Oleksiy Ostapenko

DiffuMamba: High-Throughput Diffusion LMs with Mamba Backbone

Pierre-Andre Noel

Torsten Scholak

Diffusion language models (DLMs) have emerged as a promising alternative to autoregressive (AR) generation, yet their reliance on Transforme… (see more)r backbones limits inference efficiency due to quadratic attention or KV-cache overhead. We introduce DiffuMamba, a masked diffusion language model built on a bidirectional Mamba backbone that combines the diffusion objective with linear-time sequence modeling, and DiffuMamba-H, a hybrid variant with interleaved attention. Across scales up to 1.3B parameters, our models match Transformer-based diffusion in downstream performance while achieving up to 8.2× and 4.3× higher inference throughput, respectively, on long sequences. We further present a systematic analysis of inference efficiency across modern DLM variants, combining asymptotic complexity with empirical measurements. Notably, cache-efficient block diffusion with Mamba mixers emerges as the only strategy that scales linearly with sequence length and achieves the strongest performance across all baselines, suggesting a promising direction for future diffusion-based generation systems.

2026-03-01

MM_Intelligence @ International Conference on Learning Representations (poster)

doi.org

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 (Low-Rank Adaptation) 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.

2025-07-06

colmweb.org/COLM/2025/Conference (accepted)

doi.org

openreview.net

Guiding Language Model Reasoning with Planning Tokens

Xinyi Wang

Lucas Caccia

Oleksiy Ostapenko

Xingdi Yuan

William Yang Wang

Alessandro Sordoni

Large language models (LLMs) have recently attracted considerable interest for their ability to perform complex reasoning tasks, such as cha… (see more)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 (accepted)

doi.org

openreview.net

Towards Modular LLMs by Building and Reusing a Library of LoRAs

Matheus Pereira

Lucas Caccia

Alessandro Sordoni

The growing number of parameter-efficient adaptations of a base large language model (LLM) calls for studying whether we can reuse such trai… (see more)ned adapters to improve performance for new tasks. We study how to best build a library of adapters given multi-task data and devise techniques for both zero-shot and supervised task generalization through routing in such library. We benchmark existing approaches to build this library and introduce model-based clustering, MBC, a method that groups tasks based on the similarity of their adapter parameters, indirectly optimizing for transfer across the multi-task dataset. To re-use the library, we present a novel zero-shot routing mechanism, Arrow, which enables dynamic selection of the most relevant adapters for new inputs without the need for retraining. We experiment with several LLMs, such as Phi-2 and Mistral, on a wide array of held-out tasks, verifying that MBC-based adapters and Arrow routing lead to superior generalization to new tasks. We make steps towards creating modular, adaptable LLMs that can match or outperform traditional joint training.

2023-12-31

ICML (published)

doi.org

proceedings.mlr.press

Challenging Common Assumptions About Catastrophic Forgetting and Knowledge Accumulation

Timothee LESORT

Pau Rodríguez

Md Rifat Arefin

Building learning agents that can progressively learn and accumulate knowledge is the core goal of the continual learning (CL) research fiel… (see more)d. Unfortunately, training a model on new data usually compromises the performance on past data. In the CL literature, this effect is referred to as catastrophic forgetting (CF). CF has been largely studied, and a plethora of methods have been proposed to address it on short sequences of non-overlapping tasks. In such setups, CF always leads to a quick and significant drop in performance in past tasks. Nevertheless, despite CF, recent work showed that SGD training on linear models accumulates knowledge in a CL regression setup. This phenomenon becomes especially visible when tasks reoccur. We might then wonder if DNNs trained with SGD or any standard gradient-based optimization accumulate knowledge in such a way. Such phenomena would have interesting consequences for applying DNNs to real continual scenarios. Indeed, standard gradient-based optimization methods are significantly less computationally expensive than existing CL algorithms. In this paper, we study the progressive knowledge accumulation (KA) in DNNs trained with gradient-based algorithms in long sequences of tasks with data re-occurrence. We propose a new framework, SCoLe (Scaling Continual Learning), to investigate KA and discover that catastrophic forgetting has a limited effect on DNNs trained with SGD. When trained on long sequences with data sparsely re-occurring, the overall accuracy improves, which might be counter-intuitive given the CF phenomenon. We empirically investigate KA in DNNs under various data occurrence frequencies and propose simple and scalable strategies to increase knowledge accumulation in DNNs.

