Author name cluster

Marc Dymetman

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

9 papers

2 author rows

ICLR Conference 2025 Conference Paper

Guaranteed Generation from Large Language Models

Minbeom Kim
Thibaut Thonet
Jos Rozen
Hwaran Lee
Kyomin Jung
Marc Dymetman

As large language models (LLMs) are increasingly used across various applications, there is a growing need to control text generation to satisfy specific constraints or requirements. This raises a crucial question: Is it possible to guarantee strict constraint satisfaction in generated outputs while preserving the distribution of the original model as much as possible? We first define the ideal distribution — the one closest to the original model, which also always satisfies the expressed constraint — as the ultimate goal of guaranteed generation. We then state a fundamental limitation, namely that it is impossible to reach that goal through autoregressive training alone. This motivates the necessity of combining training-time and inference-time methods to enforce such guarantees. Based on this insight, we propose GUARD, a simple yet effective approach that combines an autoregressive proposal distribution with rejection sampling. Through GUARD’s theoretical properties, we show how controlling the KL divergence between a specific proposal and the target ideal distribution simultaneously optimizes inference speed and distributional closeness. To validate these theoretical concepts, we conduct extensive experiments on two text generation settings with hard-to-satisfy constraints: a lexical constraint scenario and a sentiment reversal scenario. These experiments show that GUARD achieves perfect constraint satisfaction while almost preserving the ideal distribution with highly improved inference efficiency. GUARD provides a principled approach to enforcing strict guarantees for LLMs without compromising their generative capabilities.

Details

ICLR Conference 2024 Conference Paper

Compositional Preference Models for Aligning LMs

Dongyoung Go
Tomasz Korbak
Germán Kruszewski
Jos Rozen
Marc Dymetman

As language models (LMs) become more capable, it is increasingly important to align them with human preferences. However, the dominant paradigm for training Preference Models (PMs) for that purpose suffers from fundamental limitations, such as lack of transparency and scalability, along with susceptibility to overfitting the preference dataset. We propose Compositional Preference Models (CPMs), a novel PM framework that decomposes one global preference assessment into several interpretable features, obtains scalar scores for these features from a prompted LM, and aggregates these scores using a logistic regression classifier. Through these simple steps, CPMs allow to control which properties of the preference data are used to train the preference model and to build it based on features that are believed to underlie the human preference judgment. Our experiments show that CPMs not only improve generalization and are more robust to overoptimization than standard PMs, but also that best-of-n samples obtained using CPMs tend to be preferred over samples obtained using conventional PMs. Overall, our approach demonstrates the benefits of endowing PMs with priors about which features determine human preferences while relying on LM capabilities to extract those features in a scalable and robust way.

Details

ICML Conference 2023 Conference Paper

Aligning Language Models with Preferences through f-divergence Minimization

Dongyoung Go
Tomasz Korbak
Germán Kruszewski
Jos Rozen
Nahyeon Ryu
Marc Dymetman

Aligning language models with preferences can be posed as approximating a target distribution representing some desired behavior. Existing approaches differ both in the functional form of the target distribution and the algorithm used to approximate it. For instance, Reinforcement Learning from Human Feedback (RLHF) corresponds to minimizing a reverse KL from an implicit target distribution arising from a KL penalty in the objective. On the other hand, Generative Distributional Control (GDC) has an explicit target distribution and minimizes a forward KL from it using the Distributional Policy Gradient (DPG) algorithm. In this paper, we propose a new approach, $f$-DPG, which allows the use of any $f$-divergence to approximate any target distribution that can be evaluated. $f$-DPG unifies both frameworks (RLHF, GDC) and the approximation methods (DPG, RL with KL penalties). We show the practical benefits of various choices of divergence objectives and demonstrate that there is no universally optimal objective but that different divergences present different alignment and diversity trade-offs. We show that Jensen-Shannon divergence strikes a good balance between these objectives, and frequently outperforms forward KL divergence by a wide margin, leading to significant improvements over prior work. These distinguishing characteristics between divergences persist as the model size increases, highlighting the importance of selecting appropriate divergence objectives.

Details

TMLR Journal 2022 Journal Article

An approximate sampler for energy-based models with divergence diagnostics

Bryan Eikema
Germán Kruszewski
Christopher R Dance
Hady Elsahar
Marc Dymetman

Energy-based models (EBMs) allow flexible specifications of probability distributions. However, sampling from EBMs is non-trivial, usually requiring approximate techniques such as Markov chain Monte Carlo (MCMC). A major downside of MCMC sampling is that it is often impossible to compute the divergence of the sampling distribution from the target distribution: therefore, the quality of the samples cannot be guaranteed. Here, we introduce quasi-rejection sampling (QRS), a simple extension of rejection sampling that performs approximate sampling, but, crucially, does provide divergence diagnostics (in terms of f-divergences, such as KL divergence and total variation distance). We apply QRS to sampling from discrete EBMs over text for controlled generation. We show that we can sample from such EBMs with arbitrary precision in exchange for sampling efficiency and quantify the trade-off between the two by means of the aforementioned diagnostics.

