Flow-GRPO is the first online RL method for flow matching models, raising GenEval accuracy from 63% to 95% and text-rendering accuracy from 59% to 92% with little reward hacking.
hub Mixed citations
RAFT: Reward rAnked FineTuning for Generative Foundation Model Alignment
Mixed citation behavior. Most common role is background (64%).
33
Pith papers citing it
31
external citations · Pith
Background
64% of classified citations
abstract
Generative foundation models are susceptible to implicit biases that can arise from extensive unsupervised training data. Such biases can produce suboptimal samples, skewed outcomes, and unfairness, with potentially serious consequences. Consequently, aligning these models with human ethics and preferences is an essential step toward ensuring their responsible and effective deployment in real-world applications. Prior research has primarily employed Reinforcement Learning from Human Feedback (RLHF) to address this problem, where generative models are fine-tuned with RL algorithms guided by a human-feedback-informed reward model. However, the inefficiencies and instabilities associated with RL algorithms frequently present substantial obstacles to the successful alignment, necessitating the development of a more robust and streamlined approach. To this end, we introduce a new framework, Reward rAnked FineTuning (RAFT), designed to align generative models effectively. Utilizing a reward model and a sufficient number of samples, our approach selects the high-quality samples, discarding those that exhibit undesired behavior, and subsequently enhancing the model by fine-tuning on these filtered samples. Our studies show that RAFT can effectively improve the model performance in both reward learning and other automated metrics in both large language models and diffusion models.
hub tools
citation-role summary
background 6
method 3
baseline 1
other 1
citation-polarity summary
representative citing papers
DISA decouples partition function estimation using offline importance sampling for distribution-matching LLM-RL, matching or exceeding online baselines like FlowRL on math and code benchmarks while retaining more strategy diversity.
Introduces TBPO, which derives a Bregman-divergence density-ratio matching objective for token-level preference optimization that generalizes DPO while preserving the induced optimal policy.
PNAPO augments preference data with prior noise pairs and uses straight-line interpolation to create a tighter surrogate objective for offline alignment of rectified flow models.
BLADE uses Bayesian list-wise alignment with dynamic estimation to create a self-evolving target that overcomes limitations of static references in LLM-based recommendation, yielding sustained gains in ranking and complex metrics.
SOLACE improves text-to-image generation by using intrinsic self-confidence rewards from noise reconstruction accuracy during reinforcement learning post-training without external supervision.
IMPFM is a multi-particle flow-map sampling method with sequential posterior sharing and interaction-aware correction that targets a KL-tilted distribution for global exploration in online feedback search.
Active-GRPO reaches 0.1773 average SRxSim on TOMG-Bench MOLOPT by adaptively switching between imitation and self-reinforcement while upgrading references, outperforming GRPO and RePO.
HARVE removes the component of the reward-head vector aligned with a multi-directional hacking subspace from residual streams using a small set of contrastive examples, improving robustness on RewardHackBench across eight models without fine-tuning while preserving general capability.
Traj-Evolve combines non-parametric experience retrieval and multi-agent RL with a leave-one-out unification strategy to outperform baselines on lung cancer prediction from up to five years of multimodal EHRs, including in never-smokers.
CauSim turns scarce causal reasoning labels into scalable supervised data by having LLMs incrementally construct complex executable structural causal models.
Response times modeled as drift-diffusion processes enable consistent estimation of population-average preferences from heterogeneous anonymous binary choices.
Align-Cultura introduces the CULTURAX dataset and shows that culturally fine-tuned LLMs improve joint HHH scores by 4-6%, cut cultural failures by 18%, and gain 10-12% efficiency with minimal leakage.
Reward models used as quality scorers in text-to-image generation encode demographic biases that cause reward-guided training to sexualize female subjects, reinforce stereotypes, and reduce diversity.
PROF curates RL training data via PRM-ORM consistency to improve both final-answer accuracy and intermediate reasoning quality while reducing reliance on strong process reward models.
A human preference dataset and VideoReward model enable Flow-DPO and Flow-NRG to produce smoother, better-aligned videos from text prompts in flow-based generators.
DRaFT fine-tunes diffusion models by differentiating through sampling to maximize rewards, outperforming RL baselines and improving aesthetics on Stable Diffusion 1.4.
ReST improves LLM translation quality on benchmarks via offline RL on self-generated data, achieving gains in a compute-efficient way compared to typical RLHF.
