Alessandro Sordoni

Marc-Alexandre Côté

2025-03-26

ArXiv (prépublication)

arxiv.org

Training Plug-n-Play Knowledge Modules with Deep Context Distillation

Lucas Caccia

Alan Ansell

Edoardo Ponti

Ivan Vulić

2025-03-04

ICLR.cc/2025/Workshop/MCDC (accepté)

A Survey on Model MoErging: Recycling and Routing Among Specialized Experts for Collaborative Learning

Prateek Yadav

Colin Raffel

Mohammed Muqeeth

Lucas Caccia

Haokun Liu

Tianlong Chen

Mohit Bansal

Leshem Choshen

The availability of performant pre-trained models has led to a proliferation of fine-tuned expert models that are specialized to a particula… (voir plus)r domain or task. Model MoErging methods aim to recycle expert models to create an aggregate system with improved performance or generalization. A key component of MoErging methods is the creation of a router that decides which expert model(s) to use for a particular input or application. The promise, effectiveness, and large design space of MoErging has spurred the development of many new methods over the past few years. This rapid pace of development has made it challenging to compare different MoErging methods, which are rarely compared to one another and are often validated in different experimental setups. To remedy such gaps, we present a comprehensive survey of MoErging methods that includes a novel taxonomy for cataloging key design choices and clarifying suitable applications for each method. Apart from surveying MoErging research, we inventory software tools and applications that make use of MoErging. We additionally discuss related fields of study such as model merging, multitask learning, and mixture-of-experts models. Taken as a whole, our survey provides a unified overview of existing MoErging methods and creates a solid foundation for future work in this burgeoning field.

2024-12-31

Trans. Mach. Learn. Res. (publié)

The Markovian Thinker: Architecture-Agnostic Linear Scaling of Reasoning

Milad Aghajohari

Kamran Chitsaz

Amirhossein Kazemnejad

Reinforcement learning (RL) has recently become a strong recipe for training reasoning LLMs that produce long chains of thought (LongCoT). Y… (voir plus)et the standard RL"thinking environment", where the state is the prompt plus all prior reasoning tokens, makes the state unbounded and forces attention-based policies to pay quadratic compute as thoughts lengthen. We revisit the environment itself. We propose Markovian Thinking, a paradigm in which the policy advances reasoning while conditioning on a constant-size state, decoupling thinking length from context size. As an immediate consequence this yields linear compute with constant memory. We instantiate this idea with Delethink, an RL environment that structures reasoning into fixed-size chunks. Within each chunk, the model thinks as usual; at the boundary, the environment resets the context and reinitializes the prompt with a short carryover. Through RL, the policy learns to write a textual state near the end of each chunk sufficient for seamless continuation of reasoning after reset. Trained in this environment, an R1-Distill 1.5B model reasons in 8K-token chunks yet thinks up to 24K tokens, matching or surpassing LongCoT-RL trained with a 24K budget. With test-time scaling, Delethink continues to improve where LongCoT plateaus. The effect of linear compute is substantial: we empirically estimate at 96K average thinking length LongCoT-RL costs 27 H100-months vs. 7 for Delethink. Analysis at RL initialization shows off-the-shelf reasoning models (1.5B-120B) often sample Markovian traces zero-shot across diverse benchmarks, providing positive samples that make RL effective at scale. Our results show that redesigning the thinking environment is a powerful lever: it enables very long reasoning without quadratic overhead and opens a path toward efficient, scalable reasoning LLMs.

2024-12-31

arXiv (prépublication)

Not All LLM Reasoners Are Created Equal

Daniel Toyama

We study the depth of grade-school math (GSM) problem-solving capabilities of LLMs. To this end, we evaluate their performance on pairs of e… (voir plus)xisting math word problems together so that the answer to the second problem depends on correctly answering the first problem. Our findings reveal a significant reasoning gap in most LLMs, that is performance difference between solving the compositional pairs and solving each question independently. This gap is more pronounced in smaller, more cost-efficient, and math-specialized models. Moreover, instruction-tuning recipes and code generation have varying effects across LLM sizes, while finetuning on GSM can lead to task overfitting. Our analysis indicates that large reasoning gaps are not because of test-set leakage, but due to distraction from additional context and poor second-hop reasoning. Overall, LLMs exhibit systematic differences in their reasoning abilities, despite what their performance on standard benchmarks indicates.

