Eugene Belilovsky

Website

Github

Xiao Huang

Master's Research - Concordia University

Co-supervisor :

Paul Janson

PhD - Concordia University

Co-supervisor :

Gwen Legate

PhD - Concordia University

Co-supervisor :

Master's Research - Concordia University

Co-supervisor :

Guy Wolf

Abhinav Moudgil

PhD - Concordia University

Website

Github

Google Scholar

Adel Nabli

PhD - Concordia University

Github

Google Scholar

Geraldin Nanfack

Postdoctorate - Concordia University

Co-supervisor :

Albert Orozco Camacho

PhD - Concordia University

Co-supervisor :

PhD - Concordia University

Co-supervisor :

Benjamin Therien

PhD - Université de Montréal

Principal supervisor :

Collaborating Alumni - Université de Montréal

Principal supervisor :

PhD - Concordia University

Co-supervisor :

Publications

From Memorization to Parameter Interference: How Overtraining Experts Harms Model Merging

Stefan Horoi

Guy Wolf

Gintare Karolina Dziugaite

Modern deep learning is increasingly characterized by the use of open-weight foundation models that can be fine-tuned on specialized dataset… (see more)s. This has led to a proliferation of expert models and adapters, often shared via platforms like HuggingFace and AdapterHub. Model merging has recently emerged as an effective way to leverage these existing resources, enabling the composition of capabilities from different model checkpoints. A natural pipeline has thus formed to harness the benefits of transfer learning and amortize sunk training costs: models are pre-trained on general data, fine-tuned on specific tasks, and then multiple checkpoints are merged to obtain a more capable model. A prevailing assumption is that improvements at one stage of this pipeline propagate downstream, leading to gains at subsequent steps. In this work, we challenge that assumption by examining how expert fine-tuning affects model merging. We show that long fine-tuning of experts that optimizes for their individual performance leads to degraded merging performance across vision and language modalities, multiple model scales, and both fully fine-tuned and LoRA-adapted models. We trace this degradation to the memorization of a small set of difficult examples that dominate late fine-tuning steps. This causes negative parameter interference and encodes knowledge that is forgotten during merging. Finally, we demonstrate that task-dependent aggressive early stopping strategies can significantly improve model merging performance.

2025-12-31

International Conference on Machine Learning (Accept (regular))

Heterogeneous Low-Bandwidth Pre-Training of LLMs

Yazan Obeidi

Amir Sarfi

Joel Lidin

Paul Janson

Pre-training large language models (LLMs) increasingly requires distributed compute, yet bandwidth constraints make it difficult to scale be… (see more)yond well-provisioned datacenters-especially when model parallelism forces frequent, large inter-device communications. We study whether SparseLoCo, a low-communication data parallel method based on infrequent synchronization and sparse pseudo-gradient exchange, can be combined with low-bandwidth pipeline model parallelism via activation and activation-gradient compression. We introduce a heterogeneous distributed training framework where some participants host full replicas on high-bandwidth interconnects, while resource-limited participants are grouped to jointly instantiate a replica using pipeline parallelism with subspace-projected inter-stage communication. To make the recently introduced subspace pipeline compression compatible with SparseLoCo, we study a number of adaptations. Across large-scale language modeling experiments (178M-1B parameters) on standard pretraining corpora, we find that activation compression composes with SparseLoCo at modest cost, while selective (heterogeneous) compression consistently improves the loss-communication tradeoff relative to compressing all replicas-especially at aggressive compression ratios. These results suggest a practical path to incorporating low-bandwidth model parallelism and heterogeneous participants into LLM pre-training.

2025-12-31

arXiv (preprint)

Towards Learned Optimization Free Lunch

Abhinav Moudgil

Boris Knyazev

Learned optimizers are powerful alternatives to hand-designed rules like Adam, yet they have seen limited practical adoption since they ofte… (see more)n fail to meta-generalize beyond their training distribution and incur high meta-training cost. For instance, prior work, VeLO, scaled meta-training to 4,000 TPU months (

2025-12-31

International Conference on Learning Representations (Accept (Poster))

Understanding and Exploiting Weight Update Sparsity for Communication-Efficient Distributed RL.

Erfan Miahi

Reinforcement learning (RL) is a critical component for post-training large language models (LLMs). However, in bandwidth-constrained distri… (see more)buted RL, scalability is often bottlenecked by the synchronization of policy weights from trainers to inference workers, particularly over commodity networks or in decentralized settings. While recent studies suggest that RL updates modify only a small fraction of model parameters, these observations are typically based on coarse checkpoint differences. We present a systematic empirical study of weight-update sparsity at both step-level and multi-step granularities, examining its evolution across training dynamics, off-policy delay, and model scale. We find that update sparsity is consistently high, frequently exceeding 99% across practically relevant settings. Leveraging this structure, we propose PULSE (Patch Updates via Lossless Sparse Encoding), a simple yet highly efficient lossless weight synchronization method that transmits only the indices and values of modified parameters. PULSE is robust to transmission errors and avoids floating-point drift inherent in additive delta schemes. In bandwidth-constrained decentralized environments, our approach achieves over 100x (14 GB to ~108 MB) communication reduction while maintaining bit-identical training dynamics and performance compared to full weight synchronization. By exploiting this structure, PULSE enables decentralized RL training to approach centralized throughput, reducing the bandwidth required for weight synchronization from 20 Gbit/s to 0.2 Gbit/s to maintain high GPU utilization.

2025-12-31

arXiv (preprint)

Continual Pre-training of MoEs: How robust is your router?

