Gëstu Nous
Nous Research labo IA la bu askan wi di dawal, ñu xamee ko ci defar ay model yu ubbeeku yu siiw ñu nekk ay assistant yu am kàttan, yu amul ay tënk, ak di puus tàggat yaram bu decentralisée.
Résumé
It shows how a small team plus an open-source community can compete on model quality without owning massive infrastructure.
Plongeur bu xóot
Gëstu Nous dafa jël xeetu base yu ubbeeku, rawatina famiy Llama ak Mistral, ba noppi defar leen ñu nekk Hermes ak Capybara yuñ gëna jëfandikoo. Seen xeetu OpenHermes ak Nous Hermes nekk nañu yenn ci xeetu music yiñ gëna telesarse ci Hugging Face, ñu sargal ko ndax topp tegtal yu dëgër ak fësal steerability moo gën baña nangu lu diis. Ginaaw fine-tuning, Nous xeex na jafe-jafe bu metti: tàggat yaram buñ séddale. Seen gëstu DisTrO ak DeMo optimizer seen mébet mooy wàññi bandwidth jokkoo bi war ci diggante GPUs, ak reso Psyche di jàngat tàggat model yu mag ci gox yu tasaaroo, hardware yu lëkkaloo ak internet. They have also experimented with tool-using and reasoning-focused models, positioning themselves at the frontier of open, decentralized AI.
Gis-gis xarala
Most of Nous's models are not trained from scratch; deñuy jëfandikoo fine-tuning ak optimisation preference (lu melni DPO) ci kaw poid yu ubbeeku yi di jëfandikoo ay done yuñ defar bu baax ak ay done yu nit. Seenug liggéeyu tàggat yaram dafay song bottleneck bandwidth bi: GPU yi dañu wara weccoo yeesali gradient yu mag ci jéego bu nekk. DisTrO/DeMo dafay kompresse ak dindi yeesali yii suko defee node yi mëna tàggat ñoom ñaar ci lëkkalekaay internet yu bari yi, du soxla lëkkaloo datacenter bu dëgër.
njeextalu pexe
Pexem jaaykat
Kartu yoonu jaaykat yi ñooy wane man-man yi sa ekip mëna tabax ci kanam.
Njëgg ak budget
Anamu jënd ak jaay ak tànneefi dugal dañu am njeexital ci njëg ak risk ci diir bu xawa yàgg.
Risk ak kaaraange
Li liggéeyukaay bi di ñaax mooy tëral ni produit bi di doxee, kaaraange gi ak ubbeeku gi.
Ëlëgu Gëstu Nous
Nous mingi waaja wax ni ëlëgu IA bu ubbeeku bi defa descentralisée, fu ordinatër bi di dajaloo ci ay nit ñu bari ñu moom seen bopp, duñu dajaloo ci ay cluster hyperscale yu néew. If their low-bandwidth training methods scale, communities could collectively train frontier-class models. Xaarandi génne gu wéy ci assistant yu ubbeeku yi, dugal xaalis bu gëna xóot ci reso buñ séddale Psyche, ak xeetu xalaat yu jublu ci xalaat. Seen liggéey mën na wàññi bu baax jafe-jafe yi ci tàggat model yu mag yi ci biti xarala yu mag yi.
Doxal ci àdduna dëgg
Developpeur yi dañuy doxal xeetu Nous Hermes ak OpenHermes ci gox bi ngir assistant chat yuñ mëna doxal te duñu fay API.
Gëstukat yi dañuy wax ci pexe DisTrO ak DeMo yu Nous suñuy jàngat modelu tàggat yaram bu am yaatuwaayu bandwidth.
Hobbyist yi ak entreprise yu ndaw yi defar nañu ay done yu Nous génne ngir tabax ay assistant yuñ jagleel domen bi.
Reseau Psyche dañu koy jëfandikoo ngir jàngat xeeti tàggat ci GPU volontaire yuñ séddale ci gox yi.
Risk yi ak balustrade yi
Koom-koomu ubbite mën na raw stabilite ci def liggéeyu defar dëgg.
Njëg yi ci API wala coppite ci sàrt yi mën nañu dindi xalaat yi ci guddi gi.
Dependence ci benn jaaykat dafay yokk njëgu tëjug ak migraasioŋ.
Roadmap ngir samp gi
Saytu sa fournisseur yi nga jëfandikoo sa liggéey ak say done.
Xoolaat mbir yu nëbbu, kaaraange ak sàrti yoon balaa ngay boole.
Fexe am palaŋu fallback ci model yi wala jaaykat yi.
Xool notu génne yi suko defee coppite yi ci kàrtu yoon du jaaxal ekip yi.
Weyal di banneexu
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the Nous Research quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Gis bi ci topp
Tongyi Lab ak Gëstu Qwen
Laaj yi ñuy faral di laaj
What is Nous Research?
Nous Research labo IA la bu askan wi di dawal, ñu xamee ko ci defar ay model yu ubbeeku yu siiw ñu nekk ay assistant yu am kàttan, yu amul ay tënk, ak di puus tàggat yaram bu decentralisée. Dafay wane ni ekip bu ndaw ak askan wu ubbeeku mënee joŋante ci kalite model te duñu am infrastructure yu mag.
Lan moo gëna xamee Nous Research ci teela jog ci liggéeyam?
Nous dafa am gëdd ndax dafa defar ay model yu ubbeeku yu melni Llama ak Mistral ci famiy Hermes bu siiw bi.
Ban modelu base la Nous gëna bari ay fine-tune?
Hermes ak Capybara yu Nous dañu tabax ci kaw xeetu base yu ubbeeku, rawatina Llama ak Mistral.
Ban jafe-jafe la projet DisTrO ak DeMo yu Nous di saafara?
DisTrO ak DeMo optimizer seen mébet mooy dagg bandwidth inter-GPU bi ñuy laaj, suko defee ñu mëna tàggat ci lëkkalekaay internet yu bari.
Luy reso Psyche?
Psyche mooy reso tàggat yaram bu Nous di jàngat ni ñuy boolee GPU yu tasaaroo ngir tàggat ay model yu mag.
Ban pexe la Nous di faral di jëfandikoo ngir méngale ay modelam yuñ defar bu baax?
Nous dafay jëfandikoo fine-tuning buñ saytu ak tànneef ci kaw poid yu ubbeeku yi di jëfandikoo ay done yuñ tànn.