2024-10-08

NeurIPS.cc/2024/Workshop/Sys2-Reasoning (poster)

VinePPO: Accurate Credit Assignment in RL for LLM Mathematical Reasoning

Amirhossein Kazemnejad

Nicolas Roux

Large language models (LLMs) are increasingly required to solve complex reasoning tasks, like mathematical problems, that involve multiple r… (voir plus)easoning steps before feedback is received. Effectively identifying and prioritizing key steps by accurately assigning credit to these intermediate steps is essential for enhancing model performance. Proximal Policy Optimization (PPO), a state-of-the-art reinforcement learning algorithm for finetuning LLMs, addresses the credit assignment problem by employing value networks to predict the expected cumulative rewards of intermediate states. In this work, we identify significant limitations with this value estimation method. To address this, we propose \methodname that leverages the flexibility of language environments to compute unbiased Monte Carlo-based estimates of the intermediate values. VinePPO consistently outperforms standard PPO, doing so more efficiently and with lower divergence from the reference model. Our findings underscore the critical importance of accurate credit assignment in LLM post-training and present a simple, yet effective solution.

2024-10-08

NeurIPS.cc/2024/Workshop/MATH-AI (accepté)

VinePPO: Refining Credit Assignment in RL Training of LLMs

Amirhossein Kazemnejad

Large language models (LLMs) are increasingly applied to complex reasoning tasks that require executing several complex steps before receivi… (voir plus)ng any reward. Properly assigning credit to these steps is essential for enhancing model performance. Proximal Policy Optimization (PPO), a common reinforcement learning (RL) algorithm used for LLM finetuning, employs value networks to tackle credit assignment. However, recent approaches achieve strong results without it, raising questions about the efficacy of value networks in practice. In this work, we systematically evaluate the efficacy of value networks and reveal their significant shortcomings in reasoning-heavy LLM tasks, showing that they often produce poor estimate of expected return and barely outperform a random baseline when comparing alternative steps. This motivates our key question: Can improved credit assignment enhance RL training for LLMs? To address this, we propose VinePPO, a straightforward approach that leverages the flexibility of language environments to compute unbiased Monte Carlo-based estimates. Our method consistently outperforms PPO and other baselines across MATH and GSM8K datasets in less wall-clock time (up to 3.0x). Crucially, it achieves higher test accuracy for a given training accuracy, capturing more generalization signal per sample. These results emphasize the importance of accurate credit assignment in RL training of LLM.

2024-10-01

arXiv (prépublication)

proceedings.mlr.press

Efficient Adversarial Training in LLMs with Continuous Attacks

Stephan Günnemann

Large language models (LLMs) are vulnerable to adversarial attacks that can bypass their safety guardrails. In many domains, adversarial tra… (voir plus)ining has proven to be one of the most promising methods to reliably improve robustness against such attacks. Yet, in the context of LLMs, current methods for adversarial training are hindered by the high computational costs required to perform discrete adversarial attacks at each training iteration. We address this problem by instead calculating adversarial attacks in the continuous embedding space of the LLM, which is orders of magnitudes more efficient. We propose a fast adversarial training algorithm (C-AdvUL) composed of two losses: the first makes the model robust on continuous embedding attacks computed on an adversarial behaviour dataset; the second ensures the usefulness of the final model by fine-tuning on utility data. Moreover, we introduce C-AdvIPO, an adversarial variant of IPO that does not require utility data for adversarially robust alignment. Our empirical evaluation on five models from different families (Gemma, Phi3, Mistral, Zephyr, Llama2) and at different scales (2B, 3.8B, 7B) shows that both algorithms substantially enhance LLM robustness against discrete attacks (GCG, AutoDAN, PAIR), while maintaining utility. Our results demonstrate that robustness to continuous perturbations can extrapolate to discrete threat models. Thereby, we present a path toward scalable adversarial training algorithms for robustly aligning LLMs.

