REVIEW 30 cited by
Personalized Soups: Personalized Large Language Model Alignment via Post-hoc Parameter Merging
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
Personalized Soups: Personalized Large Language Model Alignment via Post-hoc Parameter Merging
read the original abstract
While Reinforcement Learning from Human Feedback (RLHF) aligns Large Language Models (LLMs) with general, aggregate human preferences, it is suboptimal for learning diverse, individual perspectives. In this work, we study Reinforcement Learning from Personalized Human Feedback (RLPHF) problem, wherein LLMs are aligned to multiple (sometimes conflicting) preferences by modeling alignment as a Multi-Objective Reinforcement Learning (MORL) problem. Compared to strong single-objective baselines, we show that we can achieve personalized alignment by decomposing preferences into multiple dimensions. These dimensions are defined based on personalizations that are declared as desirable by the user. In this work, we show that they can be efficiently trained independently in a distributed manner and combined effectively post-hoc through parameter merging. The code is available at https://github.com/joeljang/RLPHF.
Forward citations
Cited by 30 Pith papers
-
AlpsBench: An LLM Personalization Benchmark for Real-Dialogue Memorization and Preference Alignment
AlpsBench supplies 2500 real-dialogue sequences with verified memories to benchmark LLM extraction, updating, retrieval, and utilization of personalized information.
-
VitaBench 2.0: Evaluating Personalized and Proactive Agents in Long-Term User Interactions
VitaBench 2.0 introduces a benchmark for long-term personalized and proactive agent behavior, with results indicating substantial gaps in current frontier LLMs.
-
CARD: Cluster-level Adaptation with Reward-guided Decoding for Personalized Text Generation
CARD uses style-based user clustering and implicit preference contrasts to enable efficient personalized text generation via lightweight decoding adjustments on frozen LLMs.
-
CoPersona: Collaborative Persona Graphs for Robust LLM Personalization
CoPersona introduces a multiplex persona graph for facet-level peer alignment and a dual-branch retrieval-plus-reasoning architecture to improve LLM personalization under sparse and biased user interaction data.
-
Preference-Aware Rubric Learning for Personalized Evaluation
PARL formulates personalized LLM evaluation as a learning problem that induces preference-aware rubrics from raw user histories via discriminative RL and self-validation.
-
Personalized Turn-Level User Conversation Satisfaction Benchmark
Presents a memory-augmented turn-level satisfaction evaluator and PersTurnBench benchmark that improve agreement with human judgments over generic LLM judges and enable controlled model comparisons via replay.
-
Recon: Reconstruction-Guided Reasoning Synthesis for User Modeling
Recon scores reasoning traces via action reconstruction fidelity, achieving 54.7% win rate over post-hoc baselines and up to 70% when used to train synthesis models across four domains.
-
Implicit Safety Alignment from Crowd Preferences
A hierarchical framework extracts implicit safety criteria from crowd preferences and composes them via high-level policy to reduce safety violations in downstream RL tasks without explicit safety rewards.
-
Spectral Souping: A Unified Framework for Online Preference Alignment
Spectral Souping learns offline specialized policies for fine-grained preferences and merges them online using a discovered universal spectral representation for efficient LLM alignment.
-
Explaining and Breaking the Safety-Helpfulness Ceiling via Preference Dimensional Expansion
MORA breaks the safety-helpfulness trade-off in LLM alignment by pre-sampling single-reward prompts and rewriting them to expand multi-dimensional reward diversity, yielding 5-12.4% single-preference gains in sequenti...
-
Explaining and Breaking the Safety-Helpfulness Ceiling via Preference Dimensional Expansion
MORA breaks the safety-helpfulness ceiling in LLMs by pre-sampling single-reward prompts and rewriting them to incorporate multi-dimensional intents, delivering 5-12.4% gains in sequential alignment and 4.6% overall i...
-
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.
-
Conjecture and Inquiry: Quantifying Software Performance Requirements via Interactive Retrieval-Augmented Preference Elicitation
IRAP quantifies ambiguous performance requirements into mathematical functions via interactive retrieval-augmented preference elicitation and outperforms ten prior methods on four real-world datasets with up to 40x ga...
