A classifier trained only on transformer fine-tuning data detects an invariant memorization signature that transfers to Mamba, RWKV-4, and RecurrentGemma with AUCs of 0.963, 0.972, and 0.936.
super hub Mixed citations
Direct Preference Optimization: Your Language Model is Secretly a Reward Model
Mixed citation behavior. Most common role is background (69%).
117
Pith papers citing it
Background
69% of classified citations
abstract
While large-scale unsupervised language models (LMs) learn broad world knowledge and some reasoning skills, achieving precise control of their behavior is difficult due to the completely unsupervised nature of their training. Existing methods for gaining such steerability collect human labels of the relative quality of model generations and fine-tune the unsupervised LM to align with these preferences, often with reinforcement learning from human feedback (RLHF). However, RLHF is a complex and often unstable procedure, first fitting a reward model that reflects the human preferences, and then fine-tuning the large unsupervised LM using reinforcement learning to maximize this estimated reward without drifting too far from the original model. In this paper we introduce a new parameterization of the reward model in RLHF that enables extraction of the corresponding optimal policy in closed form, allowing us to solve the standard RLHF problem with only a simple classification loss. The resulting algorithm, which we call Direct Preference Optimization (DPO), is stable, performant, and computationally lightweight, eliminating the need for sampling from the LM during fine-tuning or performing significant hyperparameter tuning. Our experiments show that DPO can fine-tune LMs to align with human preferences as well as or better than existing methods. Notably, fine-tuning with DPO exceeds PPO-based RLHF in ability to control sentiment of generations, and matches or improves response quality in summarization and single-turn dialogue while being substantially simpler to implement and train.
hub tools
citation-role summary
background 25
method 8
baseline 1
dataset 1
citation-polarity summary
claims ledger
- abstract While large-scale unsupervised language models (LMs) learn broad world knowledge and some reasoning skills, achieving precise control of their behavior is difficult due to the completely unsupervised nature of their training. Existing methods for gaining such steerability collect human labels of the relative quality of model generations and fine-tune the unsupervised LM to align with these preferences, often with reinforcement learning from human feedback (RLHF). However, RLHF is a complex and often unstable procedure, first fitting a reward model that reflects the human preferences, and then
authors
co-cited works
representative citing papers
ORPO performs preference alignment during supervised fine-tuning via a monolithic odds ratio penalty, allowing 7B models to outperform larger state-of-the-art models on alignment benchmarks.
Formalizes Reasoning Portability (RP) and proposes RDB-CL to modulate per-sample KL regularization in RLVR for MLLM continual learning, achieving +12.0% Last accuracy over vanilla RLVR baseline by preserving reusable reasoning on high-RP samples.
Fast-Slow Training uses context optimization as fast weights alongside parameter updates as slow weights to achieve up to 3x better sample efficiency, higher performance, and less catastrophic forgetting than standard RL in continual LLM learning.
Introduces Block-R1 benchmark, Block-R1-41K dataset, and a conflict score to handle domain-specific optimal block sizes in RL post-training of diffusion LLMs.
Optimistic bilevel optimization with manifold lower-level minimizers is differentiable if the optimistic selection is unique, yielding a pseudoinverse hyper-gradient and a convergent HG-MS algorithm whose rate depends on intrinsic manifold dimension.
POISE trains a lightweight probe on the actor's internal states to predict expected rewards for RLVR, matching DAPO performance on math benchmarks with lower compute by avoiding extra rollouts or critic models.
Instruction tuning makes late-layer computation depend more on the model's own post-trained upstream state than on base-model upstream state, producing a consistent +1.68 logit interaction effect across five model families.
Topology-enhanced alignment via persistent homology on trajectories outperforms standard SFT and DPO baselines on preference metrics for LLMs.
Full development of 7B and 32B Olmo 3 models used 12.3 GWh datacenter energy and emitted 4,251 tCO2eq, with development overheads accounting for 82% of compute and reasoning models costing 17x more to post-train than instruction-tuned ones.
PEO optimizes original prompt embeddings continuously over adaptive rounds to jailbreak aligned LLMs, preserving the exact visible prompt text and outperforming discrete suffix, appended embedding, and search-based white-box attacks on harmful-behavior benchmarks.
Small VLMs show higher sycophancy (22.3% for 450M model) than larger ones (6.0% for 7B) when scoring image-text alignment on 173k fantasy portraits, quantified via a new Bluffing Coefficient metric.
A new 7x4 taxonomy organizes agentic AI security threats by architectural layer and persistence timescale, revealing under-explored upper layers and missing defenses after surveying 116 papers.
S-GRPO unifies SFT and RL for LVLMs via conditional ground-truth injection that supplies a maximal-reward anchor when group exploration fails completely.
Audio-Contrastive Preference Optimization (ACPO) mitigates audio hallucination in AVLMs via output-contrastive and input-contrastive objectives that enforce faithful audio grounding.
SCPT creates similarity-constrained preference triplets from scaffolds to train LLMs as conditional molecular editors that improve properties while keeping scaffolds intact.
SalesLLM provides an automatic evaluation framework for LLM sales dialogues that correlates 0.98 with human experts and shows top models approaching human performance while weaker ones lag.
VeriGUI adds a Thinking-Verification-Action-Expectation loop and two-stage training on synthetic failures to reduce undetected action errors and improve recovery in GUI automation.
CapTrack shows post-training causes drift beyond facts, with instruction fine-tuning producing stronger behavioral changes than preference optimization across model families.
