3

Language Models are Injective and Hence Invertible
Transformer components such as non-linear activations and normalization are inherently non-injective, suggesting that different inputs …
Giorgios Nikolaou, Tommaso Mencattini, Donato Crisostomi, Santilli, Yannis Panagakis, Emanuele Rodolà
Cite arXiv Thread
Language Models are Injective and Hence Invertible
Two-Scale Latent Dynamics for Recurrent-Depth Transformers
Recurrent-depth transformers scale test-time compute by iterating latent computations before emitting tokens. We study the geometry of …
Francesco Pappone, Donato Crisostomi, Emanuele Rodolà
arXiv
Two-Scale Latent Dynamics for Recurrent-Depth Transformers
On Task Vectors and Gradients
Task arithmetic has emerged as a simple yet powerful technique for model merging, enabling the combination of multiple finetuned models …
Luca Zhou, Daniele Solombrino, Donato Crisostomi, Maria Sofia Bucarelli, Giuseppe Alessio D'Inverno, Fabrizio Silvestri, Emanuele Rodolà
arXiv
On Task Vectors and Gradients
Implicit Inversion turns CLIP into a Decoder
CLIP is a discriminative model trained to align images and text in a shared embedding space. Due to its multimodal structure, it serves …
Antonio D'Orazio, Maria Rosaria Briglia, crisostomi, Dario Loi, rodola, Iacopo Masi
Cite arXiv GitHub
Implicit Inversion turns CLIP into a Decoder
Mergenetic: a Simple Evolutionary Model Merging Library
Model merging allows combining the capabilities of existing models into a new one - post hoc, without additional training. This has …
Adrian R. Minut, Tommaso Mencattini, Marco Santilli, Donato Crisostomi, Emanuele Rodolà
Cite arXiv GitHub
Mergenetic: a Simple Evolutionary Model Merging Library
Efficient Generation of Multimodal Fluid Simulation Data
In this work, we introduce an efficient generation procedure to produce synthetic multi-modal datasets of fluid simulations. The …
Daniele Baieri, Donato Crisostomi, Donato Crisostomi, Stefano Esposito, Filippo Maggioli, Emanuele Rodolà
Cite arXiv
Efficient Generation of Multimodal Fluid Simulation Data
STAGE: Stemmed Accompaniment Generation through Prefix-Based Conditioning
Recent advances in generative models have made it possible to create high-quality, coherent music, with some systems delivering …
Giorgio Strano, Chiara Ballanti, Donato Crisostomi, Michele Mancusi, Luca Cosmo, Emanuele Rodolà
Cite arXiv GitHub
STAGE: Stemmed Accompaniment Generation through Prefix-Based Conditioning
Activation Patching for Interpretable Steering in Music Generation
Understanding how large audio models represent music, and using that understanding to steer generation, is both challenging and …
Simone Facchiano, Giorgio Strano, Donato Crisostomi, Irene Tallini, Tommaso Mencattini, Fabio Galasso, Emanuele Rodolà
Cite arXiv
Activation Patching for Interpretable Steering in Music Generation
LoopGen: Training-Free Loopable Music Generation
Loops–short audio segments designed for seamless repetition–are central to many music genres, particularly those rooted in …
Davide Marincione, Giorgio Strano, Donato Crisostomi, Roberto Ribuoli, Emanuele Rodolà
Cite arXiv
LoopGen: Training-Free Loopable Music Generation
MASS: MoErging through Adaptive Subspace Selection
Model merging has recently emerged as a lightweight alternative to ensembling, combining multiple fine-tuned models into a single set …
Donato Crisostomi, Alessandro Zirilli, Antonio Andrea Gargiulo, Maria Sofia Bucarelli, Simone Scardapane, Fabrizio Silvestri, Iacopo Masi, Emanuele Rodolà
Cite arXiv
MASS: MoErging through Adaptive Subspace Selection