Lokhu okuthunyelwe kwesivakashi okubhalwe kanye nethimba labaholi be-Iambic Therapeutics.
I-Iambic Therapeutics kuyisiqalo sokutholwa kwezidakamizwa esinomgomo wokudala ubuchwepheshe obusha obuqhutshwa yi-AI ukuletha imithi engcono ezigulini ezinomdlavuza, ngokushesha.
Amathuluzi ethu obuhlakani bokwenziwa obuthuthukisiwe (AI) asenza sikwazi ukusesha indawo enkulu yama-molecule ezidakamizwa angenzeka ngokushesha nangempumelelo. Ubuchwepheshe bethu buyahlukahluka futhi busebenza kuzo zonke izindawo zokwelapha, izigaba zamaprotheni, kanye nezindlela zokusebenza. Ngaphandle kokudala amathuluzi e-AI ahlukene, sisungule inkundla edidiyelwe ehlanganisa isofthiwe ye-AI, idatha esekelwe emafini, ingqalasizinda yokubala esheshayo, kanye nekhono eliphezulu lekhemistri nebhayoloji. Inkundla yomibili inika amandla i-AI yethu—ngokuhlinzeka ngedatha ukuze icwengisise amamodeli ethu—futhi inikwe amandla yikho, isebenzise amathuba okwenza izinqumo okuzenzakalelayo nokucubungula idatha.
Sikala impumelelo ngekhono lethu lokukhiqiza abantu abasezingeni eliphezulu basemtholampilo ukuze sibhekane nesidingo esiphuthumayo sesiguli, ngesivinini esingakaze sibonwe ngaphambili: sithuthuke sisuka ekwethulweni kohlelo saya kwabahlolwayo basemtholampilo ezinyangeni nje ezingama-24, ngokushesha okukhulu ukwedlula labo esiqhudelana nabo.
Kulokhu okuthunyelwe, sigxila endleleni esisebenzisa ngayo Karpenter on Isevisi ye-Amazon Elastic Kubernetes (I-Amazon EKS) ukukala ukuqeqeshwa kwe-AI kanye nencazelo, okuyizici ezibalulekile zeplathifomu yokuthola i-Iambic.
Isidingo sokuqeqeshwa kwe-AI okunokwethenjelwa kanye nencazelo
Isonto ngalinye, i-Iambic yenza ukubikezela kwe-AI kuwo wonke amamodeli nezigidi zama-molecule, ihlinzeka ngamacala amabili okusetshenziswa okuyinhloko:
- Osokhemisi bezokwelapha nabanye ososayensi basebenzisa uhlelo lwethu lwewebhu, i-Insight, ukuhlola indawo yamakhemikhali, ukufinyelela nokuhumusha idatha yokuhlola, nokubikezela izici zama-molecule asanda kuklanywa. Wonke lo msebenzi wenziwa ngokuhlanganyela ngesikhathi sangempela, okwenza kube nesidingo sokunquma nge-latency ephansi kanye nokuphuma okuphakathi.
- Ngasikhathi sinye, amamodeli ethu e-AI akhiqizayo aklama ngokuzenzakalelayo ama-molecule aqondise ukuthuthuka kuzo zonke izakhiwo eziningi, asesha izigidi zabantu abazongenela ukhetho, futhi adinga ukusebenza okukhulu nokubambezeleka okumaphakathi.
Iqondiswa ubuchwepheshe be-AI kanye nabazingeli bezidakamizwa abangongoti, inkundla yethu yokuhlola ikhiqiza izinkulungwane zamamolekyuli ahlukile isonto ngalinye, futhi ngalinye libhekana nokuhlolwa okuningi kwebhayoloji. Amaphoyinti edatha akhiqiziwe acutshungulwa ngokuzenzakalelayo futhi asetshenziselwa ukushuna kahle amamodeli ethu e-AI masonto onke. Ekuqaleni, ukulungisa kahle imodeli yethu kuthathe amahora amaningi esikhathi se-CPU, ngakho-ke uhlaka lokukala imodeli yokushuna kahle kuma-GPU belubalulekile.
Amamodeli ethu okufunda okujulile anezidingo ezingezona ezincane: angamagigabhayithi ngosayizi, maningi futhi awafani, futhi adinga ama-GPU ukuze aqonde ngokushesha futhi alungise kahle. Uma sibheka nengqalasizinda yamafu, sidinga uhlelo olusivumela ukuthi sifinyelele kuma-GPU, sikhuphuke futhi sehle ngokushesha ukuze sibambe imithwalo yemisebenzi eqinile, ehlukahlukene, futhi sisebenzise izithombe ezinkulu ze-Docker.