2023-11-19

Proceedings of The 2nd Conference on Lifelong Learning Agents (published)

doi.org

proceedings.mlr.press

From IID to the Independent Mechanisms assumption in continual learning

Oleksiy Ostapenko

Pau Rodríguez

Alexandre Lacoste

Laurent Charlin

2023-01-10

AAAI.org/2023/Bridge/CCBridge (accepted)

proceedings.mlr.press

Guiding Language Model Math Reasoning with Planning Tokens

Xinyi Wang

Lucas Caccia

Oleksiy Ostapenko

Xingdi Yuan

William Yang Wang

Alessandro Sordoni

Large language models (LLMs) have recently attracted considerable interest for their ability to perform complex reasoning tasks, such as cha… (see more)in-of-thought 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. We find that while LLMs can manage individual reasoning steps well, they struggle with maintaining consistency across an entire reasoning chain. To solve this, we introduce planning tokens at the start of each reasoning step, serving as a guide for the model, and add their embeddings to the model parameters. Our approach requires a negligible increase in trainable parameters (just 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 w.r.t. standard fine-tuning baselines.

2022-12-31

arXiv.org (preprint)

doi.org

arxiv.org

Continual Learning with Foundation Models: An Empirical Study of Latent Replay

Oleksiy Ostapenko

Timothee LESORT

Pau Rodríguez

Md Rifat Arefin

Arthur Douillard

Irina Rish

Laurent Charlin

Rapid development of large-scale pre-training has resulted in foundation models that can act as effective feature extractors on a variety of… (see more) downstream tasks and domains. Motivated by this, we study the efficacy of pre-trained vision models as a foundation for downstream continual learning (CL) scenarios. Our goal is twofold. First, we want to understand the compute-accuracy trade-off between CL in the raw-data space and in the latent space of pre-trained encoders. Second, we investigate how the characteristics of the encoder, the pre-training algorithm and data, as well as of the resulting latent space affect CL performance. For this, we compare the efficacy of various pre-trained models in large-scale benchmarking scenarios with a vanilla replay setting applied in the latent and in the raw-data space. Notably, this study shows how transfer, forgetting, task similarity and learning are dependent on the input data characteristics and not necessarily on the CL algorithms. First, we show that under some circumstances reasonable CL performance can readily be achieved with a non-parametric classifier at negligible compute. We then show how models pre-trained on broader data result in better performance for various replay sizes. We explain this with representational similarity and transfer properties of these representations. Finally, we show the effectiveness of self-supervised pre-training for downstream domains that are out-of-distribution as compared to the pre-training domain. We point out and validate several research directions that can further increase the efficacy of latent CL including representation ensembling. The diverse set of datasets used in this study can serve as a compute-efficient playground for further CL research. The codebase is available under https://github.com/oleksost/latent_CL.

2022-11-27

Proceedings of The 1st Conference on Lifelong Learning Agents (published)

doi.org

proceedings.mlr.press

Attention for Compositional Modularity

Oleksiy Ostapenko

Pau Rodríguez

Alexandre Lacoste

Laurent Charlin

Modularity and compositionality are promising inductive biases for addressing longstanding problems in machine learning such as better syste… (see more)matic generalization, as well as better transfer and lower forgetting in the context of continual learning. Here we study how attention-based module selection can help achieve composi-tonal modularity – i.e. decomposition of tasks into meaningful sub-tasks which are tackled by independent architectural entities that we call modules. These sub-tasks must be reusable and the system should be able to learn them without additional supervision. We design a simple experimental setup in which the model is trained to solve mathematical equations with multiple math operations applied sequentially. We study different attention-based module selection strategies, inspired by the principles introduced in the recent literature. We evaluate the method’s ability to learn modules that can recover the underling sub-tasks (operation) used for data generation, as well as the ability to generalize compositionally. We find that meaningful module selection (i.e. routing) is the key to compositional generalization. Further, without access to the privileged information about which part of the input should be used for module selection, the routing component performs poorly for samples that are compositionally out of training distribution. We find that the the main reason for this lies in the routing component, since many of the tested methods perform well OOD if we report the performance of the best performing path at test time. Additionally, we study the role of the number of primitives, the number of training points and bottlenecks for modular specialization.