Benjamin Therien

Charles-Etienne Joseph

Zain Sarwar

Ashwinee Panda

Anirban Das

Shi-Xiong Zhang

Stephen Rawls

Sambit Sahu

2025-09-25

TMLR (accepted)

When Data Falls Short: Grokking Below the Critical Threshold

Vaibhav Singh

Rahaf Aljundi

2025-09-22

NeurIPS.cc/2025/Workshop/CCFM (poster)

ACCO: Accumulate While You Communicate for Communication-Overlapped Sharded LLM Training

Adel Nabli

Louis Fournier

Pierre Erbacher

Louis Serrano

Edouard Oyallon

Training LLMs relies on distributed implementations using multiple GPUs to compute gradients in parallel with sharded optimizers. However, s… (see more)ynchronizing gradients in data parallel setups introduces communication overhead that grows with the number of workers, limiting parallelization efficiency. Local optimization algorithms reduce communications but incur high memory costs as they prevent optimizer state sharding, hindering scalability. To address this, we propose \textbf{AC}cumulate while \textbf{CO}mmunicate (ACCO), a memory-efficient optimization algorithm for distributed LLM training. By synchronizing delayed gradients while computing new ones, ACCO reduces GPU idle time and supports heterogeneous hardware. To mitigate the convergence issues caused by delayed updates, we introduce a novel technique ensuring training dynamics align with standard distributed optimization. Compared to ZeRO-1, our approach is significantly faster and scales effectively across heterogeneous hardware.

2025-09-17

NeurIPS.cc/2025/Conference (poster)

Warming Up for Zeroth-Order Federated Pre-Training with Low Resource Clients

Gwen Legate

Federated learning enables collaborative model training across numerous edge devices without requiring participants to share data; however, … (see more)memory and communication constraints on these edge devices may preclude their participation in training. We consider a setting in which a subset of edge devices are below a critical memory or communication threshold required to conduct model updates. Under typical federated optimization algorithms, these devices are excluded from training which renders their data inaccessible and increases system induced bias. We are inspired by MeZO, a zeroth-order method used for memory-efficient fine-tuning. The increased variance inherent to zeroth-order gradient approximations has relegated previous zeroth-order optimizers exclusively to the domain of fine tuning; a limitation we seek to correct. We devise a federated, memory-efficient zeroth-order optimizer, ZOWarmUp that permits zeroth-order training from a random initialization. ZOWarmUp leverages differing client capabilities and careful variance reduction techniques to facilitate participation of under-represented, low-resource clients in model training. Like other federated zeroth-order methods, ZOWarmUp eliminates the need for edge devices to transmit their full gradients to the server and instead relies on only a small set of random seeds, rendering the up-link communication cost negligible. We present experiments using various datasets and model architectures to show that ZOWarmUp is a robust algorithm that can can be applied under a wide variety of circumstances. For systems with a high proportion of edge devices that would otherwise be excluded from training, this algorithm provides access to a greater volume and diversity of data, thus improving training outcomes.

2025-09-02

ArXiv (preprint)

Communication Efficient LLM Pre-training with SparseLoCo

Amir M. Sarfi

Benjamin Therien

Joel Lidin

2025-08-20

ArXiv (preprint)

Rethinking Prompt Optimization: Reinforcement, Diversification, and Migration in Blackbox LLMs

MohammadReza Davari

Utkarsh Garg

Weixin Cai

2025-07-13

ArXiv (preprint)

Circuit Discovery Helps To Detect LLM Jailbreaking

Despite extensive safety alignment, large language models (LLMs) remain vulnerable to jailbreak attacks that bypass safeguards to elicit har… (see more)mful content. While prior work attributes this vulnerability to safety training limitations, the internal mechanisms by which LLMs process adversarial prompts remain poorly understood. We present a mechanistic analysis of the jailbreaking behavior in a large-scale, safety-aligned LLM, focusing on LLaMA-2-7B-chat-hf. Leveraging edge attribution patching and subnetwork probing, we systematically identify computational circuits responsible for generating affirmative responses to jailbreak prompts. Ablating these circuits during the first token prediction can reduce attack success rates by up to 80\%, demonstrating its critical role in safety bypass. Our analysis uncovers key attention heads and MLP pathways that mediate adversarial prompt exploitation, revealing how important tokens propagate through these components to override safety constraints. These findings advance the understanding of adversarial vulnerabilities in aligned LLMs and pave the way for targeted, interpretable defenses mechanisms based on mechanistic interpretability.

2025-06-29

ICML.cc/2025/Workshop/R2-FM (poster)

Less is More: Undertraining Experts Improves Model Upcycling

Stefan Horoi

Gintare Karolina Dziugaite

Guy Wolf

Modern deep learning is increasingly characterized by the use of open-weight foundation models that can be fine-tuned on specialized dataset… (see more)s. This has led to a proliferation of expert models and adapters, often shared via platforms like HuggingFace and AdapterHub. To leverage these resources, numerous model upcycling methods have emerged, enabling the reuse of fine-tuned models in multi-task systems. A natural pipeline has thus formed to harness the benefits of transfer learning and amortize sunk training costs: models are pre-trained on general data, fine-tuned on specific tasks, and then upcycled into more general-purpose systems. A prevailing assumption is that improvements at one stage of this pipeline propagate downstream, leading to gains at subsequent steps. In this work, we challenge that assumption by examining how expert fine-tuning affects model upcycling. We show that long fine-tuning of experts that optimizes for their individual performance leads to degraded merging performance, both for fully fine-tuned and LoRA-adapted models, and to worse downstream results when LoRA adapters are upcycled into MoE layers. We trace this degradation to the memorization of a small set of difficult examples that dominate late fine-tuning steps and are subsequently forgotten during merging. Finally, we demonstrate that a task-dependent aggressive early stopping strategy can significantly improve upcycling performance.

2025-06-16

arXiv (Cornell University) (preprint)