Besifuna ukwakha isistimu enwebekayo ukusekela ukuqeqeshwa kwe-AI kanye nencazelo. Sisebenzisa i-Amazon EKS futhi besifuna isixazululo esingcono kakhulu sokukala ngokuzenzakalelayo izindawo zethu zabasebenzi. Sikhethe i-Karpenter ye-Kubernetes node scaling ngezizathu ezimbalwa:
- Ukuhlanganiswa kalula ne-Kubernetes, kusetshenziswa i-Kubernetes semantics ukuchaza izidingo ze-node kanye nemininingwane ye-pod yokukala
- Ukubambezeleka okuphansi kokuphuma kwama-node
- Ukuhlanganiswa kalula nengqalasizinda yethu njengethuluzi lekhodi (Terraform)
Abahlinzeki be-node basekela ukuhlanganiswa okungenamsebenzi ne-Amazon EKS nezinye izinsiza ze-AWS ezifana I-Amazon Elastic Compute Cloud (Amazon EC2) izimo kanye Isitolo se-Amazon Elastic Block imiqulu. Ama-semantics e-Kubernetes asetshenziswa abahlinzeki asekela ukuhlela okuqondisiwe kusetshenziswa ukwakhiwa kwe-Kubernetes okufana nokugcotshwa noma ukubekezelela kanye nokucaciswa okungahambisani nokuhambisana; futhi zisiza ukulawula inombolo nezinhlobo zezimo ze-GPU ezingahle zihlelwe ngu-Karpenter.
Ukubukwa kwesisombululo
Kulesi sigaba, sethula isakhiwo esijwayelekile esifana nalesi esisisebenzisela imithwalo yethu yomsebenzi, evumela ukuthunyelwa okunwebekayo kwamamodeli kusetshenziswa ukukala okuzenzakalelayo okuphumelelayo okusekelwe kumamethrikhi angokwezifiso.
Umdwebo olandelayo ubonisa isakhiwo sesixazululo.
I-architecture isebenzisa a isevisi elula ku-Kubernetes pod ngaphakathi kwe- Iqoqo le-EKS. Lokhu kungaba imodeli ecatshangwayo, ukulingisa idatha, nanoma iyiphi enye isevisi efakwe esitsheni, efinyeleleka ngesicelo se-HTTP. Isevisi ivezwa ngemuva kwe-reverse-proxy kusetshenziswa Traefik. Ummeleli obuyela emuva uqoqa amamethrikhi mayelana namakholi aya kusevisi futhi awadalule nge-API ejwayelekile yamamethrikhi ukuze Prometheus. I-Kubernetes Event Driven Autoscaler (KEDA) ilungiselelwe ukukala ngokuzenzakalelayo inombolo yamaphodi wesevisi, ngokusekelwe kumamethrikhi angokwezifiso atholakala ku-Prometheus. Lapha sisebenzisa inombolo yezicelo ngesekhondi ngalinye njengemethrikhi yangokwezifiso. Indlela efanayo yezakhiwo iyasebenza uma ukhetha imethrikhi ehlukile yomthwalo wakho womsebenzi.
I-Karpenter iqapha noma imaphi ama-pod alindile angakwazi ukusebenza ngenxa yokuntuleka kwezinsiza ezanele kuqoqo. Uma ama-pods anjalo etholwa, u-Karpenter wengeza ama-node engeziwe kuqoqo ukuze anikeze izinsiza ezidingekayo. Ngokuphambene, uma kukhona ama-node amaningi ku-cluster kunalokho okudingekayo ngama-pods ahleliwe, u-Karpenter ususa amanye ama-node abasebenzi futhi ama-pods ahlelwe kabusha, awahlanganise ezimweni ezimbalwa. Inombolo yezicelo ze-HTTP ngomzuzwana kanye nenombolo yamanodi ingabonwa kusetshenziswa a UGrafana ideshibhodi. Ukubonisa ukukala okuzenzakalelayo, sisebenzisa okukodwa noma ngaphezulu ama-pods alula okukhiqiza umthwalo, ethumela izicelo ze-HTTP kusevisi isebenzisa i-curl.
