Overview
Dhata parallelism inodzidzisa modhi imwe nekukurumidza nekuidzokorora kune akawanda maGPU, neGPU yega yega inogadzira chikamu chakasiyana che data batch. Ndiyo nzira yekushanda inoita kuti zvikwata zvikwire kusvika kune gumi nemaviri kana zviuru zveanomhanyisa.
Data Parallelism idhizaini yekuvaka inobata mhando yemhando, mutengo wezvivakwa, latency, uye kuvimbika pachiyero.
Deep Dive
Mune data parallelism, GPU yega yega inobata kopi yakafanana yehuremu hwemuenzaniso asi inogadzirisa yakasarudzika mini-batch yemienzaniso yekudzidziswa. Chishandiso chega chega chinoverengera chinopfuura chekumberi nekumashure chakazvimirira, chichigadzira chayo seti yemagradients. Uremu husati hwavandudzwa, ma gradients anoverengerwa pamaGPU ese achishandisa zvese-kuderedza kutaurirana mashandiro, saka replica yega yega inogara mukuwirirana uye inoita sekunge yakadzidziswa pane imwe hombe yakasanganiswa batch. Izvi zvinonyatso wedzera kubuda: 8 GPUs inogona kutsenga kuburikidza ne8x iyo data padanho. Iyo inobata ndeyekuti yega yega GPU inofanirwa kukwana iyo yese modhi, ma gradients, uye optimizer mamiriro mundangariro, saka yakajeka data parallelism haibatsire kana modhi yakakurisa kune chimwe chinhu.
Technical Insight
Kushanda kwakakosha ndekwese-kudzikisa, iyo inoverengera gradients pamidziyo yese uye kugovera mhedzisiro. Ring-yese-inoderedza, inoshandiswa nemaraibhurari seNCCL neHorovod, inopfuudza gradient chunks kutenderedza mhete inonzwisisika saka kutaurirana kwakazara kwakazvimirira paGPU kuverenga. PyTorch's DistributedDataParallel inoputira iyi kutaurirana neiyo yekumashure pass, ichidzima gradient sync kune ekutanga maseru apo gare gare maseru ichiri komputa, ichivanza yakawanda yetiweki latency.
Mastering Data Parallelism
Dhata parallelism inodzidzisa modhi imwe nekukurumidza nekuidzokorora kune akawanda maGPU, neGPU yega yega inogadzira chikamu chakasiyana che data batch. Ndiyo nzira yekushanda inoita kuti zvikwata zvikwire kusvika kune gumi nemaviri kana zviuru zveanomhanyisa. Data Parallelism idhizaini yekuvaka inobata mhando yemhando, mutengo wezvivakwa, latency, uye kuvimbika pachiyero. Kuvaka kunzwisisa kwakadzama, bata Data Parallelism semuenzaniso wekushandisa, kwete chinhu chimwe chete: tsanangura zvinodikanwa, kujekesa fungidziro, uye patsanura izvo zvinogona kuitwa nehurongwa hwakavimbika kubva kune zvichiri kuda kutonga kwenyanzvi.
Mukuita, zvikwata zvakasimba zvinoshandisa Data Parallelism inogadzirisa zvivakwa, data, uye sarudzo dzezvivakwa zvinopesana nekuvimbika uye mutengo. Ivo vanonyora zvakajeka maitiro ebudiriro, bvunzo vachipokana ne data rechokwadi uye mafambiro ebasa, uye iterate zvichibva pane zvakacherechedzwa maitiro ekutadza kwete kuhwina-nguva imwe chete yebhenji. Apa ndipo apo kunzwisisa kwe theoretical kunoshanduka kuve kugona kwakasimba pane chigadzirwa, mutemo, uye mashandiro.
Zvisarudzo zvezvivakwa zvinotyaira kuita uye mutengo wekushandisa kwemakore. Panguva imwecheteyo, Kukwirisa imwe bhenji kunogona kuvanza yakafara system kushaya simba. Nzira yakatsiga ndeyekubatanidza kukurumidza kuyedza nekutonga: mhanyisa vatyairi vendege, tora humbowo, buritsa matanda esarudzo, uye urambe uchivandudza chengetedzo semaitiro emuenzaniso, zvinotarisirwa nemushandisi, uye zvinodikanwa zvekutonga.
Strategic Impact
Zvisarudzo zvezvivakwa zvinotyaira kuita uye mutengo wekushandisa kwemakore.
Zvisarudzo zvezvivakwa zvinotyaira kuita uye mutengo wekushandisa kwemakore. Mukutumirwa kwemhando yepamusoro, izvi zvinoshandurirwa kuita mitemo inoyerwa yekushanda, miganhu yevaridzi, uye tsika dzekudzokorora dzinodzokororwa kuitira kuti zvikwata zvikwire kuvimba pane kukwidza kusajeka.
Dzidzo yehunyanzvi inobatsira zvikwata kusarudza murwi wakakodzera, kwete iwo mutsva chete.
Dzidzo yehunyanzvi inobatsira zvikwata kusarudza murwi wakakodzera, kwete iwo mutsva chete. Mukutumirwa kwemhando yepamusoro, izvi zvinoshandurirwa kuita mitemo inoyerwa yekushanda, miganhu yevaridzi, uye tsika dzekudzokorora dzinodzokororwa kuitira kuti zvikwata zvikwire kuvimba pane kukwidza kusajeka.
Sarudzo dzeinjiniya dziri nani dzinoderedza zviitiko zvekuvimbika mukugadzira.
Sarudzo dzeinjiniya dziri nani dzinoderedza zviitiko zvekuvimbika mukugadzira. Mukutumirwa kwemhando yepamusoro, izvi zvinoshandurirwa kuita mitemo inoyerwa yekushanda, miganhu yevaridzi, uye tsika dzekudzokorora dzinodzokororwa kuitira kuti zvikwata zvikwire kuvimba pane kukwidza kusajeka.
