About

A small research lab with a specific focus.

The name

Oblìo is Italian for oblivion. The full term — oblìo catastrofico — translates to catastrophic forgetting, the phenomenon where neural networks lose previously learned knowledge when trained on new tasks. It's the core problem we study.

Team

Davide

Co-founder

Alessandro

Co-founder

What we do

We study why fine-tuned language models lose previously learned capabilities, and we're building tools to prevent it. Our work spans continual learning strategies, LoRA/QLoRA adapter workflows, and open-source tooling — with a focus on reproducible experiments and open results.

Infrastructure

We run experiments on a mix of local hardware and cloud compute.

Machine Specs Use
MacBook Pro M4 Pro 24 GB unified Dev, prototyping, small inference
MacBook Pro M5 Pro 48 GB unified Medium inference, local experiments
Workstation NVIDIA RTX 4000, 24 GB LoRA/QLoRA fine-tuning, training
Cloud On-demand GPUs/TPUs Large-scale training

Contact

For collaboration inquiries, reach us at [email protected].

See what we're working on: research · models