Ukuthunyelwa kwesixazululo
In the isinyathelo ngesinyathelo sokuhamba, sisebenzisa I-AWS Cloud9 njengendawo yokubeka i-architecture. Lokhu kwenza zonke izinyathelo ziqedwe kusuka kusiphequluli sewebhu. Ungaphinda usebenzise isixazululo kukhompuyutha yasendaweni noma isibonelo se-EC2.
Ukwenza ukusetshenziswa kube lula futhi kuthuthukiswe ukuphindaphindeka, silandela imigomo ye yenza-uhlaka kanye nesakhiwo se- isifanekiso se-dependent-docker. Sihlanganisa i awu-do-eks iphrojekthi futhi, usebenzisa Docker, sakha isithombe sesitsha esifakwe amathuluzi adingekayo kanye nemibhalo. Ngaphakathi kwesiqukathi, sigijima kuzo zonke izinyathelo zokuya ekupheleni ukuya ekupheleni, kusukela ekudaleni iqoqo le-EKS ne-Karpenter ukuya ekulinganiseni. Izibonelo ze-EC2.
Isibonelo kulokhu okuthunyelwe, sisebenzisa okulandelayo I-manifest yeqoqo le-EKS:
apiVersion: eksctl.io/v1alpha5
kind: ClusterConfig
metadata:
name: do-eks-yaml-karpenter
version: '1.28'
region: us-west-2
tags:
karpenter.sh/discovery: do-eks-yaml-karpenter
iam:
withOIDC: true
addons:
- name: aws-ebs-csi-driver
version: v1.26.0-eksbuild.1
wellKnownPolicies:
ebsCSIController: true
managedNodeGroups:
- name: c5-xl-do-eks-karpenter-ng
instanceType: c5.xlarge
instancePrefix: c5-xl
privateNetworking: true
minSize: 0
desiredCapacity: 2
maxSize: 10
volumeSize: 300
iam:
withAddonPolicies:
cloudWatch: true
ebs: trueLe-manifest ichaza iqoqo eliqanjwe igama do-eks-yaml-karpenter ngomshayeli we-EBS CSI efakwe njengesengezo. Iqembu le-node eliphethwe elinababili c5.xlarge ama-node afakiwe ukuze aqhube ama-pods esistimu adingwa yiqoqo. Amanodi ezisebenzi asingathwe kuma-subnet ayimfihlo, futhi indawo yokugcina ye-API yeqoqo isesidlangalaleni ngokuzenzakalela.
Ungasebenzisa futhi iqoqo le-EKS elikhona esikhundleni sokulidala. Sihambisa i-Karpenter ngokulandela i- imiyalelo emibhalweni ye-Karpenter, noma ngokusebenzisa okulandelayo iskripthi, okwenza imiyalelo yokusebenzisa ngokuzenzakalelayo.
Ikhodi elandelayo ibonisa ukucushwa kwe-Karpenter esikusebenzisa kulesi sibonelo:
apiVersion: karpenter.sh/v1beta1
kind: NodePool
metadata:
name: default
spec:
template:
metadata: null
labels:
cluster-name: do-eks-yaml-karpenter
annotations:
purpose: karpenter-example
spec:
nodeClassRef:
apiVersion: karpenter.k8s.aws/v1beta1
kind: EC2NodeClass
name: default
requirements:
- key: karpenter.sh/capacity-type
operator: In
values:
- spot
- on-demand
- key: karpenter.k8s.aws/instance-category
operator: In
values:
- c
- m
- r
- g
- p
- key: karpenter.k8s.aws/instance-generation
operator: Gt
values:
- '2'
disruption:
consolidationPolicy: WhenUnderutilized
#consolidationPolicy: WhenEmpty
#consolidateAfter: 30s
expireAfter: 720h
---
apiVersion: karpenter.k8s.aws/v1beta1
kind: EC2NodeClass
metadata:
name: default
spec:
amiFamily: AL2
subnetSelectorTerms:
- tags:
karpenter.sh/discovery: "do-eks-yaml-karpenter"
securityGroupSelectorTerms:
- tags:
karpenter.sh/discovery: "do-eks-yaml-karpenter"
role: "KarpenterNodeRole-do-eks-yaml-karpenter"
tags:
app: autoscaling-test
blockDeviceMappings:
- deviceName: /dev/xvda
ebs:
volumeSize: 80Gi
volumeType: gp3
iops: 10000
deleteOnTermination: true
throughput: 125
