MLPerf 机器学习基准测试实战入门（一）NAVIDA-GNMT

最新推荐文章于 2024-09-07 23:13:35 发布
最新推荐文章于 2024-09-07 23:13:35 发布
阅读量3.4k 收藏 10
点赞数
文章标签： linux 深度学习 docker tensorflow java
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.csdn.net/qq_40875849/article/details/119955984
版权

将MLPerf训练结果库拷到本地

使用的是training_results_v0.6，而不是mlperf / training存储库中提供的参考实现。请注意，这些实现有效地用作基准实现的起点，但尚未完全优化，并且不打算用于软件框架或硬件的“实际”性能评估。

git clone https://github.com/Caiyishuai/training_results_v0.6

在此存储库中，有每个供应商提交的目录(Google，Intel，NVIDIA等)，其中包含用于生成结果的代码和脚本。在NVIDIA GPU上运行基准测试。

[root@2 ~]# cd training_results_v0.6/

[root@2 training_results_v0.6]# ls
Alibaba  CONTRIBUTING.md  Fujitsu  Google  Intel  LICENSE  NVIDIA  README.md

[root@2 training_results_v0.6]# cd NVIDIA/; ls
benchmarks  LICENSE.md  README.md  results  systems

[root@2 NVIDIA]# cd benchmarks/; ls
gnmt  maskrcnn  minigo  resnet  ssd  transformer

下载并验证数据集

[root@2 implementations]# pwd
/data/training_results_v0.6/NVIDIA/benchmarks/gnmt/implementations

[root@2 implementations]# ls
data  download_dataset2.sh  download_dataset3.sh  download_dataset.sh  pytorch  verify_dataset.sh  wget-log

[root@2 implementations]# bash download_dataset.sh

查看download_dataset.sh，可以查看数据的具体链接，如果网速较慢，可以将链接复制到其它下载器中下载，然后更改download_dataset.sh

[root@2 implementations]# cat download_dataset.sh
#! /usr/bin/env bash

# Copyright 2017 Google Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#      http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

set -e

export LANG=C.UTF-8
export LC_ALL=C.UTF-8

OUTPUT_DIR=${1:-"data"}
echo "Writing to ${OUTPUT_DIR}. To change this, set the OUTPUT_DIR environment variable."

OUTPUT_DIR_DATA="${OUTPUT_DIR}/data"

mkdir -p $OUTPUT_DIR_DATA

echo "Downloading Europarl v7. This may take a while..."
wget -nc -nv -O ${OUTPUT_DIR_DATA}/europarl-v7-de-en.tgz \
  http://www.statmt.org/europarl/v7/de-en.tgz

echo "Downloading Common Crawl corpus. This may take a while..."
wget -nc -nv -O ${OUTPUT_DIR_DATA}/common-crawl.tgz \
  http://www.statmt.org/wmt13/training-parallel-commoncrawl.tgz

echo "Downloading News Commentary v11. This may take a while..."
wget -nc -nv -O ${OUTPUT_DIR_DATA}/nc-v11.tgz \
  http://data.statmt.org/wmt16/translation-task/training-parallel-nc-v11.tgz

echo "Downloading dev/test sets"
wget -nc -nv -O  ${OUTPUT_DIR_DATA}/dev.tgz \
  http://data.statmt.org/wmt16/translation-task/dev.tgz
wget -nc -nv -O  ${OUTPUT_DIR_DATA}/test.tgz \
  http://data.statmt.org/wmt16/translation-task/test.tgz

………………

done

echo "All done."

如果通过其它方式已经下载了文件在本目录下，可以更改上述wegt代码

echo "Downloading Europarl v7. This may take a while..."
mv -i data/de-en.tgz  ${OUTPUT_DIR_DATA}/europarl-v7-de-en.tgz \
  

echo "Downloading Common Crawl corpus. This may take a while..."
mv -i data/training-parallel-commoncrawl.tgz  ${OUTPUT_DIR_DATA}/common-crawl.tgz \
  
echo "Downloading News Commentary v11. This may take a while..."
mv -i data/training-parallel-nc-v11.tgz  ${OUTPUT_DIR_DATA}/nc-v11.tgz \
  

