Subira ku makuru
Guhanga udushyaAI Understanding ibisobanuro

Imiterere-Yatesheje agaciro DPO: Kuvugurura ubumenyi bwa LLM hamwe nukuri-Kumenya Ibyatoranijwe Byatoranijwe

Abashakashatsi batanga ibitekerezo byuburyo butemewe (SD-DPO) kugirango barusheho kunoza imiterere yindimi nini (LLMs) mugukuramo ubumenyi bwabitswe.

4 min readRead the primary source
Source-provided image accompanying Style-Debiased DPO: Updating LLM Knowledge with Factuality-Aware Synthetic Preference Data
Inyandiko y'ibanzeInkomoko yanditse
Umwanditsi
arxiv.org
Ihuza ry'inkomoko
arxiv.orghttps://arxiv.org/abs/2609.16532
Ubwoko bw'inkomoko
Inyandiko y'ibanze - itangazo ryemewe, impapuro, dosiye, cyangwa urupapuro rwambere-dusoma mu buryo butaziguye.
ImirongoSobanukirwa ibi mumasegonda 60

Tangira hano

Amagambo y'ingenzi

DPO (Optimisation Yibanze)
Uburyo bwo guhugura butunganya neza moderi kuri babiri batabanje gukenera icyitegererezo cyihariye.
Ururimi runini (LLM)
Ururimi rwicyitegererezo rwahuguwe kumyandiko minini corpora kubyara no gusesengura inyandiko.
Ukuri
Nigute mubyukuri ibyifuzo byicyitegererezo bihuye nukuri kwukuri kwamakuru.
IsuzumeAI ni iki? Ikibazo

Byagenze bite

Abashakashatsi basabye uburyo bwo guhitamo uburyo bwiza bwo guhitamo (SD-DPO) kugira ngo barusheho kumenya imiterere y'ururimi runini (LLMs) mu kugarura ubumenyi bwabitswe. Bagerageje SD-DPO hejuru ya EntiGraph, uhagarariye uburyo bwo kubika impande zombi zikomeza kwitegura (CPT) kumyandiko ikomatanyirijwe muri corpus. Ibisubizo byerekanaga ko SD-DPO irenze CPT y'ibanze kuri synthique ya EntiGraph kuva muburyo bumwe kandi igasuzuma hamwe nuburyo bumwe.

Abashakashatsi basabye uburyo bwo guhitamo uburyo bwiza bwo guhitamo (SD-DPO) kugira ngo barusheho kumenya imiterere y'ururimi runini (LLMs) mu kugarura ubumenyi bwabitswe.

Bagerageje SD-DPO hejuru ya EntiGraph, uhagarariye uburyo bwo kubika impande zombi zikomeza kwitegura (CPT) kumyandiko ikomatanyirijwe muri corpus.

Ibisubizo byerekanaga ko SD-DPO irenze CPT y'ibanze kuri synthique ya EntiGraph kuva muburyo bumwe kandi igasuzuma hamwe nuburyo bumwe.

Ibisobanuro birambuye: arxiv.org β†—

Impamvu ari ngombwa

Uburyo buteganijwe, SD-DPO, burashobora kunoza ukuri kwa LLM mugukuramo ubumenyi bwabitswe. Ibi nibyingenzi byingenzi mubisabwa bisaba ubumenyi nyabwo kandi bugezweho, nkubumenyi bugezweho no guhindura. SD-DPO ifite ubushobozi bwo kunoza imikorere ya LLMs muriyi porogaramu.

Uburyo buteganijwe, SD-DPO, burashobora kunoza ukuri kwa LLM mugukuramo ubumenyi bwabitswe.

Ibi nibyingenzi byingenzi mubisabwa bisaba ubumenyi nyabwo kandi bugezweho, nkubumenyi bugezweho no guhindura.

SD-DPO ifite ubushobozi bwo kunoza imikorere ya LLMs muriyi porogaramu.

Interactive Mechanism

Uburyo bukoreshwa: Uburyo bukora

Shakisha ikoranabuhanga ryihishe inyuma yiri terambere.

Thinking Budget (Test-Time Tokens):1,024 tokens
Complex Accuracy79%Math & Code Logic
Latency3.2sTime to first full output
Inference Cost$0.0092Per query estimated
Reasoning StyleStep VerificationInternal chain depth
Active Thinking Trace:
1Deconstruct user problem into formal constraints
2Propose candidate hypotheses & step-by-step calculation
3Self-correction: Backtrack and refute subtle edge cases
4Exhaustive consistency check & final output synthesis
Core takeaway: Test-time compute fundamentally changes AI economics. Instead of only scaling during pre-training, giving reasoning models more tokens at inference time allows them to systematically solve PhD-level STEM problems.
Kugenzura Ibitekerezo Byagenzuwe+10 Points
What is AI? Quiz

A route planner searches possible journeys using explicit rules. What does this illustrate about AI?

Ibyo kureba

Uburyo buteganijwe, SD-DPO, bufite ubushobozi bwo kunoza imikorere ya LLMs muburyo bwo kuvugurura no guhindura porogaramu. Ubundi bushakashatsi burakenewe kugirango dusuzume neza imikorere ya SD-DPO no gushakisha uburyo bushobora gukoreshwa.

Uburyo buteganijwe, SD-DPO, bufite ubushobozi bwo kunoza imikorere ya LLMs muburyo bwo kuvugurura no guhindura porogaramu.

Ubundi bushakashatsi burakenewe kugirango dusuzume neza imikorere ya SD-DPO no gushakisha uburyo bushobora gukoreshwa.

Ibyavuye mu bushakashatsi byerekana ko SD-DPO ishobora kunoza ukuri kwa LLM mu kugarura ubumenyi bwabitswe.

Ibijyanye nuyobora & ibibazo

AI ni iki?Imyitwarire ya AIAbakozi ba AIModeri ya AI YasobanuweAbahinduraGerageza ibyo uzi - gerageza ikibazo cya AI kubuntuReba ijambo AI mumagambo yacuKurikiza icyerekezo cya AI cyo kurekura
Basanze ari ingirakamaro?