kayush2O6

kayush2O6

from numba import cuda
import ctypes
import numpy as np
import torch

def devndarray2torch(dev_arr):
    t = torch.empty(size=dev_arr.shape, dtype=dtyp).cuda()
    ctx = cuda.cudadrv.driver.driver.get_context()
	
    # constant value of #bytes in case of float32 = 4

kayush2O6

BEFORE INSTALLING THE CUML, PLEASE MAKE SURE YOU HAVE FOLLOWED THE ABOVE STEPS FOR CUDF. CUDF SHOULD BE WORKING...

Step 1: Install the cuml and its depandencies.
  
  !apt install libopenblas-base libomp-dev
  !pip install cuml-cuda100
  
  # import cuml at this point, will give libcuml.so not found error. #

NOTE: Step2 is optional and is just for information, you can fast forward to Step3 directly to work quickely.   

kayush2O6

Steps 1: Just to verify that you have all requirement satisfied, needed by rapids. 
  
  * check the gpu card (>=Pascal arch)
  !nvidia-smi
  
  * check cuda version installed (>=9.2)
  !nvcc -V
  
  *check the python and pip version (python==3.6)
  !python -V; pip -V

kayush2O6