REVIEW 8 cited by
Self-Consistency Preference Optimization
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Self-alignment, whereby models learn to improve themselves without human annotation, is a rapidly growing research area. However, existing techniques often fail to improve complex reasoning tasks due to the difficulty of assigning correct rewards. An orthogonal approach that is known to improve correctness is self-consistency, a method applied at inference time based on multiple sampling in order to find the most consistent answer. In this work, we extend the self-consistency concept to help train models. We thus introduce self-consistency preference optimization (ScPO), which iteratively trains consistent answers to be preferred over inconsistent ones on unsupervised new problems. We show ScPO leads to large improvements over conventional reward model training on reasoning tasks such as GSM8K and MATH, closing the gap with supervised training with gold answers or preferences, and that combining ScPO with standard supervised learning improves results even further. On ZebraLogic, ScPO finetunes Llama-3 8B to be superior to Llama-3 70B, Gemma-2 27B, and Claude-3 Haiku.
Forward citations
Cited by 8 Pith papers
-
TeamTR: Trust-Region Fine-Tuning for Multi-Agent LLM Coordination
TeamTR is a trust-region framework for multi-agent LLM fine-tuning that resamples trajectories after each update to convert quadratic compounding occupancy shift into linear scaling and yields per-update improvement l...
-
On-Policy Self-Distillation without Any Supervision
A language model improves its own math reasoning by distilling its majority-vote consensus into the prefixes of its own disagreeing answers, with no external labels.
-
Recursive Synthesis for Long-Horizon Terminal Tasks
A recursive verified synthesis pipeline generated 37,484 progressively harder terminal-agent tasks and used them to improve agent performance on three benchmarks.
-
Multi-Turn On-Policy Distillation with Prefix Replay
ReOPD offline-distills multi-turn agentic LLMs via teacher-prefix replay plus step-decay sampling, matching online OPD accuracy at ≥4× speed with zero tool calls.
-
PRInTS: Reward Modeling for Long-Horizon Information Seeking
A generative reward model that combines dense information-gain step scoring with recursive trajectory summarization improves best-of-n selection for long-horizon information-seeking agents.
-
Probability-Consistent Preference Optimization for Enhanced LLM Reasoning
PCPO selects preference pairs by combining correct-answer status with token-level probability consistency, then trains with a weighted DPO+NLL loss, yielding small and partly inconsistent gains over outcome-only metho...
-
Self-Training Large Language Models with Confident Reasoning
CORE-PO self-trains LLMs to prefer reasoning paths with high self-estimated confidence, improving answer and reasoning accuracy on several benchmarks.
-
Alignment and Safety in Large Language Models: Safety Mechanisms, Training Paradigms, and Emerging Challenges
A broad survey of LLM alignment that catalogs objectives, benchmarks, SFT/RLHF/DPO methods, and safety challenges, without contributing new experimental or theoretical results.
Discussion (0). Continue with ORCID to comment.