echo "Downloading dev/test sets"
mv -i data/dev.tgz  ${OUTPUT_DIR_DATA}/dev.tgz \
  
mv -i data/test.tgz  ${OUTPUT_DIR_DATA}/test.tgz \

执行脚本以验证是否已正确下载数据集。

[root@2 implementations]# du -sh data/
13G     data/

配置文件开始准备训练

用于执行训练作业的脚本和代码位于pytorch目录中。

[root@2 implementations]# cd pytorch/
[root@2 pytorch]# ll
total 124
-rw-r--r-- 1 root root  5047 Jan 22 15:45 bind_launch.py
-rwxr-xr-x 1 root root  1419 Jan 22 15:45 config_DGX1_multi.sh
-rwxr-xr-x 1 root root   718 Jan 25 10:50 config_DGX1.sh
-rwxr-xr-x 1 root root  1951 Jan 22 15:45 config_DGX2_multi_16x16x32.sh
-rwxr-xr-x 1 root root  1950 Jan 22 15:45 config_DGX2_multi.sh
-rwxr-xr-x 1 root root   718 Jan 22 15:45 config_DGX2.sh
-rw-r--r-- 1 root root  1372 Jan 22 15:45 Dockerfile
-rw-r--r-- 1 root root  1129 Jan 22 15:45 LICENSE
-rw-r--r-- 1 root root  6494 Jan 22 15:45 mlperf_log_utils.py
-rw-r--r-- 1 root root  4145 Jan 22 15:45 preprocess_data.py
-rw-r--r-- 1 root root 12665 Jan 22 15:45 README.md
-rw-r--r-- 1 root root    43 Jan 22 15:45 requirements.txt
-rwxr-xr-x 1 root root  2220 Jan 22 15:45 run_and_time.sh
-rwxr-xr-x 1 root root  7173 Jan 25 10:56 run.sub
drwxr-xr-x 3 root root    45 Jan 22 15:45 scripts
drwxr-xr-x 7 root root    90 Jan 22 15:45 seq2seq
-rw-r--r-- 1 root root  1082 Jan 22 15:45 setup.py
-rw-r--r-- 1 root root 25927 Jan 22 15:45 train.py
-rw-r--r-- 1 root root  8056 Jan 22 15:45 translate.py

需要配置config_ <system> .sh以反映您的系统配置。如果系统具有8个或16个GPU，则可以使用现有的config_DGX1.sh或config_DGX2.sh配置文件来启动训练作业。

要编辑的参数：DGXNGPU = 8DGXSOCKETCORES = 18DGXNSOCKET = 2

您可以使用nvidia-smi命令获取GPU信息，并使用lscpu命令获取CPU信息，尤其是：

Core(s) per socket: 18
Socket(s): 2

下载docker镜像

docker build -t mlperf-nvidia:rnn_translator .

