torch學習(四) using GPU
阿新 • • 發佈:2019-02-12
一、獲取GPU資訊
require 'cutorch'print(
cutorch.getDeviceProperties(cutorch.getDevice()) )
二、GPU和CPU之間的資料互通
1.構建GPU資料
t1 = torch.CudaTensor(100):fill(0.5)
t2 = torch.CudaTensor(100):fill(1) t1:add(t2) 2.GPU=>CPU t1_cpu = t1:float() 3.CPU=>GPU t1:zero()
t1[{}] = t1_cpu -- copies the data back to the GPU, with no new alloc
t1_new = t1_cpu:cuda() -- allocates a new tensor
三、nn使用GPU
Module引數是Torch的預設型別,如果需要Cuda的module,需要把預設型別修改為CUda。
1.構建基於Cuda的nn
require 'cunn'
-- we define an MLP
mlp = nn.Sequential()
mlp:add(nn.Linear(ninput, 1000))
mlp:add(nn.Tanh())
mlp:add(nn.Linear(1000, 1000))
mlp:add(nn.Tanh())
mlp:add(nn.Linear(1000, 1000 ))
mlp:add(nn.Tanh())
mlp:add(nn.Linear(1000, noutput))
-- and move it to the GPU: mlp:cuda() 輸入資料需要首先轉化為cuda -- input
input = torch.randn(ninput)
-- retype and feed to network:
result = mlp:forward( input:cuda() )
-- the result is a CudaTensor, if your loss is CPU-based, then you will-- need to bring it back:
result_cpu = result:float()
可以增加Copy層自動進行資料型別轉化
-- we put the mlp in a new container:
mlp_auto = nn.Sequential()
mlp_auto:add(nn.Copy('torch.FloatTensor', 'torch.CudaTensor'))
mlp_auto:add(mlp)
mlp_auto:add(nn.Copy('torch.CudaTensor', 'torch.FloatTensor'))
t2 = torch.CudaTensor(100):fill(1) t1:add(t2) 2.GPU=>CPU t1_cpu = t1:float() 3.CPU=>GPU t1:zero()
t1[{}] = t1_cpu -- copies the data back to the GPU, with no new alloc
mlp = nn.Sequential()
mlp:add(nn.Linear(ninput, 1000))
mlp:add(nn.Tanh())
mlp:add(nn.Linear(1000, 1000))
mlp:add(nn.Tanh())
mlp:add(nn.Linear(1000, 1000
mlp:add(nn.Tanh())
mlp:add(nn.Linear(1000, noutput))
-- and move it to the GPU: mlp:cuda() 輸入資料需要首先轉化為cuda -- input
input = torch.randn(ninput)
-- retype and feed to network:
result = mlp:forward( input:cuda() )
-- the result is a CudaTensor, if your loss is CPU-based, then you will-- need to bring it back:
mlp_auto = nn.Sequential()
mlp_auto:add(nn.Copy('torch.FloatTensor', 'torch.CudaTensor'))
mlp_auto:add(mlp)
mlp_auto:add(nn.Copy('torch.CudaTensor', 'torch.FloatTensor'))