2024-09-24

NeurIPS.cc/2024/Conference (spotlight)

Improving Context-Aware Preference Modeling for Language Models

Silviu Pitis

Ziang Xiao

Nicolas Roux

While finetuning language models from pairwise preferences has proven remarkably effective, the underspecified nature of natural language pr… (voir plus)esents critical challenges. Direct preference feedback is uninterpretable, difficult to provide where multidimensional criteria may apply, and often inconsistent, either because it is based on incomplete instructions or provided by diverse principals. To address these challenges, we consider the two-step preference modeling procedure that first resolves the under-specification by selecting a context, and then evaluates preference with respect to the chosen context. We decompose reward modeling error according to these two steps, which suggests that supervising context in addition to context-specific preference may be a viable approach to aligning models with diverse human preferences. For this to work, the ability of models to evaluate context-specific preference is critical. To this end, we contribute context-conditioned preference datasets and accompanying experiments that investigate the ability of language models to evaluate context-specific preference. Unlike past datasets, where context-specific preference is highly correlated with general preference, our "preference reversal" datasets disentangle context-specific and general preferences to isolate context-specific capabilities. We use our datasets to (1) show that existing preference models benefit from, but fail to fully consider, added context, (2) finetune a context-aware reward model with context-specific performance exceeding that of GPT-4 and Llama 3 70B, and (3) investigate the potential value of context-aware preference modeling.

2024-09-24

NeurIPS.cc/2024/Conference (poster)

Guiding Language Model Reasoning with Planning Tokens

Xinyi Wang

Lucas Caccia

Oleksiy Ostapenko

Xingdi Yuan

William Yang Wang

Large language models (LLMs) have recently attracted considerable interest for their ability to perform complex reasoning tasks, such as cha… (voir plus)in-of-thought (CoT) reasoning. However, most of the existing approaches to enhance this ability rely heavily on data-driven methods, while neglecting the structural aspects of the model's reasoning capacity. To encourage a more structural generation of CoT steps, we propose a hierarchical generation scheme: we let the LM generate a planning token at the start of each reasoning step, intuitively serving as a high-level plan of the current step, and add their embeddings to the model parameters. Our approach requires a negligible increase in trainable parameters (0.001%) and can be applied through either full fine-tuning or a more parameter-efficient scheme. We demonstrate our method's effectiveness by applying it to three different LLMs, showing notable accuracy improvements across three math word problem datasets and one multihop QA dataset with respect to standard fine-tuning baselines.

2024-07-09

colmweb.org/COLM/2024/Conference (accepté)

V-STaR: Training Verifiers for Self-Taught Reasoners

Xingdi Yuan

Common self-improvement approaches for large language models (LLMs), such as STaR, iteratively fine-tune LLMs on self-generated solutions to… (voir plus) improve their problem-solving ability. However, these approaches discard the large amounts of incorrect solutions generated during this process, potentially neglecting valuable information in such solutions. To address this shortcoming, we propose V-STaR that utilizes both the correct and incorrect solutions generated during the self-improvement process to train a verifier using DPO that judges correctness of model-generated solutions. This verifier is used at inference time to select one solution among many candidate solutions. Running V-STaR for multiple iterations results in progressively better reasoners and verifiers, delivering a 4% to 17% test accuracy improvement over existing self-improvement and verification approaches on common code generation and math reasoning benchmarks with LLaMA2 models.

2024-07-09

colmweb.org/COLM/2024/Conference (accepté)

Efficient Reinforcement Learning by Discovering Neural Pathways

Samin Yeasar Arnob

Riyasat Ohib

Sergey Plis

Amy Zhang