需要不少时间

[root@2 pytorch]# docker build -t mlperf-nvidia:rnn_translator .
Sending build context to Docker daemon    279kB
Step 1/12 : ARG FROM_IMAGE_NAME=nvcr.io/nvidia/pytorch:19.05-py3
Step 2/12 : FROM ${FROM_IMAGE_NAME}
19.05-py3: Pulling from nvidia/pytorch
7e6591854262: Pulling fs layer 
089d60cb4e0a: Pulling fs layer 
7e6591854262: Downloading [>                                                  ]  452.4kB/43.75MB
45085432511a: Waiting 
6ca460804a89: Waiting 
2631f04ebf64: Pulling fs layer 
86f56e03e071: Pulling fs layer 
234646620160: Waiting 
7f717cd17058: Waiting 
e69a2ba99832: Waiting 
bc9bca17b13c: Waiting 
1870788e477f: Waiting 
603e0d586945: Waiting 
717dfedf079c: Waiting 
2631f04ebf64: Waiting 
c5bd7559c3ad: Waiting 
9c461696bc09: Download complete 
059d4f560014: Waiting 
f3f14cff44df: Waiting 
603e0d586945: Downloading [==========>                                        ]  102.7MB/492.7MB
c5bd7559c3ad: Pull complete 
d82c679b8708: Pull complete 
059d4f560014: Pull complete 
f3f14cff44df: Pull complete 
96502bde320c: Pull complete 
bc5bb9379810: Pull complete 
e4d8bb046bc2: Pull complete 
4e2187010a7c: Pull complete 
9d62684b94c3: Pull complete 
e70e61e48991: Pull complete 
683f2d0d75c5: Pull complete 
d91684765fac: Pull complete 
ceb6cf7ee657: Pull complete 
8d2533535f88: Pull complete 
15c2061baa94: Pull complete 
fe35706ec086: Pull complete 
ef06e50267e2: Pull complete 
24569ba3e1d3: Pull complete 
c49dc7cbf15c: Pull complete 
34e55507c797: Pull complete 
c26e49a3c2c6: Pull complete 
7f6410878ec9: Pull complete 
97f3bcccbcdf: Pull complete 
3f9a50c314fa: Pull complete 
d6c800c70bb2: Pull complete 
9c785de98406: Pull complete 
acb71385d77d: Pull complete 
ea9fb98cc638: Pull complete 
08e43405860a: Pull complete 
02899df1d7b5: Pull complete 
66e5d0f2b0fa: Pull complete 
46bb7884fc3b: Pull complete 
af50c16f8064: Pull complete 
a8c14d818405: Pull complete 
8c3f313defdf: Pull complete 
Digest: sha256:6614fa29720fc253bcb0e99c29af2f93caff16976661f241ec5ed5cf08e7c010
Status: Downloaded newer image for nvcr.io/nvidia/pytorch:19.05-py3
 ---> 7e98758d4777
Step 3/12 : RUN apt-get update && apt-get install -y --no-install-recommends         infiniband-diags         pciutils &&     rm -rf /var/lib/apt/lists/*
 ---> Running in 7b374edf0b57
Get:1 http://archive.ubuntu.com/ubuntu xenial InRelease [247 kB]
Get:2 http://security.ubuntu.com/ubuntu xenial-security InRelease [109 kB]
Get:3 http://security.ubuntu.com/ubuntu xenial-security/main amd64 Packages [1905 kB]
Get:4 http://archive.ubuntu.com/ubuntu xenial-updates InRelease [109 kB]
Get:5 http://archive.ubuntu.com/ubuntu xenial-backports InRelease [107 kB]
Get:6 http://archive.ubuntu.com/ubuntu xenial/main amd64 Packages [1558 kB]
Get:7 http://archive.ubuntu.com/ubuntu xenial/restricted amd64 Packages [14.1 kB]
Get:8 http://archive.ubuntu.com/ubuntu xenial/universe amd64 Packages [9827 kB]
Get:9 http://security.ubuntu.com/ubuntu xenial-security/restricted amd64 Packages [15.9 kB]
Get:10 http://security.ubuntu.com/ubuntu xenial-security/universe amd64 Packages [982 kB]
Get:11 http://security.ubuntu.com/ubuntu xenial-security/multiverse amd64 Packages [8820 B]
Get:12 http://archive.ubuntu.com/ubuntu xenial/multiverse amd64 Packages [176 kB]
Get:13 http://archive.ubuntu.com/ubuntu xenial-updates/main amd64 Packages [2414 kB]
Get:14 http://archive.ubuntu.com/ubuntu xenial-updates/restricted amd64 Packages [16.4 kB]