2022-10-19

NeurIPS.cc/2022/Workshop/Attention (poster)

openreview.net

Sequoia: A Software Framework to Unify Continual Learning Research

Pau Rodríguez

J. Hurtado

Dominic Zhao

Ryan Lindeborg

Timothee LESORT

David Vázquez

Laurent Charlin

Irina Rish

Massimo Caccia

The field of Continual Learning (CL) seeks to develop algorithms that accumulate knowledge and skills over time through interaction with non… (see more)-stationary environments. In practice, a plethora of evaluation procedures (settings) and algorithmic solutions (methods) exist, each with their own potentially disjoint set of assumptions. This variety makes measuring progress in CL difficult. We propose a taxonomy of settings, where each setting is described as a set of assumptions. A tree-shaped hierarchy emerges from this view, where more general settings become the parents of those with more restrictive assumptions. This makes it possible to use inheritance to share and reuse research, as developing a method for a given setting also makes it directly applicable onto any of its children. We instantiate this idea as a publicly available software framework called Sequoia, which features a wide variety of settings from both the Continual Supervised Learning (CSL) and Continual Reinforcement Learning (CRL) domains. Sequoia also includes a growing suite of methods which are easy to extend and customize, in addition to more specialized methods from external libraries. We hope that this new paradigm and its first implementation can help unify and accelerate research in CL. You can help us grow the tree by visiting www.github.com/lebrice/Sequoia.

2021-08-01

ArXiv (preprint)

doi.org

openreview.net

Continual Learning via Local Module Composition

Oleksiy Ostapenko

Pau Rodríguez

Massimo Caccia

Laurent Charlin

Modularity is a compelling solution to continual learning (CL), the problem of modeling sequences of related tasks. Learning and then compos… (see more)ing modules to solve different tasks provides an abstraction to address the principal challenges of CL including catastrophic forgetting, backward and forward transfer across tasks, and sub-linear model growth. We introduce local module composition (LMC), an approach to modular CL where each module is provided a local structural component that estimates a module's relevance to the input. Dynamic module composition is performed layer-wise based on local relevance scores. We demonstrate that agnosticity to task identities (IDs) arises from (local) structural learning that is module-specific as opposed to the task- and/or model-specific as in previous works, making LMC applicable to more CL settings compared to previous works. In addition, LMC also tracks statistics about the input distribution and adds new modules when outlier samples are detected. In the first set of experiments, LMC performs favorably compared to existing methods on the recent Continual Transfer-learning Benchmark without requiring task identities. In another study, we show that the locality of structural learning allows LMC to interpolate to related but unseen tasks (OOD), as well as to compose modular networks trained independently on different task sequences into a third modular network without any fine-tuning. Finally, in search for limitations of LMC we study it on more challenging sequences of 30 and 100 tasks, demonstrating that local module selection becomes much more challenging in presence of a large number of candidate modules. In this setting best performing LMC spawns much fewer modules compared to an oracle based baseline, however, it reaches a lower overall accuracy. The codebase is available under https://github.com/oleksost/LMC.

2020-12-31

Advances in Neural Information Processing Systems 34 (NeurIPS 2021) (published)

doi.org

openreview.net

Online Fast Adaptation and Knowledge Accumulation (OSAKA): a New Approach to Continual Learning

Massimo Caccia

Pau Rodríguez

Lucas Caccia

Alexandre Lacoste

Continual learning studies agents that learn from streams of tasks without forgetting previous ones while adapting to new ones. Two recent c… (see more)ontinual-learning scenarios have opened new avenues of research. In meta-continual learning, the model is pre-trained to minimize catastrophic forgetting of previous tasks. In continual-meta learning, the aim is to train agents for faster remembering of previous tasks through adaptation. In their original formulations, both methods have limitations. We stand on their shoulders to propose a more general scenario, OSAKA, where an agent must quickly solve new (out-of-distribution) tasks, while also requiring fast remembering. We show that current continual learning, meta-learning, meta-continual learning, and continual-meta learning techniques fail in this new scenario. We propose Continual-MAML, an online extension of the popular MAML algorithm as a strong baseline for this scenario. We empirically show that Continual-MAML is better suited to the new scenario than the aforementioned methodologies, as well as standard continual learning and meta-learning approaches.

2019-12-31

NeurIPS (published)

doi.org

arxiv.org

Mila Ventures Founder in Residence

TRAIL: Responsible AI for Professionals and Leaders

AI Advantage: Productivity in Public Service

Oleksiy Ostapenko

Publications

Mila Ventures Founder in Residence

TRAIL: Responsible AI for Professionals and Leaders

AI Advantage: Productivity in Public Service

Popular keywords:

Oleksiy Ostapenko

Publications