detailedMonitoring: trueSichaza i-Karpenter NodePool ezenzakalelayo enezidingo ezilandelayo:
- I-Karpenter ingaqala izimo kusuka kokubili
spotfuthion-demandamachibi omthamo - Izimo kumele zisuke ku-"
c” (ukubala kulungiselelwe), “m” (inhloso jikelele), “r” (inkumbulo yenziwe kahle), noma “g"Futhi"p” (i-GPU isheshisiwe) ikhompuyutha imindeni - Ukwenziwa kwezehlakalo kufanele kube kukhulu kuno-2; Ngokwesibonelo,
g3kuyamukeleka, kodwag2ayiyona
I-NodePool ezenzakalelayo iphinda ichaze izinqubomgomo zokuphazamiseka. Amanodi angasetshenziswa kancane azokhishwa ukuze ama-pods ahlanganiswe ukuze asebenze kumanodi ambalwa noma amancane. Kungenjalo, singamisa amanodi angenalutho ukuze akhishwe ngemva kwesikhathi esishiwo. I expireAfter ukulungiselelwa kucacisa isikhathi esiphezulu sokuphila kwanoma iyiphi i-node, ngaphambi kokuthi imiswe futhi ishintshwe uma kunesidingo. Lokhu kusiza ukunciphisa ubungozi bokuphepha kanye nokugwema izinkinga ezijwayelekile kumanodi anesikhathi eside, njengokuhlukana kwefayela noma ukuvuza kwenkumbulo.
Ngokuzenzakalelayo, i-Karpenter inikeza amanodi anevolumu encane yempande, enganele ekusebenziseni i-AI noma imithwalo yomsebenzi yokufunda ngomshini (ML). Ezinye zezithombe zeziqukathi zokufunda ezijulile zingaba amashumi ama-GB ngosayizi, futhi sidinga ukwenza isiqiniseko sokuthi kunesikhala esanele sokulondoloza kumanodi ukuze siqhube ama-pods sisebenzisa lezi zithombe. Ukwenza lokho, sichaza EC2NodeClass nge blockDeviceMappings, njengoba kukhonjisiwe kukhodi eyandulele.
U-Karpenter unesibopho sokukala okuzenzakalelayo ezingeni leqoqo. Ukuze ulungiselele ukukala okuzenzakalelayo ezingeni le-pod, sisebenzisa i-KEDA ukuze sichaze insiza yangokwezifiso ebizwa ScaledObject, njengoba kukhonjisiwe kukhodi elandelayo:
apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
name: keda-prometheus-hpa
namespace: hpa-example
spec:
scaleTargetRef:
name: php-apache
minReplicaCount: 1
cooldownPeriod: 30
triggers:
- type: prometheus
metadata:
serverAddress: http://prometheus- server.prometheus.svc.cluster.local:80
metricName: http_requests_total
threshold: '1'
query: rate(traefik_service_requests_total{service="hpa-example-php-apache-80@kubernetes",code="200"}[2m])I-manifest eyandulele ichaza a ScaledObject okuthiwa keda-prometheus-hpa, enesibopho sokukala ukuthunyelwa kwe-php-apache futhi njalo igcina okungenani isifaniso esisodwa sisebenza. Ikala ama-pods alokhu kuthunyelwa ngokusekelwe kumethrikhi http_requests_total etholakala ku-Prometheus etholwe ngombuzo oshiwo, futhi kuhloswe ukukhuphula ama-pods ukuze i-pod ngayinye inikeze isicelo esingaphezu kwesisodwa ngomzuzwana. Yehlisa izifaniso ngemva kokuba umthwalo wesicelo ube ngaphansi komkhawulo isikhathi eside kunamasekhondi angu-30.
The i-deployment spec ngokwesibonelo isevisi yethu iqukethe okulandelayo izicelo zezinsiza kanye nemikhawulo:
resources:
limits:
cpu: 500m
nvidia.com/gpu: 1
requests:
cpu: 200m
nvidia.com/gpu: 1Ngalokhu kumisa, i-pod ngayinye yesevisi izosebenzisa i-NVIDIA GPU eyodwa. Uma ama-pod amasha edalwa, azoba sesimweni sokulinda kuze kutholakale i-GPU. U-Karpenter wengeza amanodi e-GPU kuqoqo njengoba kudingeka ukuze kuhlaliswe ama-pods alindile.