Get:15 http://archive.ubuntu.com/ubuntu xenial-updates/universe amd64 Packages [1534 kB]
Get:16 http://archive.ubuntu.com/ubuntu xenial-updates/multiverse amd64 Packages [26.4 kB]
Get:17 http://archive.ubuntu.com/ubuntu xenial-backports/main amd64 Packages [10.9 kB]
Get:18 http://archive.ubuntu.com/ubuntu xenial-backports/universe amd64 Packages [12.6 kB]
Fetched 19.1 MB in 30s (621 kB/s)
Reading package lists...
Reading package lists...
Building dependency tree...
Reading state information...
The following additional packages will be installed:
  libibmad5 libibnetdisc5 libibumad3 libosmcomp3 libpci3
The following NEW packages will be installed:
  infiniband-diags libibmad5 libibnetdisc5 libibumad3 libosmcomp3 libpci3
  pciutils
0 upgraded, 7 newly installed, 0 to remove and 120 not upgraded.
Need to get 574 kB of archives.
After this operation, 2638 kB of additional disk space will be used.
Get:1 http://archive.ubuntu.com/ubuntu xenial-updates/main amd64 libpci3 amd64 1:3.3.1-1.1ubuntu1.3 [24.3 kB]
Get:2 http://archive.ubuntu.com/ubuntu xenial-updates/main amd64 pciutils amd64 1:3.3.1-1.1ubuntu1.3 [254 kB]
Get:3 http://archive.ubuntu.com/ubuntu xenial/universe amd64 libibumad3 amd64 1.3.10.2-1 [16.7 kB]
Get:4 http://archive.ubuntu.com/ubuntu xenial/universe amd64 libibmad5 amd64 1.3.12-1 [29.9 kB]
Get:5 http://archive.ubuntu.com/ubuntu xenial/universe amd64 libosmcomp3 amd64 3.3.19-1 [22.2 kB]
Get:6 http://archive.ubuntu.com/ubuntu xenial/universe amd64 libibnetdisc5 amd64 1.6.6-1 [22.8 kB]
Get:7 http://archive.ubuntu.com/ubuntu xenial/universe amd64 infiniband-diags amd64 1.6.6-1 [205 kB]
debconf: unable to initialize frontend: Dialog
debconf: (TERM is not set, so the dialog frontend is not usable.)
debconf: falling back to frontend: Readline
debconf: unable to initialize frontend: Readline
debconf: (This frontend requires a controlling tty.)
debconf: falling back to frontend: Teletype
dpkg-preconfigure: unable to re-open stdin: 
Fetched 574 kB in 5s (112 kB/s)
Selecting previously unselected package libpci3:amd64.
(Reading database ... 21560 files and directories currently installed.)
Preparing to unpack .../libpci3_1%3a3.3.1-1.1ubuntu1.3_amd64.deb ...
Unpacking libpci3:amd64 (1:3.3.1-1.1ubuntu1.3) ...
Selecting previously unselected package pciutils.
Preparing to unpack .../pciutils_1%3a3.3.1-1.1ubuntu1.3_amd64.deb ...
Unpacking pciutils (1:3.3.1-1.1ubuntu1.3) ...
Selecting previously unselected package libibumad3.
Preparing to unpack .../libibumad3_1.3.10.2-1_amd64.deb ...
Unpacking libibumad3 (1.3.10.2-1) ...
Selecting previously unselected package libibmad5.
Preparing to unpack .../libibmad5_1.3.12-1_amd64.deb ...
Unpacking libibmad5 (1.3.12-1) ...
Selecting previously unselected package libosmcomp3.
Preparing to unpack .../libosmcomp3_3.3.19-1_amd64.deb ...
Unpacking libosmcomp3 (3.3.19-1) ...
Selecting previously unselected package libibnetdisc5.
Preparing to unpack .../libibnetdisc5_1.6.6-1_amd64.deb ...
Unpacking libibnetdisc5 (1.6.6-1) ...
Selecting previously unselected package infiniband-diags.
Preparing to unpack .../infiniband-diags_1.6.6-1_amd64.deb ...
Unpacking infiniband-diags (1.6.6-1) ...
Processing triggers for libc-bin (2.23-0ubuntu11) ...
Setting up libpci3:amd64 (1:3.3.1-1.1ubuntu1.3) ...
Setting up pciutils (1:3.3.1-1.1ubuntu1.3) ...
Setting up libibumad3 (1.3.10.2-1) ...
Setting up libibmad5 (1.3.12-1) ...
Setting up libosmcomp3 (3.3.19-1) ...
