Fabrizio Sandri

I’m a Machine Learning Engineer with a strong passion for open-source, currently working at SpazioDati. Previously I was a Machine Learning Engineer at Airbus Defence and Space, working on cutting-edge aerospace projects.

I earned both my Bachelor’s and Master’s degrees (cum laude) from the University of Trento in 2022 and 2024, respectively.

With a particular keenness for open source, I took part in the Google Summer of Code 2022 as a software developer for the R project for statistical computing. I had the privilege of serving as a mentor at the University of Tennessee, as part of the Google Award for Inclusion. Over the years, I have actively contributed to diverse open source projects, including contributions to NASA FPrime, DeepState, Data & Analytics Framework, fueling my dedication to pushing the boundaries of innovation through collaborative endeavors.

By day I build AI systems; by night I photograph galaxies. Astrophotography is where my curiosity goes when the sky is clear and the code can wait. → Astrophotography gallery

News

Aug 17, 2025	2SSP: A Two-Stage Framework for Structured Pruning of LLMs has been accepted at TMLR 2025.
Mar 05, 2025	2SSP: A Two-Stage Framework for Structured Pruning of LLMs has been accepted at SLLM @ ICLR2025
Feb 01, 2024	Joined Airbus Defence and Space.
May 20, 2022	Selected for the Google Summer of Code 2022 – Read more on my blog

Selected publications

TMLR2025

2SSP: A Two-Stage Framework for Structured Pruning of LLMs

Fabrizio Sandri, Elia Cunegatti, and Giovanni Iacca

Transactions on Machine Learning Research, Aug 2025

Abs PDF Code

We propose a novel Two-Stage framework for Structured Pruning (2SSP) for pruning Large Language Models (LLMs), which combines two different strategies of pruning, namely Width and Depth Pruning. The first stage (Width Pruning) removes entire neurons, hence their corresponding rows and columns, aiming to preserve the connectivity among the pruned structures in the intermediate state of the Feed-Forward Networks in each Transformer block. The second stage (Depth Pruning), instead, removes entire Attention submodules. We also propose a novel mechanism to balance the sparsity rate of the two stages w.r.t. to the desired global sparsity. We test 2SSP on four LLM families and three sparsity rates (25%, 37.5%, and 50%), measuring the resulting perplexity over three language modeling datasets as well as the performance over six downstream tasks. Our method consistently outperforms five state-of-the-art competitors over three language modeling and six downstream tasks, with an up to two-order-of-magnitude gain in terms of pruning time. The code is available at available at https://github.com/FabrizioSandri/2SSP.
SLLM @ ICLR2025

2SSP: A Two-Stage Framework for Structured Pruning of LLMs

Fabrizio Sandri, Elia Cunegatti, and Giovanni Iacca

In Accepted at Sparsity in LLMs (SLLM) Workshop at ICLR 2025, Mar 2025

Abs PDF Code

We propose a novel Two-Stage framework for Structured Pruning (2SSP) for pruning Large Language Models (LLMs), which combines two different strategies of pruning, namely Width and Depth Pruning. The first stage (Width Pruning) removes entire neurons, hence their corresponding rows and columns, aiming to preserve the connectivity among the pruned structures in the intermediate state of the Feed-Forward Networks in each Transformer block. The second stage (Depth Pruning), instead, removes entire Attention submodules. We also propose a novel mechanism to balance the sparsity rate of the two stages w.r.t. to the desired global sparsity. We test 2SSP on four LLM families and three sparsity rates (25%, 37.5%, and 50%), measuring the resulting perplexity over three language modeling datasets as well as the performance over six downstream tasks. Our method consistently outperforms five state-of-the-art competitors over three language modeling and six downstream tasks, with an up to two-order-of-magnitude gain in terms of pruning time. The code is available at available at https://github.com/FabrizioSandri/2SSP.
Preprint

2SSP: A Two-Stage Framework for Structured Pruning of LLMs

Fabrizio Sandri, Elia Cunegatti, and Giovanni Iacca

arXiv preprint arXiv:2501.17771, Jan 2025

Abs arXiv PDF Code

We propose a novel Two-Stage framework for Structured Pruning (2SSP) for pruning Large Language Models (LLMs), which combines two different strategies of pruning, namely Width and Depth Pruning. The first stage (Width Pruning) removes entire neurons, hence their corresponding rows and columns, aiming to preserve the connectivity among the pruned structures in the intermediate state of the Feed-Forward Networks in each Transformer block. The second stage (Depth Pruning), instead, removes entire Attention submodules. We also propose a novel mechanism to balance the sparsity rate of the two stages w.r.t. to the desired global sparsity. We test 2SSP on four LLM families and three sparsity rates (25%, 37.5%, and 50%), measuring the resulting perplexity over three language modeling datasets as well as the performance over six downstream tasks. Our method consistently outperforms five state-of-the-art competitors over three language modeling and six downstream tasks, with an up to two-order-of-magnitude gain in terms of pruning time. The code is available at available at https://github.com/FabrizioSandri/2SSP.