A i-pod ekhiqiza umthwalo ithumela izicelo ze-HTTP kusevisi ngefrikhwensi esethwe ngaphambilini. Sandisa inani lezicelo ngokwandisa inani lezifaniso ku- ukuthunyelwa kwejeneretha yokulayisha.
Umjikelezo wokukala ogcwele ngokuhlanganisa i-node esekelwe ekusetshenzisweni ubonakala ngeso lengqondo kudeshibhodi ye-Grafana. Ideshibhodi elandelayo ibonisa inani lamanodi kuqoqo ngokohlobo lwesibonelo (phezulu), inani lezicelo ngomzuzwana (phansi kwesokunxele), kanye nenani lamaphodi (phansi kwesokudla).
Siqala ngezimo ezimbili nje ze-c5.xlarge CPU iqoqo elakhiwe ngazo. Bese sikhipha isibonelo sesevisi esisodwa, esidinga i-GPU eyodwa. U-Karpenter wengeza isibonelo se-g4dn.xlarge ukuze kuhlangatshezwane nalesi sidingo. Sibe sesihambisa ijeneretha yokulayisha, okubangela ukuthi i-KEDA ingeze ama-pod wesevisi futhi u-Karpenter wengeza izimo ze-GPU ezengeziwe. Ngemva kokuthuthukisa, izwe lihlala endaweni eyodwa ye-p3.8xlarge enama-GPU angu-8 kanye nesenzakalo esisodwa se-g5.12xlarge esinama-GPU angu-4.
Uma sikala ukuthunyelwa okukhiqiza umthwalo kuma-replicas angu-40, i-KEDA idala ama-pods esevisi eyengeziwe ukuze kugcinwe umthwalo wesicelo odingekayo nge-pod ngayinye. U-Karpenter wengeza amanodi e-g4dn.metal kanye ne-g4dn.12xlarge kuqoqo ukuze anikeze ama-GPU adingekayo kuma-pods engeziwe. Esimeni esilinganiselwe, iqoqo liqukethe ama-node we-GPU angu-16 futhi linikeza izicelo ezingaba ngu-300 ngomzuzwana. Uma sehlisa ijeneretha yokulayisha ibe yi-replica engu-1, inqubo yokuhlehla iyenzeka. Ngemuva kwesikhathi sokupholisa, i-KEDA yehlisa inani lamaphodi esevisi. Khona-ke njengoba ama-pods ambalwa egijima, u-Karpenter ususa ama-node angasetshenziswa kancane eqenjini futhi ama-pods wesevisi ayahlanganiswa ukuze asebenze kuma-node ambalwa. Lapho i-pod yokukhiqiza umthwalo isuswa, i-pod yesevisi eyodwa kusenzakalo esisodwa se-g4dn.xlarge ene-1 GPU ihlala isebenza. Uma sisusa i-pod yesevisi futhi, iqoqo lishiywa esimweni sokuqala ngamanodi amabili we-CPU kuphela.
Singakubona lokhu kuziphatha lapho NodePool inesilungiselelo consolidationPolicy: WhenUnderutilized.
Ngalesi silungiselelo, i-Karpenter imisa ngokuguquguqukayo iqoqo elinamanodi ambalwa ngangokunokwenzeka, kuyilapho ihlinzeka ngezinsiza ezanele ukuze wonke ama-pods asebenze futhi futhi yehlisa izindleko.
Ukuziphatha kokukala okuboniswe kudeshibhodi elandelayo kuyabonwa lapho i NodePool inqubomgomo yokuhlanganisa isethwe ukuze WhenEmpty, kanye consolidateAfter: 30s.
Kulesi simo, ama-node ayamiswa kuphela uma kungekho ama-pods agijima kuwo ngemva kwesikhathi sokupholisa. Ijika lokukala libonakala libushelelezi, uma liqhathaniswa nenqubomgomo yokuhlanganisa esekelwe ekusetshenzisweni; nokho, kungabonakala ukuthi amanodi amaningi asetshenziswa esimweni esilinganisiwe (22 vs. 16).