PAPA directly optimizes diffusion models via real-time user feedback for personalized preference alignment, drawing from variational inference, with an efficiency-enhanced variant EPAPA.
FlowAWR derives an advantage-weighted rectification for optimal velocity fields in flow models, claiming 2-5x faster convergence than DiffusionNFT on SD3.5-Medium.
SCPO is a steerable training method for reward models that improves minority cultural preference accuracy by up to 7 points and is up to 280% more data-efficient than standard finetuning on PRISM and GlobalOpinionQA datasets.
Lightweight autoregressive graph generation model uses structure-guided ordering for efficient serialization and two-phase training to boost novelty while maintaining validity.
GCSL reframes LLM fine-tuning as supervised pursuit of quality thresholds using natural-language goals, outperforming SFT and DPO on toxicity, code, and recommendation tasks.
ReMedi boosts LLM performance on EHR clinical predictions by up to 19.9% F1 through ground-truth-guided rationale regeneration and fine-tuning.
citing papers explorer
-
Flow-GRPO: Training Flow Matching Models via Online RL
Flow-GRPO is the first online RL method for flow matching models, raising GenEval accuracy from 63% to 95% and text-rendering accuracy from 59% to 92% with little reward hacking.
-
DISA: Offline Importance Sampling for Distribution-Matching LLM-RL
DISA decouples partition function estimation using offline importance sampling for distribution-matching LLM-RL, matching or exceeding online baselines like FlowRL on math and code benchmarks while retaining more strategy diversity.
-
TokenRatio: Principled Token-Level Preference Optimization via Ratio Matching
Introduces TBPO, which derives a Bregman-divergence density-ratio matching objective for token-level preference optimization that generalizes DPO while preserving the induced optimal policy.
-
Offline Preference Optimization for Rectified Flow with Noise-Tracked Pairs
PNAPO augments preference data with prior noise pairs and uses straight-line interpolation to create a tighter surrogate objective for offline alignment of rectified flow models.
-
Beyond Static Best-of-N: Bayesian List-wise Alignment for LLM-based Recommendation
BLADE uses Bayesian list-wise alignment with dynamic estimation to create a self-evolving target that overcomes limitations of static references in LLM-based recommendation, yielding sustained gains in ranking and complex metrics.
-
Improving Text-to-Image Generation with Intrinsic Self-Confidence Rewards
SOLACE improves text-to-image generation by using intrinsic self-confidence rewards from noise reconstruction accuracy during reinforcement learning post-training without external supervision.
-
Sequentially-Controlled Interactive Multi-Particle Flow-Maps for Online Feedback-Driven Search
IMPFM is a multi-particle flow-map sampling method with sequential posterior sharing and interaction-aware correction that targets a KL-tilted distribution for global exploration in online feedback search.
-
Active-GRPO: Adaptive Imitation and Self-Improving Reasoning for Molecular Optimization
Active-GRPO reaches 0.1773 average SRxSim on TOMG-Bench MOLOPT by adaptively switching between imitation and self-reinforcement while upgrading references, outperforming GRPO and RePO.
-
HARVE: Hacking-Aware Reward-Head Vector Editing for Robust Reward Models
HARVE removes the component of the reward-head vector aligned with a multi-directional hacking subspace from residual streams using a small set of contrastive examples, improving robustness on RewardHackBench across eight models without fine-tuning while preserving general capability.
-
Traj-Evolve: A Self-Evolving Multi-Agent System for Patient Trajectory Modeling in Lung Cancer Early Detection
Traj-Evolve combines non-parametric experience retrieval and multi-agent RL with a leave-one-out unification strategy to outperform baselines on lung cancer prediction from up to five years of multimodal EHRs, including in never-smokers.
-
CauSim: Scaling Causal Reasoning with Increasingly Complex Causal Simulators
CauSim turns scarce causal reasoning labels into scalable supervised data by having LLMs incrementally construct complex executable structural causal models.
-
Response Time Enhances Alignment with Heterogeneous Preferences
Response times modeled as drift-diffusion processes enable consistent estimation of population-average preferences from heterogeneous anonymous binary choices.
-
AlignCultura: Towards Culturally Aligned Large Language Models?
Align-Cultura introduces the CULTURAX dataset and shows that culturally fine-tuned LLMs improve joint HHH scores by 4-6%, cut cultural failures by 18%, and gain 10-12% efficiency with minimal leakage.