Setting up libibnetdisc5 (1.6.6-1) ...
Setting up infiniband-diags (1.6.6-1) ...
Processing triggers for libc-bin (2.23-0ubuntu11) ...
Removing intermediate container 7b374edf0b57
 ---> 91942ef1e039
Step 4/12 : WORKDIR /workspace/rnn_translator
 ---> Running in 150b2d9df1cc
Removing intermediate container 150b2d9df1cc
 ---> 17720ab57857
Step 5/12 : COPY requirements.txt .
 ---> fc25fbdf0006
Step 6/12 : RUN pip install --no-cache-dir https://github.com/mlperf/training/archive/6289993e1e9f0f5c4534336df83ff199bd0cdb75.zip#subdirectory=compliance  && pip install --no-cache-dir -r requirements.txt
 ---> Running in 88b21caded36
Collecting https://github.com/mlperf/training/archive/6289993e1e9f0f5c4534336df83ff199bd0cdb75.zip#subdirectory=compliance
  Downloading https://github.com/mlperf/training/archive/6289993e1e9f0f5c4534336df83ff199bd0cdb75.zip
Building wheels for collected packages: mlperf-compliance
  Building wheel for mlperf-compliance (setup.py): started
  Building wheel for mlperf-compliance (setup.py): finished with status 'done'
  Stored in directory: /tmp/pip-ephem-wheel-cache-c_6ttc8p/wheels/9e/73/0a/3c481ccbda248a195828b8ea5173e83b8394051d8c40e08660
Successfully built mlperf-compliance
Installing collected packages: mlperf-compliance
  Found existing installation: mlperf-compliance 0.0.10
    Uninstalling mlperf-compliance-0.0.10:
      Successfully uninstalled mlperf-compliance-0.0.10
Successfully installed mlperf-compliance-0.6.0
Requirement already satisfied: mlperf-compliance==0.6.0 in /opt/conda/lib/python3.6/site-packages (from -r requirements.txt (line 1)) (0.6.0)
Requirement already satisfied: sacrebleu==1.2.10 in /opt/conda/lib/python3.6/site-packages (from -r requirements.txt (line 2)) (1.2.10)
Requirement already satisfied: typing in /opt/conda/lib/python3.6/site-packages (from sacrebleu==1.2.10->-r requirements.txt (line 2)) (3.6.6)
Removing intermediate container 88b21caded36
 ---> 346646500f0f
Step 7/12 : COPY seq2seq/csrc seq2seq/csrc
 ---> 936e5bc1a41e
Step 8/12 : COPY setup.py .
 ---> 090cc90c4cb5
Step 9/12 : RUN pip install .
 ---> Running in 0547065d6492
Processing /workspace/rnn_translator
Requirement already satisfied: mlperf-compliance==0.6.0 in /opt/conda/lib/python3.6/site-packages (from gnmt==0.6.0) (0.6.0)
Requirement already satisfied: sacrebleu==1.2.10 in /opt/conda/lib/python3.6/site-packages (from gnmt==0.6.0) (1.2.10)
Requirement already satisfied: typing in /opt/conda/lib/python3.6/site-packages (from sacrebleu==1.2.10->gnmt==0.6.0) (3.6.6)
Building wheels for collected packages: gnmt
  Building wheel for gnmt (setup.py): started
  Building wheel for gnmt (setup.py): still running...
  Building wheel for gnmt (setup.py): finished with status 'done'
  Stored in directory: /tmp/pip-ephem-wheel-cache-_jrlxic9/wheels/84/b6/f1/20addc378b275e39e227da5ee58c19f8e2433a88fd6e5fbf7b
Successfully built gnmt
Installing collected packages: gnmt
Successfully installed gnmt-0.6.0
Removing intermediate container 0547065d6492
 ---> 7a7bb07a7855
Step 10/12 : COPY . .
 ---> dfa84645d44d
Step 11/12 : ENV LANG C.UTF-8
 ---> Running in 992046e4ef3e
Removing intermediate container 992046e4ef3e
 ---> d1e6862fe916
Step 12/12 : ENV LC_ALL C.UTF-8
 ---> Running in c67514666b6d
Removing intermediate container c67514666b6d
 ---> 2d4231f91c86
Successfully built 2d4231f91c86
Successfully tagged mlperf-nvidia:rnn_translator
View Code