Sekukonke, ukuhlanganisa i-pod ne-cluster auto scaling kuqinisekisa ukuthi iqoqo lilinganisa ngokuguquguqukayo nomthwalo womsebenzi, labe izinsiza lapho kudingekile futhi lizisuse lapho zingasetshenziswa, ngaleyo ndlela kukhulisa ukusetshenziswa nokunciphisa izindleko.
imiPhumela
I-Iambic isebenzise lesi sakhiwo ukunika amandla ukusetshenziswa kahle kwama-GPU ku-AWS futhi ithuthe imithwalo yomsebenzi isuka ku-CPU iye ku-GPU. Ngokusebenzisa izimo ezinamandla ze-EC2 GPU, i-Amazon EKS, ne-Karpenter, sikwazile ukunika amandla ukucatshangwa okusheshayo kwamamodeli ethu asekelwe ku-physics kanye nezikhathi zokuhlola okusheshayo kososayensi abasetshenzisiwe abathembele ekuqeqeshweni njengesevisi.
Ithebula elilandelayo lifingqa amanye amamethrikhi esikhathi alokhu kufuduka.
| Umsebenzi | Ama-CPU | Ama-GPU |
| Ukucabanga kusetshenziswa amamodeli okusabalalisa amamodeli e-ML asuselwa ku-physics | 3,600 imizuzwana |
100 imizuzwana (ngenxa yokuqoqwa kwemvelo kwama-GPU) |
| Ukuqeqeshwa kwemodeli ye-ML njengesevisi | 180 imizuzu | 4 imizuzu |
Ithebula elilandelayo lifingqa ezinye zesikhathi sethu namamethrikhi ezindleko.
| Umsebenzi | Ukusebenza/Izindleko | |
| Ama-CPU | Ama-GPU | |
| Ukuqeqeshwa kwemodeli ye-ML |
240 imizuzu isilinganiso esingu-$0.70 ngomsebenzi ngamunye wokuqeqesha |
20 imizuzu isilinganiso esingu-$0.38 ngomsebenzi ngamunye wokuqeqesha |
Isifinyezo
Kulokhu okuthunyelwe, sibonise ukuthi i-Iambic isebenzise kanjani i-Karpenter ne-KEDA ukukala ingqalasizinda yethu ye-Amazon EKS ukuze ihlangabezane nezidingo ze-latency zokuchazwa kwe-AI yethu kanye nemithwalo yemisebenzi yokuqeqesha. I-Karpenter ne-KEDA zingamathuluzi anamandla omthombo ovulekile asiza ukukala ngokuzenzakalelayo amaqoqo e-EKS nemithwalo yomsebenzi esebenza kuwo. Lokhu kusiza ukukhulisa izindleko zokubala ngenkathi uhlangabezana nezimfuneko zokusebenza. Ungahlola ikhodi futhi usebenzise izakhiwo ezifanayo endaweni yakho ngokulandela uhambo oluphelele kulokhu. GitHub repo.
Mayelana Ababhali
Matthew Welborn ungumqondisi we-Machine Learning e-Iambic Therapeutics. Yena nethimba lakhe basebenzisa i-AI ukuze kusheshiswe ukuhlonzwa nokuthuthukiswa kwezindlela zokwelapha ezintsha, okuletha imithi esindisa impilo ezigulini ngokushesha.
Paul Whittemore ungunjiniyela Omkhulu e-Iambic Therapeutics. Usekela ukulethwa kwengqalasizinda yenkundla yokutholwa kwezidakamizwa eqhutshwa yi-Iambic AI.
Alex Iankoulski ungumakhi Oyinhloko Wezixazululo, i-ML/AI Frameworks, egxile ekusizeni amakhasimende ukuhlela umthwalo wawo we-AI esebenzisa iziqukathi kanye nengqalasizinda yekhompuyutha esheshisiwe ku-AWS.
- I-SEO Powered Content & PR Distribution. Khuliswa Namuhla.
- I-PlatoData.Network Vertical Generative Ai. Zinike Amandla. Finyelela Lapha.
- I-PlatoAiStream. I-Web3 Intelligence. Ulwazi Lukhulisiwe. Finyelela Lapha.
- I-PlatoESG. Ikhabhoni, I-CleanTech, Amandla, Environment, Ilanga, Ukuphathwa Kwemfucuza. Finyelela Lapha.
- I-PlatoHealth. I-Biotech kanye ne-Clinical Trials Intelligence. Finyelela Lapha.
- Source: https://aws.amazon.com/blogs/machine-learning/scale-ai-training-and-inference-for-drug-discovery-through-amazon-eks-and-karpenter/