-
Bias at the End of the Score
Reward models used as quality scorers in text-to-image generation encode demographic biases that cause reward-guided training to sexualize female subjects, reinforce stereotypes, and reduce diversity.
-
Beyond Correctness: Harmonizing Process and Outcome Rewards through RL Training
PROF curates RL training data via PRM-ORM consistency to improve both final-answer accuracy and intermediate reasoning quality while reducing reliance on strong process reward models.
-
Improving Video Generation with Human Feedback
A human preference dataset and VideoReward model enable Flow-DPO and Flow-NRG to produce smoother, better-aligned videos from text prompts in flow-based generators.
-
Directly Fine-Tuning Diffusion Models on Differentiable Rewards
DRaFT fine-tunes diffusion models by differentiating through sampling to maximize rewards, outperforming RL baselines and improving aesthetics on Stable Diffusion 1.4.
-
Reinforced Self-Training (ReST) for Language Modeling
ReST improves LLM translation quality on benchmarks via offline RL on self-generated data, achieving gains in a compute-efficient way compared to typical RLHF.
-
PAPA: Online Personalized Active Preference Alignment
PAPA directly optimizes diffusion models via real-time user feedback for personalized preference alignment, drawing from variational inference, with an efficiency-enhanced variant EPAPA.
-
FlowAWR: Online Adaptive Flow Reinforcement via Advantage-Weighted Rectification
FlowAWR derives an advantage-weighted rectification for optimal velocity fields in flow models, claiming 2-5x faster convergence than DiffusionNFT on SD3.5-Medium.
-
Steerable Cultural Preference Optimization of Reward Models
SCPO is a steerable training method for reward models that improves minority cultural preference accuracy by up to 7 points and is up to 280% more data-efficient than standard finetuning on PRISM and GlobalOpinionQA datasets.
-
Scaling Novel Graph Generation via Lightweight Structure-Guided Autoregressive Models
Lightweight autoregressive graph generation model uses structure-guided ordering for efficient serialization and two-phase training to boost novelty while maintaining validity.
-
Goal-Conditioned Supervised Learning for LLM Fine-Tuning
GCSL reframes LLM fine-tuning as supervised pursuit of quality thresholds using natural-language goals, outperforming SFT and DPO on toxicity, code, and recommendation tasks.
-
ReMedi: Reasoner for Medical Clinical Prediction
ReMedi boosts LLM performance on EHR clinical predictions by up to 19.9% F1 through ground-truth-guided rationale regeneration and fine-tuning.
-
Failure Modes of Maximum Entropy RLHF
Derives SimPO from MaxEnt RL and reports that MaxEnt RL in online RLHF exhibits frequent overoptimization and unstable KL dynamics across scales, unlike stable KL-constrained baselines.
-
TrustLLM: Trustworthiness in Large Language Models
TrustLLM defines eight trustworthiness principles, creates a six-dimension benchmark, and evaluates 16 LLMs showing proprietary models generally lead but some open-source ones are close while over-calibration can hurt utility.
-
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.
-
BV-Blend: Uncertainty-Weighted Historical Baselines for Stable Critic-Free RL with Verifiable Rewards
BV-Blend blends prompt-local and semantic-cluster historical reward statistics via SEM-derived weights to stabilize critic-free RL advantage estimation.
-
Curr-RLCER:Curriculum Reinforcement Learning For Coherence Explainable Recommendation
Curr-RLCER applies curriculum reinforcement learning with coherence-driven rewards to align generated explanations with predicted ratings in explainable recommendation systems.
-
A Survey on LLM-as-a-Judge
A survey on LLM-as-a-Judge that reviews reliability strategies, proposes evaluation methods, and introduces a novel benchmark for assessing such systems.
-
A Survey of Reinforcement Learning for Large Reasoning Models
A survey compiling RL methods, challenges, data resources, and applications for enhancing reasoning in large language models and large reasoning models since DeepSeek-R1.
-
Harmful Fine-tuning Attacks and Defenses for Large Language Models: A Survey
Survey of harmful fine-tuning attacks on LLMs, their variants, defense strategies, mechanical analysis, and evaluation methodologies.
-
A Comprehensive Overview of Large Language Models
A survey paper providing an overview of Large Language Models, their background, and recent advances in the field.