我们可以查看一下dockfile文件

# Copyright (c) 2018-2019, NVIDIA CORPORATION. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

ARG FROM_IMAGE_NAME=nvcr.io/nvidia/pytorch:19.05-py3
FROM ${FROM_IMAGE_NAME}

# Install dependencies for system configuration logger
RUN apt-get update && apt-get install -y --no-install-recommends \
        infiniband-diags \
        pciutils && \
    rm -rf /var/lib/apt/lists/*

# Install Python dependencies
WORKDIR /workspace/rnn_translator

COPY requirements.txt .
RUN pip install --no-cache-dir https://github.com/mlperf/training/archive/6289993e1e9f0f5c4534336df83ff199bd0cdb75.zip#subdirectory=compliance \
 && pip install --no-cache-dir -r requirements.txt

# Copy & build extensions
COPY seq2seq/csrc seq2seq/csrc
COPY setup.py .
RUN pip install .

# Copy GNMT code
COPY . .

# Configure environment variables
ENV LANG C.UTF-8
ENV LC_ALL C.UTF-8

Docker问题：Dockerfile的From之前不能使用ARG，允许这种用法是在docker 17.05.0-ce (2017-05-04)之后才引入的，查看我的docker版本

[root@2 pytorch]# docker version
Client: Docker Engine - Community
 Version:           20.10.2
 API version:       1.41
 Go version:        go1.13.15
 Git commit:        2291f61
 Built:             Mon Dec 28 16:17:48 2020
 OS/Arch:           linux/amd64
 Context:           default
 Experimental:      true
…………

 对于本测试，将使用config_DGX1.sh并因此将DGXSYTEM指定为DGX1。还要将PULL设置为0，以指示使用本地映像而不是从存储库中提取docker映像。创建了一个新目录“ logs”来存储基准日志文件，并在启动基准运行时提供数据目录路径，如下所示：

[root@2 pytorch]# DATADIR=/data/training_results_v0.6/NVIDIA/benchmarks/gnmt/implementations/data 

[root@2 pytorch]# LOGDIR=/data/training_results_v0.6/NVIDIA/benchmarks/gnmt/implementations/logs 

[root@2 pytorch]# PULL=0 DGXSYSTEM=DGX1 

[root@2 pytorch]# ./run.sub

如果报错以下，表示计算机没有GPU

[root@2 pytorch]# ./run.sub
mlperf-nvidia:rnn_translator
nvidia-docker | 2021/01/25 13:48:39 Error: Could not load UVM kernel module. Is nvidia-modprobe installed?
ERR: Base container launch failed.

查看GPU信息nvidia-smi

如果一切顺利，它将执行基准测试10次，并将日志文件存储在指定目录中。由于在配置文件中指定了8个GPU，因此将看到所有8个GPU被用于训练GNMT模型。可以使用此命令watch -d -n 1 nvidia-smi来定期监视GPU使用情况。

run.sub里也可以设置PULL=0

# Pull latest image
PULL=0
if [[ "${PULL}" != "0" ]]; then
  DOCKERPULL="docker pull $CONT"
  pids=();
  for hostn in ${hosts[@]}; do
    timeout -k 600s 600s \
      $(eval echo $SRUN) $DOCKERPULL &
    pids+=($!);
  done
  wait "${pids[@]}"
  success=$? ; if [ $success -ne 0 ]; then echo "ERR: Image pull failed."; exit $success ; fi
fi

