp32.yaml ./
在准备 bottom-up-attention 出现问题
首先，在lib/下 make

python setup.py build_ext --inplace
running build_ext
skipping 'utils/bbox.c' Cython extension (up-to-date)
skipping 'nms/cpu_nms.c' Cython extension (up-to-date)
skipping 'nms/gpu_nms.cpp' Cython extension (up-to-date)
skipping 'pycocotools/_mask.c' Cython extension (up-to-date)
rm -rf build

其次，Build Caffe and pycaffe
执行：make -j8 && make pycaffe
出现

NVCC src/caffe/util/im2col.cu
NVCC src/caffe/util/math_functions.cu
make: /usr/local/cuda/bin/nvcc: Command not found
make: /usr/local/cuda/bin/nvcc: Command not found
Makefile:594: recipe for target '.build_release/cuda/src/caffe/util/im2col.o' failed
make: *** [.build_release/cuda/src/caffe/util/im2col.o] Error 127
make: *** Waiting for unfinished jobs....
Makefile:594: recipe for target '.build_release/cuda/src/caffe/util/math_functions.o' failed
make: *** [.build_release/cuda/src/caffe/util/math_functions.o] Error 127
NVCC src/caffe/layers/concat_layer.cu
make: /usr/local/cuda/bin/nvcc: Command not found
Makefile:594: recipe for target '.build_release/cuda/src/caffe/layers/concat_layer.o' failed
make: *** [.build_release/cuda/src/caffe/layers/concat_layer.o] Error 127

但我的caffe是有的

caffe version 1.0.0

所以请麻烦指点一下

from vl-bert.

@campper 看起来是你没有装好cuda。。。无法找到nvcc的命令。你可以通过下面的命令来测试cuda是不是配置好了：

nvcc -V

如果装好了，会显示cuda版本。

from vl-bert.

你好，我的机器上有cuda

from vl-bert.

你可以看看which cuda，看是不是cuda路径和上面编译调用的/usr/local/cuda/bin/nvcc不一致

from vl-bert.

你可以看看which cuda，看是不是cuda路径和上面编译调用的/usr/local/cuda/bin/nvcc不一致

好的，非常感谢，我再试试

from vl-bert.

我重新安装了cuda10 和 对应的nvcc
重新进入到bottom_up_attention/caffe后 又将cuda10/include/cudnn.h 拷贝到 bottom-up-attention/caffe/include/caffe/util/ ，然后报 找不到nccl.h
我又按照网上教程安装了nccl
git clone https://github.com/NVIDIA/nccl.git
cd nccl
sudo make install -j4
sudo ldconfig
之后比较顺利的对caffe进行了编译，但又报

make: Nothing to be done for 'all'.
CXX/LD -o python/caffe/_caffe.so python/caffe/_caffe.cpp
python/caffe/_caffe.cpp:1:10: fatal error: Python.h: No such file or directory
#include <Python.h> // NOLINT(build/include_alpha)
^~~~~~~~~~
compilation terminated.
Makefile:507: recipe for target 'python/caffe/_caffe.so' failed
make: *** [python/caffe/_caffe.so] Error 1

from vl-bert.

你可以试试

locate Python.h

from vl-bert.

你好，我通过sudo apt install libboost-python-dev 解决了

然后我在测试caffe是否安装成功时，报

我在bottom_up_attention 看readme.md要求的python环境是2.7 整个系统环境是cuda9.0，不太明白caffe为什么要libcudart.so.10.0

from vl-bert.

caffe

终于搞定了。python2进入python后 pip install -r requirements 把该装的包装好就不错了。不容易。
谢谢，还想请教一下，我能不能直接通过model/pretrain_model 对自己的图片进行预测？

from vl-bert.

你是指bottom_up_attention吗？

from vl-bert.

你是指bottom_up_attention吗？
不是，是否能直接对我自己收集的图片通过您提供的模型进行image caption,还是必须要训练一遍。

from vl-bert.

我们的预训练模型并没有image captioning的功能，你需要自己在你的数据训练。

from vl-bert.