Home

Inspiring the development of artificial intelligence for the benefit of all 

A professor talks to his students in a café/lounge.

Located in the heart of Quebec’s AI ecosystem, Mila is a community of more than 1,200 researchers specializing in machine learning and dedicated to scientific excellence and innovation.

About

Featured

Faculty 

Founded in 1993 by Professor Yoshua Bengio, Mila today brings together over 130 professors affiliated with Université de Montréal, McGill University, Polytechnique Montréal and HEC Montréal. Mila also welcomes professors from Université Laval, Université de Sherbrooke and École de technologie supérieure (ÉTS). 

Browse the online directory

Photo of Yoshua Bengio

Latest Publications

Group Membership Bias
Ali Vardasbi
Maarten de Rijke
Mostafa Dehghani
Simple and Scalable Strategies to Continually Pre-train Large Language Models
Adam Ibrahim
Benjamin Thérien
Kshitij Gupta
Mats Leon Richter
Quentin Gregory Anthony
Timothee LESORT
LORD: Low Rank Decomposition Of Monolingual Code LLMs For One-Shot Compression
Ayush Kaushal
Tejas Vaidhya
Low Rank Decomposition of matrix - splitting a large matrix into a product of two smaller matrix offers a means for compression that reduces… (see more) the parameters of a model without sparsification, and hence delivering more speedup on modern hardware. Moreover, unlike quantization, the compressed linear layers remain fully differentiable and all the parameters trainable, while being able to leverage the existing highly efficient kernels over floating point matrices. We study the potential to compress Large Language Models (LLMs) for monolingual Code generation via Low Rank Decomposition (LoRD) and observe that ranks for the linear layers in these models can be reduced by upto 39.58% with less than 1% increase in perplexity. We then use Low Rank Decomposition (LoRD) to compress StarCoder 16B to 13.2B parameter with no drop and to 12.3B with minimal drop in HumanEval Pass@1 score, in less than 10 minutes on a single A100. The compressed models speeds up inference by up to 22.35% with just a single line of change in code over huggingface's implementation with pytorch backend. Low Rank Decomposition (LoRD) models remain compatible with state of the art near-lossless quantization method such as SpQR, which allows leveraging further compression gains of quantization. Lastly, QLoRA over Low Rank Decomposition (LoRD) model further reduces memory requirements by as much as 21.2% over vanilla QLoRA while offering similar gains from parameter efficient fine tuning. Our work shows Low Rank Decomposition (LoRD) as a promising new paradigm for LLM compression.
A Bayesian Non-Stationary Heteroskedastic Time Series Model for Multivariate Critical Care Data
Zayd Omar
David A. Stephens
Alexandra M. Schmidt

AI for Humanity

Socially responsible and beneficial development of AI is a fundamental component of Mila’s mission. As a leader in the field, we wish to contribute to social dialogue and the development of applications that will benefit society.

Learn more

A person looks up at a starry sky.