PDF Details

ICML Conference 2022 Conference Paper

Controlling Conditional Language Models without Catastrophic Forgetting

Tomasz Korbak
Hady Elsahar
Germán Kruszewski
Marc Dymetman

Machine learning is shifting towards general-purpose pretrained generative models, trained in a self-supervised manner on large amounts of data, which can then be applied to solve a large number of tasks. However, due to their generic training methodology, these models often fail to meet some of the downstream requirements (e. g. , hallucinations in abstractive summarization or style violations in code generation). This raises the important question of how to adapt pre-trained generative models to meet all requirements without destroying their general capabilities ("catastrophic forgetting"). Recent work has proposed to solve this problem by representing task-specific requirements through energy-based models (EBMs) and approximating these EBMs using distributional policy gradients (DPG). Despite its effectiveness, this approach is however limited to unconditional distributions. In this paper, we extend DPG to conditional tasks by proposing Conditional DPG (CDPG). We evaluate CDPG on four different control objectives across three tasks (translation, summarization and code generation) and two pretrained models (T5 and GPT-Neo). Our results show that fine-tuning using CDPG robustly moves these pretrained models closer towards meeting control objectives and — in contrast with baseline approaches — does not result in catastrophic forgetting.

Details

NeurIPS Conference 2022 Conference Paper

On Reinforcement Learning and Distribution Matching for Fine-Tuning Language Models with no Catastrophic Forgetting

Tomasz Korbak
Hady Elsahar
Germán Kruszewski
Marc Dymetman

The availability of large pre-trained models is changing the landscape of Machine Learning research and practice, moving from a "training from scratch" to a "fine-tuning'' paradigm. While in some applications the goal is to "nudge'' the pre-trained distribution towards preferred outputs, in others it is to steer it towards a different distribution over the sample space. Two main paradigms have emerged to tackle this challenge: Reward Maximization (RM) and, more recently, Distribution Matching (DM). RM applies standard Reinforcement Learning (RL) techniques, such as Policy Gradients, to gradually increase the reward signal. DM prescribes to first make explicit the target distribution that the model is fine-tuned to approximate. Here we explore the theoretical connections between the two paradigms and show that methods such as KL-control developed in the RM paradigm can also be construed as belonging to DM. We further observe that while DM differs from RM, it can suffer from similar training difficulties, such as high gradient variance. We leverage connections between the two paradigms to import the concept of baseline into DM methods. We empirically validate the benefits of adding a baseline on an array of controllable language generation tasks such as constraining topic, sentiment, and gender distributions in texts sampled from a language model. We observe superior performance in terms of constraint satisfaction, stability, and sample efficiency.

PDF Details

ICLR Conference 2021 Conference Paper

A Distributional Approach to Controlled Text Generation

Muhammad Khalifa
Hady Elsahar
Marc Dymetman

We propose a Distributional Approach for addressing Controlled Text Generation from pre-trained Language Models (LM). This approach permits to specify, in a single formal framework, both “pointwise’” and “distributional” constraints over the target LM — to our knowledge, the first model with such generality —while minimizing KL divergence from the initial LM distribution. The optimal target distribution is then uniquely determined as an explicit EBM (Energy-BasedModel) representation. From that optimal representation, we then train a target controlled Autoregressive LM through an adaptive distributional variant of PolicyGradient. We conduct a first set of experiments over pointwise constraints showing the advantages of our approach over a set of baselines, in terms of obtaining a controlled LM balancing constraint satisfaction with divergence from the pretrained LM. We then perform experiments over distributional constraints, a unique feature of our approach, demonstrating its potential as a remedy to the problem of Bias in Language Models. Through an ablation study, we show the effectiveness of our adaptive technique for obtaining faster convergence. Code available at https://github.com/naver/gdc

Details

IJCAI Conference 2017 Conference Paper

Symbolic Priors for RNN-based Semantic Parsing

Chunyang Xiao
Marc Dymetman
Claire Gardent

Seq2seq models based on Recurrent Neural Networks (RNNs) have recently received a lot of attention in the domain of Semantic Parsing. While in principle they can be trained directly on pairs (natural language utterances, logical forms), their performance is limited by the amount of available data. To alleviate this problem, we propose to exploit various sources of prior knowledge: the well-formedness of the logical forms is modeled by a weighted context-free grammar; the likelihood that certain entities present in the input utterance are also present in the logical form is modeled by weighted finite-state automata. The grammar and automata are combined together through an efficient intersection algorithm to form a soft guide (“background”) to the RNN. We test our method on an extension of the Overnight dataset and show that it not only strongly improves over an RNN baseline, but also outperforms non-RNN models based on rich sets of hand-crafted features.

PDF Details

IJCAI Conference 1997 Conference Paper

Charts, interaction-free grammars, and the compact representation of ambiguity

Marc Dymetman

Recently researchers working in the LFG framework have proposed algorithms for taking advantage of the implicit context-free components of a unification grammar [Maxwell and Kaplan, 1996]. This paper clarifies the mathematical foundations of these techniques, provides a uniform framework in which they can be formally studied and eliminates the need for special purpose runtime datastructures recording ambiguity. The paper posits the identity: Ambiguous Feature Structures = Grammars, which states that (finitely) ambiguous representations are best seen as unification grammars of a certain type, here called "interactionfree" grammars, which generate in a backtrackfree way each of the feature structures subsumed by the ambiguous representation. This work extends a line of research [Billot and Lang, 1989; Lang, 1994] which stresses the connection between charts and grammars: a chart can be seen as a specialization of the reference grammar for a given input string. We show how this specialization grammar can be transformed into an interaction-free form which has the same practicality as a listing of the individual solutions, but is produced in less time and space.

PDF