強化学習でRuntimeError: CUDA error: out of memoryが出る

以下のプログラムは途中までのものですが、google colabだと動いたのですが、会社のGPUを積んだマシンだとエラーがでました。エラーが出た原因が知りたいです。
動かしたプログラム

import gym
from creversi.gym_reversi.envs import ReversiVecEnv
from creversi import *

import os
import datetime
import math
import random
import numpy as np
from collections import namedtuple
from itertools import count
from tqdm import tqdm_notebook as tqdm
import matplotlib.pyplot as plt

import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F

BATCH_SIZE = 256

vecenv = ReversiVecEnv(BATCH_SIZE)

# if gpu is to be used
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

######################################################################
# Replay Memory

Transition = namedtuple('Transition',
                        ('state', 'action', 'next_state', 'next_actions', 'reward'))


class ReplayMemory(object):

    def __init__(self, capacity):
        self.capacity = capacity
        self.memory = []
        self.position = 0

    def push(self, *args):
        """Saves a transition."""
        if len(self.memory) < self.capacity:
            self.memory.append(None)
        self.memory[self.position] = Transition(*args)
        self.position = (self.position + 1) % self.capacity

    def sample(self, batch_size):
        return random.sample(self.memory, batch_size)

    def __len__(self):
        return len(self.memory)

######################################################################
# DQN

k = 192
fcl_units = 256
class DQN(nn.Module):

    def __init__(self):
        super(DQN, self).__init__()
        self.conv1 = nn.Conv2d(2, k, kernel_size=3, padding=1)
        self.bn1 = nn.BatchNorm2d(k)
        self.conv2 = nn.Conv2d(k, k, kernel_size=3, padding=1)
        self.bn2 = nn.BatchNorm2d(k)
        self.conv3 = nn.Conv2d(k, k, kernel_size=3, padding=1)
        self.bn3 = nn.BatchNorm2d(k)
        self.conv4 = nn.Conv2d(k, k, kernel_size=3, padding=1)
        self.bn4 = nn.BatchNorm2d(k)
        self.conv5 = nn.Conv2d(k, k, kernel_size=3, padding=1)
        self.bn5 = nn.BatchNorm2d(k)
        self.conv6 = nn.Conv2d(k, k, kernel_size=3, padding=1)
        self.bn6 = nn.BatchNorm2d(k)
        self.conv7 = nn.Conv2d(k, k, kernel_size=3, padding=1)
        self.bn7 = nn.BatchNorm2d(k)
        self.conv8 = nn.Conv2d(k, k, kernel_size=3, padding=1)
        self.bn8 = nn.BatchNorm2d(k)
        self.conv9 = nn.Conv2d(k, k, kernel_size=3, padding=1)
        self.bn9 = nn.BatchNorm2d(k)
        self.conv10 = nn.Conv2d(k, k, kernel_size=3, padding=1)
        self.bn10 = nn.BatchNorm2d(k)
        self.fcl1 = nn.Linear(k * 64, fcl_units)
        self.fcl2 = nn.Linear(fcl_units, 65)

    def forward(self, x):
        x = F.relu(self.bn1(self.conv1(x)))
        x = F.relu(self.bn2(self.conv2(x)))
        x = F.relu(self.bn3(self.conv3(x)))
        x = F.relu(self.bn4(self.conv4(x)))
        x = F.relu(self.bn5(self.conv5(x)))
        x = F.relu(self.bn6(self.conv6(x)))
        x = F.relu(self.bn7(self.conv7(x)))
        x = F.relu(self.bn8(self.conv8(x)))
        x = F.relu(self.bn9(self.conv9(x)))
        x = F.relu(self.bn10(self.conv10(x)))
        x = F.relu(self.fcl1(x.view(-1, k * 64)))
        x = self.fcl2(x)
        return x.tanh()

def get_states(envs):
    features_vec = np.zeros((BATCH_SIZE, 2, 8, 8), dtype=np.float32)
    for i, env in enumerate(envs):
        env.board.piece_planes(features_vec[i])
    return torch.from_numpy(features_vec).to(device)

この最後のreturn torch.from_numpy(features_vec).to(device)のところで、、google colabでは
エラーが出ないのですが、会社のGPUを積んだLinuxだと
RuntimeError: CUDA error: out of memory
CUDA kernel errors might be asynchronously reported at some other API call,so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
というエラーがでます。ググったりして調べたのですが、原因が分かりませんでした。
会社のGPUの環境は、下記の通りです。
-bash-4.2$ lspci | grep -i nvidia
3b:00.0 VGA compatible controller: NVIDIA Corporation TU102 [GeForce RTX 2080 Ti] (rev a1)
3b:00.1 Audio device: NVIDIA Corporation TU102 High Definition Audio Controller (rev a1)
3b:00.2 USB controller: NVIDIA Corporation TU102 USB 3.1 Host Controller (rev a1)
3b:00.3 Serial bus controller [0c80]: NVIDIA Corporation TU102 USB Type-C UCSI Controller (rev a1)
af:00.0 VGA compatible controller: NVIDIA Corporation GP102 [GeForce GTX 1080 Ti] (rev a1)
af:00.1 Audio device: NVIDIA Corporation GP102 HDMI Audio Controller (rev a1)

ご教示いただけますと幸いです。
何卒、よろしくお願い申し上げます。

episteme

2021/12/10 02:46

単にデバイス(グラボ)上のメモリーが足りんのじゃないですか?

onosan

2021/12/10 02:56

ありがとうございます。 nvidia-smiをためすと以下のようになりますが、足りていないでしょうか。 +-----------------------------------------------------------------------------+ | NVIDIA-SMI 440.82 Driver Version: 440.82 CUDA Version: 10.2 | |-------------------------------+----------------------+----------------------+ | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | |===============================+======================+======================| | 0 GeForce RTX 208... Off | 00000000:3B:00.0 Off | N/A | | 28% 36C P2 61W / 250W | 11013MiB / 11019MiB | 0% Default | +-------------------------------+----------------------+----------------------+ | 1 GeForce GTX 108... Off | 00000000:AF:00.0 Off | N/A | | 20% 24C P8 8W / 250W | 10MiB / 11178MiB | 0% Default | +-------------------------------+----------------------+----------------------+ +-----------------------------------------------------------------------------+ | Processes: GPU Memory | | GPU PID Type Process name Usage | |=============================================================================| | 0 101628 C .../local/anaconda3/envs/37prod/bin/python 6897MiB | | 0 169129 C .../local/anaconda3/envs/37prod/bin/python 833MiB | | 0 181127 C /home/s80141/.conda/envs/s80141/bin/python 959MiB | | 0 188149 C .../local/anaconda3/envs/37prod/bin/python 2313MiB | +-----------------------------------------------------------------------------+

episteme

2021/12/10 02:58

いやそんなの知らんよgoogle colabでどんだけメモリ使ったんかわからんのだから。

行動規範の内容に同意します

回答1件

ベストアンサー

CUDA error: out of memory なので、単純に GPU のメモリが足りない
無料の Goolge Colab で割り当てられる Tesla K80 とかでも 24GB のメモリがあったと思いますが、市販のゲーミング用の GPU だと多くてもメモリ量が 12GB とかなので、業務用の GPU に比べるとメモリ量が少ない

以下のバッチサイズをエラーがでなくなるまで小さい値にしてください