HAGAHA Farsamada
U-dejinta Gobolka Optimizer ee CPU iyo NVMe
Khiyaamo kaydinta xusuusta oo dhigata kaydinta culus ee tababarka (gobolada hagaajinta, gradients, mararka qaarkood miisaan) CPU RAM ama NVMe SSDs halkii ay ka ahaan lahaayeen xusuusta GPU-da yar.
Dhageyso hagahanAI-waa laga sheekeeyay · ~2 min
Boggaan2 daqiiqo akhri
Dulmar
It lets people train far larger models than their GPU's memory would otherwise allow.
quusid qoto dheer
Markaad tababarto shabakad neural ah oo leh hagaajin sida Adam oo kale, halbeeg kastaa wuxuu sido shandad dheeri ah: laba tirakoob oo socda (xoog iyo kala duwanaansho), oo lagu daray koobi sax ah oo miisaan ah, oo lagu daray gradientkiisa. Tababarka saxda ah ee isku dhafan tani waxay wadarta guud ahaan kartaa 16 bytes halkiibeegba, iyada oo ka dhigaysa 2 bytes miisaanka laftiisa. Soo dejintu waxay ka rarisaa alaabtaas GPU-da. CPU offload durdurrada kor u qaadaya gobolada nidaamka caadiga ah RAM ee baska PCIe, halka NVMe offload iyaga u riixdo ilaa dariiqa hoos si degdeg ah-state disks. Waxaa caan ku ah DeepSpeed's ZeRO-Infinity iyo ZeRO-Offload, farsamadu waxay ka baayacmushtareysaa xawaaraha ceyriinka ah ee awoodda, taasoo u oggolaanaysa hal GPU ama koox yar oo qaab-qaabeeya oo leh balaayiin cabbirro ah.
Dhaqangelinta Adduunka-dhabta ah
DeepSpeed’s ZeRO-Offload tutorial trains a ten-billion-parameter GPT-2 configuration on a 32 GB V100, with CPU optimizer offloading and activation checkpointing.
A research team evaluates ZeRO-Infinity with CPU and NVMe tiers for a model that exceeds GPU-only capacity.
A fine-tuning team configures CPU optimizer offloading and measures host memory use and update time.
A team compares larger GPUs with offloading, accounting for CPU, storage, and transfer costs.
Mustaqbalka Wax-ka-qabashada Gobolka ee CPU iyo NVMe
Sida moodellada ay uga sii kordhayaan xusuusta GPU-da, soo-dejintu waxay noqonaysaa heer aan caadi ahayn. Filo is dhexgalka adag oo leh isku xirnaanta dhaqsaha badan sida NVLink-C2C iyo barkadaha xusuusta ee CXL ee xumeeya soohdinta CPU-GPU, oo lagu daray jadwalayaal caqli badan oo saadaaliya gobolada la hor marinayo. Nashqada xusuusta midaysan sida Grace Hopper waxay yaraynaysaa ciqaabta PCIe, qaab-dhismeedyaduna waxay riixayaan samaynta soo dejinta heerarka kala duwan ee ku dhawaad daah-furnaanta si ay hiwaayadaha u hagaajiyaan moodooyinka waaweyn ee qalabka suubban.
Ku dhaqangelinta
Maxay muhiim u tahay
- Qiimaha iyo miisaaniyada. Go'aamada qaab-dhismeedku waxay horseedaan waxqabadka iyo kharashka hawlgalka sannadaha.
- Go'aamo cad. Waxbarashada farsamada waxay ka caawisaa kooxaha inay doortaan xidhmo sax ah, ma aha oo kaliya kan ugu cusub.
- Xakamaynta tayada. Doorashooyinka injineernimada ee wanaagsan waxay yareeyaan shilalka la isku halleyn karo ee wax soo saarka.
Iska ilaali
- Hagaajinta hal bartilmaameed waxay qarin kartaa daciifnimada nidaamka ballaaran.
