PyTorch bindings for Warp-ctc

This is an extension onto the original repo found here.

安装方法

git clone https://github.com/SeanNaren/warp-ctc.git 或点击链接直接下载Zip cd warp-ctc git reset ac045b6 mkdir build; cd build cmake .. make

cd ../pytorch_binding python setup.py install

以上成功之后 直接将 warp-ctc/warp-ctc-cuda10.1/pytorch_binding/build/lib.linux-x86_64-3.7 下的 warpctc_pytorch 文件夹 复制到~/anaconda3/lib/python3.7/site-packages 里面

（该文件夹里面有__init__.py 和 _warp_ctc.cpython-37m-x86_64-linux-gnu.so 两个文件） （以上路径和文件名可能根据Python和系统环境有所差别。） 复制完成之后可以测试import

###以上成功之后添加以下2行到：.bashrc 不用该操作 ###export LD_LIBRARY_PATH=/anaconda3/lib/python3.7/site-packages/warp-ctc/pytorch_binding:$LD_LIBRARY_PATH ###export WARP_CTC_PATH="/anaconda3/lib/python3.7/site-packages/warp-ctc/build"

注意问题： 一、如果当前的cuda是只是使用conda安装的cudatoolkit版本，则需再安装完整的cuda版本才能完成cmake编译gpu版本 否则只会编译cpu版本。 安装方法可以使用： 1、到 https://developer.nvidia.com/cuda-toolkit-archive 自行下载 run安装包安装

二、当前环境
    LinuxMint19.2
    gcc-7 G++7
    Python 3.7
    pytorch 1.3.1
    CUDA 10.1

从 git clone https://github.com/SeanNaren/warp-ctc.git 下来的编译代码比较旧 CUDA编译不过 需要按照下面内容进行修改：

SeanNaren/deepspeech.pytorch#397 zhenglilei commented on 15 Oct ##---- src/reduce.cu Line 44 to : shuff = __shfl_down_sync(0xFFFFFFFF, x, offset);

include/contrib/moderngpu/include/device/intrinsics.cuh Line 115 to : var = __shfl_up_sync(0xFFFFFFFF, var, delta, width); Line 125 to : p.x = __shfl_up_sync(0xFFFFFFFF, p.x, delta, width); Line 126 to : p.y = __shfl_up_sync(0xFFFFFFFF, p.y, delta, width); Line 143 to : "shfl.up.sync.b32 r0|p, %1, %2, %3, %4;" Line 158 to : "shfl.up.sync.b32 r0|p, %1, %2, %3, %4;"

works fine with CUDA 10.1

This is the correct solution by Oct. 2019.

Installation

Install PyTorch v0.4.

WARP_CTC_PATH should be set to the location of a built WarpCTC (i.e. libwarpctc.so). This defaults to ../build, so from within a new warp-ctc clone you could build WarpCTC like this:

git clone https://github.com/SeanNaren/warp-ctc.git
cd warp-ctc
mkdir build; cd build
cmake ..
make

Now install the bindings:

cd pytorch_binding
python setup.py install

If you try the above and get a dlopen error on OSX with anaconda3 (as recommended by pytorch):

cd ../pytorch_binding
python setup.py install
cd ../build
cp libwarpctc.dylib /Users/$WHOAMI/anaconda3/lib

This will resolve the library not loaded error. This can be easily modified to work with other python installs if needed.

Example to use the bindings below.

import torch
from warpctc_pytorch import CTCLoss
ctc_loss = CTCLoss()
# expected shape of seqLength x batchSize x alphabet_size
probs = torch.FloatTensor([[[0.1, 0.6, 0.1, 0.1, 0.1], [0.1, 0.1, 0.6, 0.1, 0.1]]]).transpose(0, 1).contiguous()
labels = torch.IntTensor([1, 2])
label_sizes = torch.IntTensor([2])
probs_sizes = torch.IntTensor([2])
probs.requires_grad_(True)  # tells autograd to compute gradients for probs
cost = ctc_loss(probs, labels, probs_sizes, label_sizes)
cost.backward()

Documentation

CTCLoss(size_average=False, length_average=False)
    # size_average (bool): normalize the loss by the batch size (default: False)
    # length_average (bool): normalize the loss by the total number of frames in the batch. If True, supersedes size_average (default: False)

forward(acts, labels, act_lens, label_lens)
    # acts: Tensor of (seqLength x batch x outputDim) containing output activations from network (before softmax)
    # labels: 1 dimensional Tensor containing all the targets of the batch in one large sequence
    # act_lens: Tensor of size (batch) containing size of each output sequence from the network
    # label_lens: Tensor of (batch) containing label length of each example