7 papers
Shieldstral
Antonia Calvi, Avinash Sooriyarachchi, Giada Pistilli +274
We introduce Shieldstral, a 3B-parameter policy-adaptive multimodal safety classifier that matches or outperforms models nearly 7 its size on text safety benchmarks and set…
Efficient LLM Moderation with Multi-Layer Latent Prototypes
Maciej ChrabÄ szcz, Filip Szatkowski, Bartosz Wójcik +3
Although modern LLMs are aligned with human values during post-training, robust moderation remains essential to prevent harmful outputs at deployment time. Existing approaches suff…
Beyond Classification: Dynamic Adapter Routing for Continual Multimodal Retrieval
Alicja Dobrzeniecka, Filip Szatkowski, Sebastian Cygert +2
While retrieval is a core function of vision-language models, continually updating these models for retrieval tasks remains critically underexplored. Existing work often approaches…
Rethinking Calibration for Early-Exit Neural Networks
Piotr Kubaty, Filip Szatkowski, Grzegorz ChoczyÅski +2
Early-exit neural networks (EENNs) accelerate inference by allowing intermediate classifiers to stop computation once predictions are confident enough. Most methods rely on confide…
Universal Properties of Activation Sparsity in Modern Large Language Models
Filip Szatkowski, Patryk BÄdkowski, Alessio Devoto +5
Activation sparsity is an intriguing property of deep neural networks that has been extensively studied in ReLU-based models, due to its advantages for efficiency, robustness, and…
ExpertSim: Fast Particle Detector Simulation Using Mixture-of-Generative-Experts
Patryk BÄdkowski, Jan DubiÅski, Filip Szatkowski +3
Simulating detector responses is a crucial part of understanding the inner workings of particle collisions in the Large Hadron Collider at CERN. Such simulations are currently perf…