$30 off During Our Annual Pro Sale. View Details »
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Rancherでkubeflow構築
Search
nakayamam
March 16, 2019
Technology
3
19k
Rancherでkubeflow構築
Rancher Meetup #07 in Osakaでの発表資料です
nakayamam
March 16, 2019
Tweet
Share
More Decks by nakayamam
See All by nakayamam
rancher/system-toolsを試してみた
nakayamam
0
380
Other Decks in Technology
See All in Technology
RAG/Agent開発のアップデートまとめ
taka0709
0
180
[JAWS-UG 横浜支部 #91]DevOps Agent vs CloudWatch Investigations -比較と実践-
sh_fk2
2
260
MLflowダイエット大作戦
lycorptech_jp
PRO
1
130
文字列の並び順 / Unicode Collation
tmtms
3
590
1人1サービス開発しているチームでのClaudeCodeの使い方
noayaoshiro
0
120
AWS Bedrock AgentCoreで作る 1on1支援AIエージェント 〜Memory × Evaluationsによる実践開発〜
yusukeshimizu
6
400
Python 3.14 Overview
lycorptech_jp
PRO
1
120
モダンデータスタック (MDS) の話とデータ分析が起こすビジネス変革
sutotakeshi
0
500
MLflowで始めるプロンプト管理、評価、最適化
databricksjapan
1
240
AIプラットフォームにおけるMLflowの利用について
lycorptech_jp
PRO
1
150
多様なデジタルアイデンティティを攻撃からどうやって守るのか / 20251212
ayokura
0
460
評価駆動開発で不確実性を制御する - MLflow 3が支えるエージェント開発
databricksjapan
1
190
Featured
See All Featured
Put a Button on it: Removing Barriers to Going Fast.
kastner
60
4.1k
Understanding Cognitive Biases in Performance Measurement
bluesmoon
32
2.7k
Fight the Zombie Pattern Library - RWD Summit 2016
marcelosomers
234
17k
Product Roadmaps are Hard
iamctodd
PRO
55
12k
Practical Tips for Bootstrapping Information Extraction Pipelines
honnibal
25
1.6k
A Modern Web Designer's Workflow
chriscoyier
698
190k
Leading Effective Engineering Teams in the AI Era
addyosmani
8
1.3k
jQuery: Nuts, Bolts and Bling
dougneiner
65
8.3k
How to Think Like a Performance Engineer
csswizardry
28
2.4k
Art, The Web, and Tiny UX
lynnandtonic
303
21k
Rails Girls Zürich Keynote
gr2m
95
14k
Music & Morning Musume
bryan
46
7k
Transcript
RancherͰkubeflowߏங Rancher Meetup #07 in Osaka Masaki-Nakayama
ࣗݾհ • Masaki-Nakayama @nakayamam2 • KAGOYA JAPAN • Rancher Meetup,
CNJP Kansai
kubeflowʁ IUUQTXXXLVCFqPXPSHEPDTBCPVULVCFqPX
kubeflow? KubernetesͷͨΊͷػցֶशπʔϧΩοτ
kubeflow? ԼهͷํʹΦεεϝ by ެࣜ • TensorFlowϞσϧΛ͞·͟·ͳڥʢϩʔΧϧɺΦϯϓ ϨɺΫϥυͳͲʣͰτϨʔχϯά/ఏڙ͍ͨ͠ • TensorFlowτϨʔχϯάδϣϒΛཧ͢ΔͨΊʹJupyter ϊʔτϒοΫΛ͍͍ͨ
• TensorFlowΛଞͷϓϩηεͱΈ߹Θ͍ͤͨ
kubeflow? kubeflowͷμογϡϘʔυ
kubeflow? Լهͷͷ͕͋Β͔͡Ίೖ͍ͬͯ·͢ • JupyterHub : Jupyter NotebookʹϢʔβʔೝূՃͯ͠ෳਓͰ͑ΔΑ ͏ʹͨ͠ͷ • TFjob
Dashboard: k8sͰTensorFlowτϨʔχϯάδϣϒΛཧͰ͖Δ • Katib Dashboard: ϋΠύʔύϥϝʔλʔνϡʔχϯάͷπʔϧ https://www.slideshare.net/Oshima0x3fd/katib ͕ৄͦ͠͏
ߏங·ͰಓͷΓͦ͏ɾɾɾ
͋ΔRancherΧλϩάΛݟ͍ͯΔͱɾɾ
͋ʂ
None
͔ͯ͠͠Chainer͙͑͢Δɾɾɾʁ
ྲྀΕ 1. GPUΫϥελʔͷߏங on GKE 2. ΫϥελʔΛRancherΠϯϙʔτ 3. RancherΧλϩάͰkubeflowσϓϩΠ ※ͪͳΈʹGKEʹೖΕΔ͚ͩͳΒઐ༻ͷϫϯΫϦοΫσϓϩΠ͕༻ҙ͞ΕͯΔͷͰ
ͦͪΒΛ͏ํ͕ૣ͍͔ https://deploy.kubeflow.cloud/#/
GPUΫϥελʔͷߏங on GKE
GPUબΜͰ࡞ KZVQZUFSIVC͕Ϧιʔε Λ৯͏ͷͰεϖοΫ͕͋ Μ·Γ͍ͱࢮʹ·͢
GPUΫϥελʔ͕Ͱ͖·ͨ͠
ΫϥελʔΛ RancherΠϯϙʔτ
None
ࣗݾॺ໊ͳͷͰͪ͜Β Λ࣮ߦ DMVTUFSBENJOΛϢʔβʔ ʹCJOEJOH
ΠϯϙʔτͰ͖·ͨ͠
RancherΧλϩάͰ kubeflowσϓϩΠ
None
֤ػೳͷ0/0''͕Ͱ͖ΔͬΆ͍ʁ
֤ػೳͷ0/0''͕Ͱ͖ΔͬΆ͍ʁ
None
6*
None
Ϣʔβʔ࡞͢Δ
ΠϝʔδΛࢦఆͯ͠TQBXO
ϫʔΫεϖʔε͕࡞͞ΕΔ
ϫʔΫεϖʔε͕ग़དྷ্͕Δ
QZUIPOίʔυΛ͙࣮͢ߦͰ͖Δ
None
UFOTPSqPXͷδϣϒͷ࡞͕Ͱ͖Δ
> kubectl get crd NAME AGE backendconfigs.cloud.google.com 2h scalingpolicies.scalingpolicy.kope.io 2h
studyjobs.kubeflow.org 1h tfjobs.kubeflow.org 1h ͋Εɺchainer operator͕͍ͳ͍ɾɾɾ
None
{"log":"2019/03/14 17:57:17 info: manifest \"kubeflow/templates/ chainer-rbac.yaml\" is empty. Skipping. \n","stream":"stderr","time":"2019-03-14T17:57:17.702047111Z"}
{"log":"2019/03/14 17:57:17 info: manifest \"kubeflow/templates/ chainer-crd.yaml\" is empty. Skipping. \n","stream":"stderr","time":"2019-03-14T17:57:17.702898257Z"} {"log":"2019/03/14 17:57:17 info: manifest \"kubeflow/templates/ chainer-operator.yaml\" is empty. Skipping. \n","stream":"stderr","time":"2019-03-14T17:57:17.702904447Z"} RancherαʔόʔͷϩάΛݟΔͱ Ͳ͏manifestʹө͞Ε͍ͯͳ͍Α͏ͩ
issueग़͠ͱ͖·ͨ͠ʂ
Thanks!