-
Beyond Compromise: Pareto-Lenient Consensus for Efficient Multi-Preference LLM Alignment
PLC uses dynamic lenient gradient updates in a game-theoretic setup to let multi-preference LLM optimization escape local equilibria and reach better global Pareto fronts.
-
PersonaVLM: Long-Term Personalized Multimodal LLMs
PersonaVLM adds memory extraction, multi-turn retrieval-based reasoning, and personality inference to multimodal LLMs, yielding 22.4% gains on a new long-term personalization benchmark and outperforming GPT-4o.
-
VC-Soup: Value-Consistency Guided Multi-Value Alignment for Large Language Models
VC-Soup uses a cosine-similarity consistency metric to filter data, trains value-consistent policies, and applies linear merging with Pareto filtering to improve multi-value LLM alignment trade-offs.
-
GDPO: Group reward-Decoupled Normalization Policy Optimization for Multi-reward RL Optimization
GDPO decouples per-reward normalization in multi-reward RL to avoid advantage collapse and improve convergence over GRPO on tool-calling, math, and coding tasks.
-
Learning from Natural Language Feedback for Personalized Question Answering
VAC replaces scalar rewards with natural language feedback in an alternating training loop between a feedback model and a policy model, yielding better personalized QA on the LaMP-QA benchmark.
-
A Roadmap to Pluralistic Alignment
The paper formalizes three types of pluralistic AI models and three benchmark classes, arguing that current alignment techniques may reduce rather than increase distributional pluralism.
-
REAR: Test-time Preference Realignment through Reward Decomposition
REAR decomposes the reward into question and preference components, rescales their balance, and expresses the result as a linear combination of token log-probabilities for efficient integration with best-of-N and tree search.
-
Personalization Meets Safety:Mechanisms,Risks,and Mitigations in Personalized LLMs
A survey that maps safety risks in personalized LLMs, introduces a unified taxonomy, and highlights three structural inadequacies in existing research on user-invariant safety, isolated techniques, and short-term evaluations.
-
From Empathy to Personalized Empathy: Adapting Empathetic Strategies to Individual Users
Introduces personalized empathy task, PersonaEmp dataset from long-term interactions, and PereGRM reward framework that combines empathy evaluation with dynamic criteria for improved adaptation to user personas.
-
Federated Variational Preference Alignment with Gumbel-Softmax Prior for Personalized User Preferences
FedVPA-GP applies variational preference learning in a federated setting with a mixture prior and orthogonal loss to disentangle user preferences on the HH-RLHF dataset.
-
In-Context Reward Adaptation for Robust Preference Modeling
Transformer model with response-time auxiliary input adapts reward models to unseen human preference domains via in-context learning from demonstrations.
-
Palette: A Modular, Controllable, and Efficient Framework for On-demand Authorized Safety Alignment Relaxation in LLMs
Palette identifies refusal directions via multi-objective search, internalizes them through lightweight adaptation, and supports on-demand multi-domain authorization via independent learning and parameter merging.
-
The Deterministic Horizon: Impossibility Results as Design Specifications for Trustworthy AI Systems
Converts impossibility theorems into architecture-dependent accuracy ceilings and design rules for transformers and other AI subfields, with the Deterministic Horizon measured at 19-31 across twelve models.
-
CLIPer: Tailoring Diverse User Preference via Classifier-Guided Inference-Time Personalization
CLIPer uses classifier guidance during inference to personalize LLM generations across single and multi-dimensional user preferences without extensive fine-tuning.
-
Test-Time Alignment via Hypothesis Reweighting
HyRe personalizes reward models at test time by reweighting an ensemble of heads trained on aggregate preferences, using few target examples to outperform uniform averaging and prior methods on RewardBench and 32 tasks.
-
Multi-Objective Exploration and Preference Optimization via Mutual Information
MI-EPO maximizes joint conditional mutual information among responses, feedback, and preference vectors, using probabilistic routing to improve alignment and controllability in multi-objective LLM optimization.
-
Model Merging in LLMs, MLLMs, and Beyond: Methods, Theories, Applications and Opportunities
The paper introduces a new taxonomy for model merging methods and reviews their applications in LLMs, MLLMs, continual learning, multi-task learning, and other subfields while outlining open challenges.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.