Ytukyg

1 I am using the following code. I run this on a GPU machine with below config: nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2017 NVIDIA Corporation Built on Fri_Sep__1_21:08:03_CDT_2017 Cuda compilation tools, release 9.0, V9.0.176 I am using pytorch version 1.0.0.dev20181123. I face this error: RuntimeError: DataLoader worker (pid 23646) is killed by signal: Illegal instruction. Thinking that its a shared memory problem, I tried the solution here at: https://www.lucidarme.me/increase-shared-memory-limit/ It didn't help. Any pointers? import argparse import json import os import torch #=====START: ADDED FOR DISTRIBUTED====== from distributed import init_distributed, apply_gradient_allreduce, reduce_tensor from torch.utils.data.distributed import DistributedSampler #=====END:...

6 3 $begingroup$ Question A question paper on mathematics contains $12$ questions divided into three parts A, B and C, each containing $4$ questions. In how many ways can an examinee answer $5$ questions, selecting at least one from each part. Attempt Firstly, I selected three questions (one from each part) and it can be done $4 cdot 4 cdot 4$ ways. And hence the remaining two positions for two questions can be given in $^9{mathrm C}_2$ since there is no restrictions now. So, total ways is $36 times 64=2304$ . But, in the answer given in the solution manual is $624$ . And the process described is: A B C 1 1 3 1 2 2 which is then arranged for part to be $3 times (4 times 4 times 4 + 4 times 6 times 6) = 624$ . Why my process is incorrect? I understand the second solution ...