Ubuyobozi bwa tekiniki

Gukwirakwiza Leta Kurekura kuri CPU na NVMe

Amayeri yo kuzigama yibitse ahagarika kubika ibitabo biremereye byamahugurwa (optimizer leta, gradients, rimwe na rimwe uburemere) muri CPU RAM cyangwa kuri NVMe SSDs aho kuba GPU yibuke.

Umva iki gitaboAI-byavuzwe · ~2 min

Kuriyi page2 min soma
  1. Incamake
  2. Kwibira cyane
  3. Gushyira mu bikorwa Isi
  4. Kazoza ka Optimizer Leta Yipakurura kuri CPU na NVMe
  5. Kubishyira mubikorwa
  6. Komeza Ubushakashatsi
  7. Ibibazo bikunze kubazwa

Incamake

It lets people train far larger models than their GPU's memory would otherwise allow.

Kwibira cyane

Iyo uhuguye urusobe rw'imitsi hamwe na optimizer nka Adam, buri kintu cyose gitwara imizigo yinyongera: imibare ibiri ikora (umuvuduko na variance), wongeyeho kopi yuzuye yuburemere, wongeyeho na gradient yayo. Mu mahugurwa avanze-yuzuye neza ibi birashobora kuba hafi ya 16 bytes kuri parameter, bikagabanya 2 bytes kuburemere ubwabwo. Kurekura bimura iyo mitwaro kuri GPU. CPU yamanura imitekerereze ya optimizer ivuga muri sisitemu isanzwe RAM hejuru ya bisi ya PCIe, mugihe NVMe yamanuye ibasunika kugeza kumanuka yihuta ya disiki ikomeye. Yamamaye cyane muri ZeRO-Infinity na ZeRO-Offload ya DeepSpeed, tekinike igurisha umuvuduko muke kubushobozi, ikareka GPU imwe cyangwa cluster ntoya ihuza neza na miliyari zingana.

Gushyira mu bikorwa Isi

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.

Kazoza ka Optimizer Leta Yipakurura kuri CPU na NVMe

Nka moderi ikomeza kwiyongera kububiko bwa GPU, gupakurura ibyiciro bigenda biba ibisanzwe aho kuba exotic. Tegereza guhuza cyane hamwe n’imikoranire yihuse nka NVLink-C2C na CXL yibuka yibidendezi bihindura imipaka ya CPU-GPU, wongeyeho gahunda zubwenge ziteganya ibizaba mbere yo gutangira. Ububiko-bwibumbiye hamwe nka Grace Hopper bugabanya ibihano bya PCIe, kandi urwego rugenda rutera gukora ibintu byinshi byo gukuramo ibintu hafi ya mucyo kugirango abakunda ibintu bashobore guhuza neza moderi nini ku byuma byoroheje.

Kubishyira mubikorwa

Impamvu ari ngombwa

  • Igiciro na bije. Ibyemezo byubwubatsi bitwara imikorere nigiciro cyimikorere kumyaka.
  • Ibyemezo bisobanutse. Ubuhanga bwa tekinike bufasha amakipe guhitamo umurongo ukwiye, ntabwo ari shyashya gusa.
  • Kugenzura ubuziranenge. Guhitamo neza bya injeniyeri bigabanya ibintu byizewe mubikorwa.

Witondere

  • Gutezimbere igipimo kimwe gishobora guhisha intege nke za sisitemu.
  • Ibikorwa Remezo no kubungabunga akenshi usanga bidahabwa agaciro.
  • Icyuho cyumutekano no kwitegereza birashobora kwiyongera uko sisitemu igenda igorana.

Intambwe ikurikira

  1. Sobanura ubukererwe, ubuziranenge, nigiciro cyibiciro mbere yo kubishyira mubikorwa.
  2. Ibipimo byerekana umutwaro ufatika hamwe namakuru yimiterere.
  3. Gukurikirana ibikoresho kubikosa, drift, ningaruka zabakoresha.
  4. Tegura inzira yo gusubiza ibyabaye mbere yo gupima.

Komeza Ubushakashatsi

Ibaruwa y'amakuru y'ubuntu

Ibuka amakuru ya buri munsi ya AI

Inkuru eshatu zizewe za AI buri gitondo cy'iminsi y'akazi, zanditse mu rurimi rworoshye. Ubuntu iteka ryose, nta kwamamaza.

Imeri imwe buri munsi y'akazi. Kwiyimura mu buryo bumwe. Ntitwigera tugurisha cyangwa ngo dusangire aderesi yawe.

Isuzume

Take the Optimizer State Offloading to CPU and NVMe quiz

Ibisubizo ako kanya kuri buri kibazo, n'icyemezo gishobora gusangirwa gifite ID yagenzurwa umaze gutsinda isomo.

Tangira ikibazo

Fasha uburezi bwa AI bw'ubuntu. AI Understanding ni umuryango utegamiye kuri leta wa 501(c)(3) — nta kwamamaza, nta kwishyura, iteka ryose. Tanga impano

Ibibazo bikunze kubazwa

What is Optimizer State Offloading to CPU and NVMe?

Amayeri yo kuzigama yibitse ahagarika kubika ibitabo biremereye byamahugurwa (optimizer leta, gradients, rimwe na rimwe uburemere) muri CPU RAM cyangwa kuri NVMe SSDs aho kuba GPU yibuke. Bituma abantu bitoza moderi nini cyane kuruta ububiko bwa GPU bwakwemerera.

Niyihe ntego yibanze yo gupakurura optimizer leta kuri CPU cyangwa NVMe?

Kurekura byimura optimizer iremereye kuri GPU muri RAM RAM cyangwa NVMe, kubohora ububiko bwa GPU kuburyo moderi nini zishobora gutozwa ibyuma bimwe.

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.

Nibihe bikoresho biranga abantu benshi bakwirakwiza optimizer?

DeepSpeed's ZeRO-Offload na ZeRO-Infinity yatangije kandi ikwirakwiza ibimanuka byo gukuramo ibintu kuri CPU na NVMe kurwego.

Ni ikihe giciro nyamukuru cyibikorwa byo gupakurura kuri CPU cyangwa NVMe?

Kwimura leta hanze ya GPU bisobanura kohereza amakuru hejuru ya PCIe cyangwa kuri NVMe, itinda kurenza ububiko bwa GPU, bityo ibicuruzwa byinjira bikamanuka keretse byuzuye na compte.

Nigute sisitemu yo gupakurura ihisha byinshi mubukererwe?

Abategura bategura ibice bikenewe hejuru ya PCIe mugihe GPU ikiri kubara, guhuza itumanaho hamwe no kubara kubitinda bya mask.