- Kaabayaasha dhaqaalaha iyo dayactirka inta badan waa la dhayalsadaa.
- Nabadgelyada iyo daldaloolada u fiirsashada ayaa kori kara marka nidaamyadu noqdaan kuwo aad u adag.
Tallaabooyinka xiga
- Qeex daahida, tayada, iyo bartilmaameedyada qiimaha ka hor inta aan la hirgelin.
- Benchmark marka la eego culeyska dhabta ah iyo xaaladaha xogta.
- La socodka qalabka khaladaadka, leexashada, iyo saamaynta isticmaalaha.
- U diyaari dib-u-noqoshada iyo dariiqyada jawaab-celinta dhacdada ka hor inta aanad miisaan.
Sii wad Sahaminta
Warqad bilaash ah
Hel warbixinta maalinlaha ah ee AI
Saddex sheeko oo la xaqiijiyay oo AI ah maalin kasta oo shaqo ah, oo lagu qoray Ingiriis fudud. Bilaash had iyo jeer, aan xayaysiis lahayn.
Hal iimayl maalin kasta oo shaqo ah. Ka bax hal gujin. Marna ma iibinno ama ma wadaagno ciwaankaaga.
Is tijaabi
Take the Optimizer State Offloading to CPU and NVMe quiz
Fikrad degdeg ah oo ku saabsan jawaab kasta, iyo shahaado la wadaajin karo oo leh aqoonsi la xaqiijin karo markaad ka gudubto koorso.
Taageer waxbarashada AI bilaashka ah. AI Understanding waa 501(c)(3) aan faa'iido doon ahayn — xayaysiis la'aan, albaab lacag ah ma jiro, had iyo jeer. Samee deeq
Su'aalaha soo noqnoqda
What is Optimizer State Offloading to CPU and NVMe?
Khiyaamo kaydinta xusuusta oo dhigata kaydinta culus ee tababarka (gobolada hagaajinta, gradients, mararka qaarkood miisaan) CPU RAM ama NVMe SSDs halkii ay ka ahaan lahaayeen xusuusta GPU-da yar. Waxay u ogolaataa dadka inay tababaraan moodallo aad uga weyn kuwa xusuusta GPU-da ay si kale u oggolaan karto.
Waa maxay himilada koowaad ee u dajinta dawladaha kor u qaadaya CPU ama NVMe?
Soo dejintu waxay u dhaqaajisaa gobolada wanaajiyaha culus ee GPU-ga CPU RAM ama NVMe, iyada oo xoraynaysa xusuusta GPU si moodooyinka waaweyn loogu tababaro isla qalab isku mid ah.
For an Adam setup with two FP32 moments and an FP32 master weight, how much memory do those three values use per parameter?
Each FP32 value uses four bytes: two moments plus one master weight total twelve. A sixteen-byte mixed-precision estimate includes other tensors as well.
Qaabkee ayaa sifeynaya soo dejin heersare ah oo caan baxay?
DeepSpeed's ZeRO-Offload iyo ZeRO-Infinity waxay soo bandhigtay oo caan ka dhigtay kor u qaadida gobolada CPU iyo NVMe cabir ahaan.
Waa maxay qiimaha waxqabadka ugu weyn ee u dejinta CPU ama NVMe?
Ka guurista gobolada ka baxsan GPU waxay la macno tahay ku wareejinta xogta PCIe ama NVMe, taas oo ka gaabinaysa xusuusta GPU-da, markaa wax soo saarku wuu yaraadaa ilaa lagu daro xisaabinta.
Sidee bay nidaamyada rarista u qariyaan inta badan daahitaanka wareejinta?
Jadwaleeyayaashu waxay ka sii hormariyaan qaybaha loo baahan yahay PCIe halka GPUgu wali xisaabinayo, isgaadhsiinta isdhaafsan ee xisaabinta si loo daboolo daahitaanka.
Sii wad waxbarashada
Tilmaamaha la xidhiidha
Tilmaamayaal badan ayaa loo doortay mawduucan