参考链接：https://segmentfault.com/a/1190000022834920?utm_source=tag-newest

《自然语言处理实战入门》 ---- 笔试、面试题：机器学习基础(61-80)
shiter编写程序的艺术
02-25 1万+
51. 假如使用一个较复杂的回归模型来拟合样本数据，使用 Ridge 回归，调试正则化参数 λ，来降低模型复杂度。若 λ 较小时，关于偏差（bias）和方差（variance），下列说法正确的是？ A. 若 λ 较小时，偏差减小，方差减小 B. 若 λ 较小时，偏差减小，方差增大 C. 若 λ 较小时，偏差增大，方差减小 D. 若 λ 较小时，偏差增大，方差增大 答案：B 解析：见 题5。 52. 下列关于 Ridge 回归，说法正确的是（多选）？ A. 若 λ=0，则等价于一般的线性回归 B. 若 λ=.
《自然语言处理实战入门》 ---- 【Generative AI重制版】总目录
shiter编写程序的艺术
12-03 5762
随着人工智能的快速发展，自然语言处理和机器学习技术的应用愈加广泛。为使大家对该领域整体概况有一个系统、明晰的认识，同时入门一些工程实践，也借CSDN为NLP的学习，开发者们搭建一个交流的平台。我希望能够通过这个专栏《自然语言处理实战入门》和广大NLP爱好者一起学习自然语言处理技术，分享学习资料，打破NLP 技术 的实战应用壁垒。................................................
MLPerf Rules总结
HanBlogs
10-24 5313
文章目录1. MLPerf Inference Rules1.1 Inference Division1.2 Test Scenarios1.3 General rules1.4 Data Sets Rules1.5 Model Rules1.6 总结2. MLPerf Submission Rules (Training and Inference)2.1 Submission Result S...
全球权威MLPerf基准测试再发榜，浪潮AI服务器创18项AI性能纪录
CSDN云计算
10-22 540
美国东部时间10月21日，全球备受瞩目的权威AI基准测试MLPerf公布今年的推理测试榜单，浪潮AI服务器NF5488A5一举创造18项性能纪录，在数据中心AI推理性能上遥遥领先其他厂商产品。 MLPerf是当前全球最具影响力的AI计算基准评测组织，由图灵奖得主大卫·帕特森（David Patterson）联合谷歌、斯坦福、哈佛大学等单位共同成立，每年组织全球AI训练和AI推理性能测试并发榜。此次MLPerf的AI推理基准测试有全球23家公司和单位参与，在数据中心及边缘等场景进行AI计算产品的性能比.
MLPerf 基准测试——基础知识笔记
最新发布
weixin_58277783的博客
09-07 4692
介绍了 MLPerf 的相关基础概念和知识，帮助初学者快速理解和使用MLPerf基准测试。MLPerf基准测试是机器学习性能标准基准套件，旨在衡量和比较机器学习推理和训练任务的性能。由一系列基准测试组成，涵盖多个应用领域和硬件平台。
MLPerf inference基准测试步骤——Google Colab篇_onnx
weixin_58277783的博客
01-20 4000
使用MLPerf inference机器学习性能标准基准套件对机器的推理性能进行测试，旨在衡量和比较机器学习推理和训练任务的性能。下面的测试步骤为使用谷歌免费服务器 Google Colab 进行的测试记录。
硬件性能基准测试工具MLPerf、SPEC的特点和应用
易之阴阳，量子纠缠，道之一体，缘起性空
04-16 1271
MLPerf特点：应用：SPEC (Standard Performance Evaluation Corporation)特点：应用：
[Python从零到壹] 十三.机器学习之聚类算法四万字总结全网首发（K-Means、BIRCH、树状聚类、MeanShift）
杨秀璋的专栏
07-06 1万+
欢迎大家来到“Python从零到壹”，在这里我将分享约200篇Python系列文章，带大家一起去学习和玩耍，看看Python这个有趣的世界。所有文章都将结合案例、代码和作者的经验讲解，真心想把自己近十年的编程经验分享给大家，希望对您有所帮助，文章中不足之处也请海涵。Python系列整体框架包括基础语法10篇、网络爬虫30篇、可视化分析10篇、机器学习20篇、大数据分析20篇、图像识别30篇、人工智能40篇、Python安全20篇、其他技巧10篇。您的关注、点赞和转发就是对秀璋最大的支持，知识无价人有情，希望
机器学习笔记 - 机器学习/深度学习实战案例合集
学以致用 知行合一
06-05 2155
案例1、使用opencv的DNN模块调用caffe的GoogLeNet图像分类模型 案例2、使用Opencv的DNN模块调用Tensorflow的mobilenet对象检测模型 案例3、基于tensorflow2.0的手写数字识别，并导出pb模型供OpenCV的C++版本的DNN模块调用 案例4、基于OpenCV图像处理和手写数字识别进行数独求解 案例5、自定义数据集 + 标准前馈神经网络分类进行训练 案例6、自定义数据集 + 卷积神经网络分类进行训练 案例7、使用Opencv的DNN模块调用Caffe的人