PIAST iteratively optimizes few-shot examples in prompts via Monte Carlo Shapley value estimation, outperforming prior automatic prompting methods and setting new SOTA on classification, simplification, and GSM8K with modest compute.
EyeMulator augments CodeLLM fine-tuning loss with token weights derived from human eye-tracking scan paths, producing large gains on code translation and summarization across StarCoder, Llama-3.2 and DeepSeek-Coder.
Controlled study shows mixed training curricula improve aggregate F1 on memory QA benchmarks while out-of-domain data transfers targeted skills like temporal reasoning, with per-question-type effects exceeding aggregate differences.
DITTO uses RL with verbal feedback to train LLMs for human behavior simulation, reporting 36% average gains over base models and outperforming GPT-5.4 on 6 of 10 SOUL benchmark tasks.
AutoVecCoder combines VecPrompt for automated intrinsic knowledge synthesis and VecRL for efficiency-aligned RL to train an 8B LLM that achieves SOTA on SimdBench SSE/AVX subsets and sometimes exceeds -O3 compiler results.
citing papers explorer
-
Adaptive Prompt Embedding Optimization for LLM Jailbreaking
PEO optimizes original prompt embeddings continuously over adaptive rounds to jailbreak aligned LLMs, preserving the exact visible prompt text and outperforming discrete suffix, appended embedding, and search-based white-box attacks on harmful-behavior benchmarks.
-
SkillEvolver: Skill Learning as a Meta-Skill
A meta-skill authors and refines prose-and-code skills for agents by learning from post-deployment failures with an overfit audit, achieving 56.8% accuracy on SkillsBench tasks versus 43.6% for human-curated skills.
-
Separable Expert Architecture: Toward Privacy-Preserving LLM Personalization via Composable Adapters and Deletable User Proxies
A separable expert architecture uses base models, LoRA adapters, and deletable per-user proxies to enable privacy-preserving personalization and deterministic unlearning in LLMs.
-
Dual-Axis Generative Reward Model Toward Semantic and Turn-taking Robustness in Interactive Spoken Dialogue Models
A generative reward model supplies separate semantic and turn-taking scores for spoken dialogues to enable more reliable reinforcement learning.
-
Mitigating LLM biases toward spurious social contexts using direct preference optimization
Debiasing-DPO reduces bias to spurious social contexts by 84% and improves predictive accuracy by 52% on average for LLMs evaluating U.S. classroom transcripts.
-
Preference Goal Tuning: Post-Training as Latent Control for Frozen Policies
PGT optimizes latent goal embeddings for frozen policies via trajectory-level preference objectives, reporting 72-81.6% relative gains on 17 Minecraft tasks and 13.4% better OOD performance than fine-tuning.
-
Agent Q: Advanced Reasoning and Learning for Autonomous AI Agents
Agent Q integrates MCTS-guided search, self-critique, and off-policy DPO to train LLM agents that outperform behavior cloning and reinforced fine-tuning baselines in WebShop and achieve up to 95.4% success in real-world booking scenarios.
-
Inference Scaling Laws: An Empirical Analysis of Compute-Optimal Inference for Problem-Solving with Language Models
Empirical analysis shows scaling inference compute via strategies like tree search can be more efficient than scaling model parameters, with 7B models plus novel search outperforming 34B models.
-
ECPO: Evidence-Coupled Policy Optimization for Evidence-Certified Candidate Ranking
ECPO is a listwise policy optimization method that couples ranking utility with span-level evidence certificate validity and a deterministic verifier reward on MAVEN-ERE and RAMS datasets.
-
Signal Reshaping for GRPO in Weak-Feedback Agentic Code Repair
Reshaping outcome rewards, process signals, and rollout comparability in GRPO raises strict compile-and-semantic accuracy in agentic code repair from 0.385 to 0.535 under weak feedback.
-
Explanation Quality Assessment as Ranking with Listwise Rewards
Explanation quality assessment is recast as ranking with listwise and pairwise losses that outperform regression, allow small models to match large ones on curated data, and enable stable convergence in reinforcement learning.
-
Mind DeepResearch Technical Report
MindDR combines a Planning Agent, DeepSearch Agent, and Report Agent with SFT cold-start, Search-RL, Report-RL, and preference alignment to reach competitive scores on research benchmarks using 30B-scale models.
-
Voxtral TTS
Voxtral TTS produces expressive multilingual speech from 3-second reference audio with a hybrid autoregressive-plus-flow-matching architecture and a new VQ-FSQ tokenizer, achieving 68.4% win rate over ElevenLabs in human evaluations.
-
Trustworthy LLMs: a Survey and Guideline for Evaluating Large Language Models' Alignment
Survey organizes LLM trustworthiness into seven categories and 29 sub-categories, measures eight sub-categories on popular models, and finds that more aligned models generally score higher but with varying effectiveness.
-
DuIVRS-2: An LLM-based Interactive Voice Response System for Large-scale POI Attribute Acquisition
DuIVRS-2 deploys an LLM-driven IVR pipeline that processes 0.4 million calls per day at 83.9 percent task success rate using FSM-guided augmentation, selective CoT generation, and cooperative policy iteration.
-
Agentic AI Security: Threats, Defenses, Evaluation, and Open Challenges
A survey that taxonomizes threats to agentic AI, reviews benchmarks and evaluation methods, discusses technical and governance defenses, and identifies open challenges.
- Adaptive Teacher Exposure for Self-Distillation in LLM Reasoning