Alessandro Sordoni

2025-12-11

ArXiv (preprint)

Effect of Document Packing on the Latent Multi-Hop Reasoning Capabilities of Large Language Models

The standard practice for training large language models involves packing multiple documents together to optimize computational efficiency. … (see more)However, the impact of this process on the models' capabilities remains largely unexplored. To address this gap, we investigate how different document-packing strategies influence the latent multi-hop reasoning abilities of LLMs. Our findings indicate that packing can improve model performance compared to training on individual documents, at the expense of more compute. To further understand the underlying mechanisms, we conduct an ablation study, identifying key factors that explain the advantages of packing. Ultimately, our research deepens the understanding of LLM training dynamics and provides practical insights for optimizing model development.

2025-11-30

arXiv (published)

Gistify! Codebase-Level Understanding via Runtime Execution

Hyunji Lee

Minseon Kim

Chinmay Singh

Matheus Pereira

Atharv Sonwane

Isadora White

Elias Stengel-Eskin

Mohit Bansal

Zhengyan Shi

Marc-Alexandre Côté

Xingdi Yuan

Lucas Caccia

As coding agents are increasingly deployed in large codebases, the need to automatically design challenging, codebase-level evaluation is ce… (see more)ntral. We propose Gistify, a task where a coding LLM must create a single, minimal, self-contained file that can reproduce a specific functionality of a codebase. The coding LLM is given full access to a codebase along with a specific entrypoint (e.g., a python command), and the generated file must replicate the output of the same command ran under the full codebase, while containing only the essential components necessary to execute the provided command. Success on Gistify requires both structural understanding of the codebase, accurate modeling of its execution flow as well as the ability to produce potentially large code patches. Our findings show that current state-of-the-art models struggle to reliably solve Gistify tasks, especially ones with long executions traces.

2025-10-29

ArXiv (preprint)

BugPilot: Complex Bug Generation for Efficient Learning of SWE Skills

Atharv Sonwane

Isadora White

Hyunji Lee

Matheus Pereira

Lucas Caccia

Minseon Kim

Zhengyan Shi

Chinmay Singh

Marc-Alexandre Cot'e

Xingdi Yuan

2025-10-22

ArXiv (preprint)

The Markovian Thinker

Milad Aghajohari

Kamran Chitsaz

Reinforcement learning (RL) has recently become a strong recipe for training reasoning LLMs that produce long chains of thought (LongCoT). Y… (see more)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.

2025-10-08

ArXiv (preprint)

The Markovian Thinker

Milad Aghajohari

Kamran Chitsaz

Reinforcement learning (RL) has recently become a strong recipe for training reasoning LLMs that produce long chains of thought (LongCoT). Y… (see more)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.

2025-10-08

ArXiv (preprint)

VinePPO: Refining Credit Assignment in RL Training of LLMs

Large language models (LLMs) are increasingly applied to complex reasoning tasks that require executing several complex steps before receivi… (see more)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.

2025-10-06

Proceedings of the 42nd International Conference on Machine Learning (published)

proceedings.mlr.press

VinePPO: Refining Credit Assignment in RL Training of LLMs

Large language models (LLMs) are increasingly applied to complex reasoning tasks that require executing several complex steps before receivi… (see more)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.

2025-10-06

Proceedings of the 42nd International Conference on Machine Learning (published)

proceedings.mlr.press

openreview.net

Exploring Sparse Adapters for Scalable Merging of Parameter Efficient Experts

Minseon Kim

Riyasat Ohib

Lucas Caccia

Merging parameter-efficient task experts has recently gained growing attention as a way to build modular architectures that can be rapidly a… (see more)dapted on the fly for specific downstream tasks, without requiring additional fine-tuning. Typically, LoRA serves as the foundational building block of such parameter-efficient modular architectures, leveraging low-rank weight structures to reduce the number of trainable parameters. In this paper, we study the properties of sparse adapters, which train only a subset of weights in the base neural network, as potential building blocks of modular architectures. First, we propose a simple method for training highly effective sparse adapters, which is conceptually simpler than existing methods in the literature and surprisingly outperforms both LoRA and full fine-tuning in our setting. Next, we investigate the merging properties of these sparse adapters by merging adapters for up to 20 natural language processing tasks, thus scaling beyond what is usually studied in the literature. Our findings demonstrate that sparse adapters yield superior in-distribution performance post-merging compared to LoRA or full model merging. Achieving strong held-out performance remains a challenge for all methods considered.

2025-07-09

ArXiv (preprint)

Instilling Parallel Reasoning into Language Models

Matthew Macfarlane

Minseon Kim

Nebojsa Jojic

Weijia Xu

Lucas Caccia

Xingdi Yuan

Wanru Zhao

Zhengyan Shi

Sequential chain-of-thought reasoning significantly improves the performance of Large language models (LLMs) on complex tasks. However, sequ… (see more)ential reasoning has structural limitations: Long chains are expensive due to attention's quadratic complexity, and multiple diverse strategies cannot be considered simultaneously. To address this we propose a method that instills parallel reasoning capabilities in LLMs by distilling parallel reasoning traces from a teacher model. This approach enables models to decompose problems, explore diverse strategies via concurrent reasoning traces, and aggregate trace outputs for the final answer. Evaluating on a variety of math and puzzle benchmarks such as MATH 500, AIME and Countdown, we show our approach can decompose parallelizable problems, and that the performance scales with the number of parallel traces. The resulting model can dynamically allocate reasoning strategies based on problem complexity, outperforming standard sampling methods.

2025-07-09

ICML.cc/2025/Workshop/AI4MATH (poster)

openreview.net

Learning to Solve Complex Problems via Dataset Decomposition

Wanru Zhao

Lucas Caccia

Zhengyan Shi

Minseon Kim

Xingdi Yuan

Weijia Xu

Marc-Alexandre Côté

Curriculum learning is a class of training strategies that organizes the data being exposed to a model by difficulty, gradually from simpler… (see more) to more complex examples. This research explores a reverse curriculum generation approach that recursively decomposes complex datasets into simpler, more learnable components. We propose a teacher-student framework where the teacher is equipped with the ability to reason step-by-step, which is used to recursively generate easier versions of examples, enabling the student model to progressively master difficult tasks. We propose a novel scoring system to measure data difficulty based on its structural complexity and conceptual depth, allowing curriculum construction over decomposed data. Experiments on math datasets (MATH and AIME) demonstrate that models trained with curricula generated by our approach exhibit superior performance compared to standard training on original datasets.

2025-07-09

ICML.cc/2025/Workshop/AI4MATH (poster)

openreview.net

Medical Red Teaming Protocol of Language Models: On the Importance of User Perspectives in Healthcare Settings

Jean-Philippe Corbeil

Minseon Kim

Francois Beaulieu

Paul Vozila

2025-07-09

ArXiv (preprint)