14 papers
Formalizing Mathematics at Scale
Ahmad Rammal, Niket Patel, Fabian Gloeckle +5
We present AutoformBot, a multi-agent system for building an Autoformalized Textbook Library At Scale (Atlas) in Lean 4. AutoformBot orchestrates thousands of LLM agents, equipped…
Efficient RL Training for LLMs with Experience Replay
Charles Arnal, Vivien Cabannes, Taco Cohen +2
While Experience Replay - the practice of storing rollouts and reusing them multiple times during training - is a foundational technique in general RL, it remains largely unexplore…
Automatic Textbook Formalization
Fabian Gloeckle, Ahmad Rammal, Charles Arnal +4
We present a case study where an automatic AI system formalizes a textbook with more than 500 pages of graduate-level algebraic combinatorics to Lean. The resulting formalization r…
Provable Benefits of In-Tool Learning for Large Language Models
Sam Houliston, Ambroise Odonnat, Charles Arnal +1
Tool-augmented language models, equipped with retrieval, memory, or external APIs, are reshaping AI, yet their theoretical advantages remain underexplored. In this paper, we addres…
Asymmetric REINFORCE for off-Policy Reinforcement Learning: Balancing positive and negative rewards
Charles Arnal, Gaëtan Narozniak, Vivien Cabannes +3
Reinforcement learning (RL) is increasingly used to align large language models (LLMs). Off-policy methods offer greater implementation simplicity and data efficiency than on-polic…
Unveiling Simplicities of Attention: Adaptive Long-Context Head Identification
Konstantin Donhauser, Charles Arnal, Mohammad Pezeshki +3
The ability to process long contexts is crucial for many natural language processing tasks, yet it remains a significant challenge. While substantial progress has been made in enha…
Easing Optimization Paths: a Circuit Perspective
Ambroise Odonnat, Wassim Bouaziz, Vivien Cabannes
Gradient descent is the method of choice for training large artificial intelligence systems. As these systems become larger, a better understanding of the mechanisms behind gradien…
A Mechanistic Study of Transformers Training Dynamics
Ambroise Odonnat, Wassim Bouaziz, Vivien Cabannes
Large-scale pretraining of transformers has been central to the success of foundation models. However, the scale of those models limits our understanding of the mechanisms at play…
Iteration Head: A Mechanistic Study of Chain-of-Thought
Vivien Cabannes, Charles Arnal, Wassim Bouaziz +3
Chain-of-Thought (CoT) reasoning is known to improve Large Language Models both empirically and in terms of theoretical approximation power. However, our understanding of the inner…
Learning Associative Memories with Gradient Descent
Vivien Cabannes, Berfin Simsek, Alberto Bietti
This work focuses on the training dynamics of one associative memory module storing outer products of token embeddings. We reduce this problem to the study of a system of particles…
Mode Estimation with Partial Feedback
Charles Arnal, Vivien Cabannes, Vianney Perchet
The combination of lightly supervised pre-training and online fine-tuning has played a key role in recent AI developments. These new learning pipelines call for new theoretical fra…
Touring sampling with pushforward maps
Vivien Cabannes, Charles Arnal
The number of sampling methods could be daunting for a practitioner looking to cast powerful machine learning methods to their specific problem. This paper takes a theoretical stan…
Scaling Laws for Associative Memories
Vivien Cabannes, Elvis Dohmatob, Alberto Bietti
Learning arguably involves the discovery and memorization of abstract rules. The aim of this paper is to study associative memory mechanisms. Our model is based on high-dimensional…
Open Problem: Learning with Variational Objectives on Measures
Vivien Cabannes, Carles Domingo-Enrich
The theory of statistical learning has focused on variational objectives expressed on functions. In this note, we discuss motivations to write similar objectives on measures, in pa…