I-DPO Esuswe Isitayela: Ibuyekeza Ulwazi lwe-LLM Ngedatha Yezintandokazi Zokwenziwa Kweqiniso-Aware

Idokhumenti yomthombo oyinhlokoUmthombo urekhodiwe Sep 16, 2026 04:44 UTC

Umshicileli: arxiv.org
Isixhumanisi somthombo: arxiv.orghttps://arxiv.org/abs/2609.16532
Uhlobo lomthombo: Idokhumenti eyisisekelo — isimemezelo esisemthethweni, iphepha, ukugcwalisa, noma ikhasi lomuntu wokuqala esilifunda ngokuqondile.

UmongoQonda lokhu ngemizuzwana engama-60

Qala lapha

Imigomo ebalulekile

I-DPO (Direct Preference Optimization): Indlela yokuqeqesha eshuna kahle imodeli ngokuqondile kumapheya athandwayo ngaphandle kokudinga imodeli yomklomelo ehlukile.
Imodeli Yolimi Olukhulu (LLM): Imodeli yolimi eqeqeshwe ku-massive text corpora ukuze ikhiqize futhi ihlaziye umbhalo.
Iqiniso: Ukunemba okungakanani izimangalo zemodeli zifana nolwazi lomhlaba wangempela olungaqinisekiswa.

Kwenzekeni

Abacwaningi bahlongoze i-style-debiased direct preference optimization (SD-DPO) ukuze kuthuthukiswe ukunemba kwamamodeli wezilimi ezinkulu (LLMs) ekubuyiseni ulwazi olugciniwe. Bahlole i-SD-DPO phezu kwe-EntiGraph, indlela emele uhlangothi lokulondoloza esebenzisa ukuqeqeshwa kwangaphambi kwesikhathi (CPT) kumbhalo owenziwe kusuka ku-corpus. Imiphumela ibonise ukuthi i-SD-DPO idlula i-CPT yesisekelo kudatha yokwenziwa ye-EntiGraph kusukela kumodeli yesisekelo efanayo futhi ihlola ngenqubo efanayo.

Abacwaningi bahlongoze i-style-debiased direct preference optimization (SD-DPO) ukuze kuthuthukiswe ukunemba kwamamodeli wezilimi ezinkulu (LLMs) ekubuyiseni ulwazi olugciniwe.

Bahlole i-SD-DPO phezu kwe-EntiGraph, indlela emele uhlangothi lokulondoloza esebenzisa ukuqeqeshwa kwangaphambi kwesikhathi (CPT) kumbhalo owenziwe kusuka ku-corpus.

Imiphumela ibonise ukuthi i-SD-DPO idlula i-CPT yesisekelo kudatha yokwenziwa ye-EntiGraph kusukela kumodeli yesisekelo efanayo futhi ihlola ngenqubo efanayo.

Imininingwane yomthombo: arxiv.org ↗

Kungani kubalulekile

Indlela ehlongozwayo, i-SD-DPO, ingathuthukisa ukunemba kwama-LLM ekubuyiseni ulwazi olugciniwe. Lokhu kubaluleke kakhulu ezinhlelweni ezidinga ulwazi olunembile nolwakamuva, olufana nokubuyekeza ulwazi nokuhlela. I-SD-DPO inamandla okuthuthukisa ukusebenza kwama-LLM kulezi zinhlelo zokusebenza.

Indlela ehlongozwayo, i-SD-DPO, ingathuthukisa ukunemba kwama-LLM ekubuyiseni ulwazi olugciniwe.

Lokhu kubaluleke kakhulu ezinhlelweni ezidinga ulwazi olunembile nolwakamuva, olufana nokubuyekeza ulwazi nokuhlela.

I-SD-DPO inamandla okuthuthukisa ukusebenza kwama-LLM kulezi zinhlelo zokusebenza.

Interactive Mechanism

I-Interactive Mechanism: Indlela Esebenza Ngayo Ngempela

Hlola ubuchwepheshe obuyisisekelo ngemuva kwalokhu kuthuthukiswa ngokuhlanganyela.

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.

I-Interactive Concept Check+10 Points

What is AI? Quiz

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

Ongakubuka ngokulandelayo

Indlela ehlongozwayo, i-SD-DPO, inamandla okuthuthukisa ukusebenza kwama-LLM ekubuyekezeni ulwazi nasekuhleleni izinhlelo zokusebenza. Ucwaningo olwengeziwe luyadingeka ukuze kuhlolwe ngokugcwele ukusebenza kwe-SD-DPO kanye nokuhlola okungenzeka ukuthi isetshenziswa.

Indlela ehlongozwayo, i-SD-DPO, inamandla okuthuthukisa ukusebenza kwama-LLM ekubuyekezeni ulwazi nasekuhleleni izinhlelo zokusebenza.

Ucwaningo olwengeziwe luyadingeka ukuze kuhlolwe ngokugcwele ukusebenza kwe-SD-DPO kanye nokuhlola okungenzeka ukuthi isetshenziswa.

Imiphumela yocwaningo iphakamisa ukuthi i-SD-DPO ingathuthukisa ukunemba kwama-LLM ekubuyiseni ulwazi olugciniwe.

Imihlahlandlela ehlobene nemibuzo

Yini i-AI?Ukuziphatha kwe-AI Ama-AI Agents Amamodeli e-AI Achaziwe Ama-Transformers Hlola okwaziyo — zama imibuzo ye-AI yamahhala Bheka igama le-AI kuhlu lwethu lwamagama Landela i-tracker yokukhishwa kwemodeli ye-AI

Uthole lokhu kuwusizo?