Real-World Implementation
Kudzidzisa ResNet mufananidzo classifier mhiri 8 GPUs mune imwe sevha uchishandisa PyTorch DistributedDataParallel, yega yega GPU inobata makumi matatu nemaviri e256-mufananidzo batch.
Kuyera BERT pretraining mhiri kwemazana eGPUs neHorovod, uchishandisa mhete zvese-kuderedza kuwiriranisa gradients nhanho imwe neimwe.
Kunyatsogadzirisa modhi yekurudziro pane akawanda-node cluster apo imwe neimwe node inogadzira akasiyana-siyana-yekudyidzana shards.
Kushandisa TensorFlow's MirroredStrategy kuparadzira kudzidziswa kwemuenzaniso wechiratidzo pane akawanda maGPU pane imwe nzvimbo yekushandira ine mashoma kodhi shanduko.
Maitiro Ekuita
Data Parallelism mukuita
Kudzidzisa ResNet mufananidzo classifier mhiri 8 GPUs mune imwe sevha uchishandisa PyTorch DistributedDataParallel, yega yega GPU inobata makumi matatu nemaviri e256-mufananidzo batch.
Kudzidzira ResNet mufananidzo classifier mhiri 8 GPUs mune imwe sevha uchishandisa PyTorch DistributedDataParallel, yega yega GPU inobata makumi matatu nembiri ye256-image batch Matimu anowanzo kuwana mhedzisiro iri nani kana achinge atsanangura emhando yepamusoro kumberi, chengetedza nzira yekukwira kwevanhu yemakesi emupendero, uye kuteedzera zvese zvakawanikwa zvechigadzirwa uye mutengo wekukanganisa nekufamba kwenguva.
Data Parallelism mukuita
Kuyera BERT pretraining mhiri kwemazana eGPUs neHorovod, uchishandisa mhete zvese-kuderedza kuwiriranisa gradients nhanho imwe neimwe.
Kuyera BERT pretraining mhiri kwemazana eGPUs neHorovod, uchishandisa mhete-yese-kudzikisa kuwiriranisa gradients danho rega rega Matimu anowanzo kuwana mibairo iri nani kana achinge atsanangura emhando yepamusoro kumberi, chengetedza nzira yekukwira kwevanhu yemakesi emupendero, uye kuteedzera zvese zvakawanikwa zvekubereka uye kukanganisa mutengo nekufamba kwenguva.
Data Parallelism mukuita
Kunyatsogadzirisa modhi yekurudziro pane akawanda-node cluster apo imwe neimwe node inogadzira akasiyana-siyana-yekudyidzana shards.
Kunyatsogadzirisa modhi yekurudziro pane akawanda-node cluster apo imwe neimwe node inogadzira akasiyana mushandisi-yekudyidzana shards Matimu anowanzo kuwana mhedzisiro iri nani kana achinge atsanangura emhando yepamusoro kumberi, chengetedza nzira yekukwira kwevanhu yemakesi emupendero, uye kuteedzera zvese zvakawanikwa zvechigadzirwa uye mutengo wekukanganisa nekufamba kwenguva.
Data Parallelism mukuita
Kushandisa TensorFlow's MirroredStrategy kuparadzira kudzidziswa kwemuenzaniso wechiratidzo pane akawanda maGPU pane imwe nzvimbo yekushandira ine mashoma kodhi shanduko.
Kushandisa TensorFlow's MirroredStrategy kuparadzira kudzidziswa kwemuenzaniso wechiratidzo pane akawanda maGPU pane imwe nzvimbo yekushandira ine kadiki shanduko yekodhi Matimu anowanzo kuwana mhedzisiro iri nani kana achinge atsanangura emhando yepamusoro kumberi, chengetedza nzira yekukwira kwevanhu yemakesi emupendero, uye kuteedzera zvese zvakawanikwa zvechigadzirwa uye mutengo wekukanganisa nekufamba kwenguva.
Njodzi & Guardrails
Kugadzirisa imwe bhenji kunogona kuvanza yakafara system kushaya simba.
Infrastructure uye mari yekugadzirisa inowanzotarisirwa pasi.
Chengetedzo uye kucherechedzwa mapundu anogona kukura sezvo masisitimu anowedzera kuoma.
Implementation Roadmap
Tsanangura latency, mhando, uye mutengo zvinangwa usati waitwa.
Tsanangura latency, mhando, uye mutengo zvinangwa usati waitwa. Bata nhanho yega yega segedhi rehumbowo: kana maitiro asina kusangana, imbomira kuburitsa, vhara gaka, uye wobva wawedzera kushandiswa.
Benchmark pasi pechokwadi mutoro uye data mamiriro.
Benchmark pasi pechokwadi mutoro uye data mamiriro. Bata nhanho yega yega segedhi rehumbowo: kana maitiro asina kusangana, imbomira kuburitsa, vhara gaka, uye wobva wawedzera kushandiswa.
Chishandiso chekutarisa zvikanganiso, kudonha, uye mushandisi maitiro.
Chishandiso chekutarisa zvikanganiso, kudonha, uye mushandisi maitiro. Bata nhanho yega yega segedhi rehumbowo: kana maitiro asina kusangana, imbomira kuburitsa, vhara gaka, uye wobva wawedzera kushandiswa.
Gadzirira nzira dzekudzosera kumashure uye dzezviitiko usati wawedzera.
Gadzirira nzira dzekudzosera kumashure uye dzezviitiko usati wawedzera. Bata nhanho yega yega segedhi rehumbowo: kana maitiro asina kusangana, imbomira kuburitsa, vhara gaka, uye wobva wawedzera kushandiswa.