〖Python APP 自动化测试实战篇⑩〗- app自动化总结与展望
热门推荐
易编橙 · 终身成长社群，相遇已是上上签！
07-09 4万+
该章节其实没有太多的内容，其实就是针对近期的 App 自动化的一些知识点的总结，以及未来的 App 自动化的未来领域我们可以往哪个方向发展。
MLPerf Benchmarks - MLPerf 基准测试
程永强
10-17 2425
MLPerf Benchmarks - MLPerf 基准测试
今天聊一聊MLPerf 基准测试
C_19870的博客
06-29 2260
其目标是为机器学习研究者和开发者提供一个统一的标准和基准测试工具，以促进机器学习系统的性能提升和相互比较。MLPerf基准测试的核心目标是提供一套公开、验证过的基准测试套件，用于衡量机器学习系统在不同任务上的性能。本文将深入介绍MLPerf基准测试的背景、目标和方法，以及它对于推动机器学习性能评估的重要作用。同时，公开透明的基准测试结果为用户选择合适的机器学习系统提供了依据，推动整个行业的发展和进步。通过提供统一的评估标准和具有代表性的基准测试套件，它促进了机器学习系统的发展和比较，激发了创新和技术进步。
MLPerf 机器学习基准测试实战入门（二）object_detection
蔡军帅
01-25 1142
object_detection使用Mask R-CNN with ResNet50 backbone进行模型训练，参考链接为https://github.com/Caiyishuai/training/tree/master/object_detection。 将MLPerf库拷到本地 mkdir -p mlperf cd mlperf git clone https://github.co...
MLPerf storage基准测试
fakerth的博客
06-03 2645
MLPerf™ 基准测试由来自学术界、研究实验室和行业的 AI 领导者联盟 MLCommons 开发，旨在对硬件、软件和服务的训练和推理性能进行无偏评估。它们都在规定的条件下进行。为了保持在行业趋势的前沿，MLPerf 不断发展，定期举行新的测试，并添加代表 AI 技术水平的新工作负载。每个基准测试套件均由工作组专家社区定义，他们为 AI 系统制定公平的基准测试。工作组定义要运行的 AI 模型、运行模型所依据的数据集、设置允许对模型进行哪些更改的规则，并测量给定硬件运行模型的速度。
MLPerf 机器学习基准测试实战入门（三）recommendation
蔡军帅
02-22 912
该任务对带有神经协同过滤模型的MovieLens 2000万(ml- 2000万)数据集进行隐式反馈的推荐基准。该模型根据关于用户是否与特定物品交互的二进制信息进行训练。 实现参考来自：https://codeload.github.com/mlcommons/training/zip/master 将MLPerf库拷到本地 sudo apt-get install unzip curl ...
MLPerf-train训练评估-gnmt
roymustang的博客
01-26 1117
文章目录准备测试 参考Getting start with MLPerf Hands-on 准备 软件环境 ubuntu 16.04 CUDA10.2 docker nvidia-docker2 测试硬件 特斯拉-v100 测试 首先下载别人的测试结果 test@test-X11DPG-OT:~$ mkdir githubfile test@test-X11DPG-OT:~$ cd githubfile test@test-X11DPG-OT:~/githubfile$ git clone https:
MLPerf Training 参考实现
蔡军帅
02-02 1788
一、MLPerf Training 参考实现 1.MLPerf 简介 MLPerf 官网：https://mlperf.org/ 2020年7月29日，MLPerf联盟发布了MLPerf Training v0.7的结果，这是MLPerf基准测试的第三轮结果。v0.5为第一轮，v0.6是第二轮。MLPerf基准测试是衡量机器学习性能的行业标准，显示了行业的巨大进步和日益增长的多样性，包括多个新...
MLPerf世界纪录技术分享：通过模型压缩优化取得最佳性能
CSDN云计算
06-06 771
本文将重点介绍浪潮在比赛中使用的模型压缩算法的设计思路、实现方式及效果。
Ubuntu14.04安装mlpy（机器学习库）---python
liyuqian199695的博客
01-08 2148
Ubuntu安装mlpy（机器学习库）---python 1、mlpy安装要求 （1）需要先安装配置好以下软件： GCCPython且版本>= 2.6或为 3.XNumPy且版本>= 1.3.0SciPy 且版本>= 0.7.0GSL 且版本>= 1.11 mlpy下载地址：http://sourceforg
评论
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包

打赏作者

蔡军帅

你的鼓励将是我创作的最大动力

¥1 ¥2 ¥4 ¥6 ¥10 ¥20
扫码支付：¥1
获取中
扫码支付

您的余额不足，请更换扫码支付或充值

打赏作者

实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值