Luca CanaleinSarus BlogDiscovering new knowledge with Differential PrivacyIn a previous post, we showed that a LLM can be fine tuned to learn new knowledge from unstructured natural language. We trained a Q-LoRA…3d ago3d ago
Luca CanaleinSarus BlogFine-tuning Mistral 7-B with QLora for new knowledge learningThis post is a small technical report to evaluate the capacity of a LLM to learn some new knowledge, it has never faced before. To do so…Jul 11Jul 11
Luca CanaleinSarus BlogAn open-source dataset to test LLM private knowledge understandingSarus builds privacy-preserving AI solutions. Many of our AI use cases involve:Jul 5Jul 5
Luca CanaleinSarus BlogLLMs have become excellent at memorizing: it’s a problem for privacyRecent advances in generative AI have brought a lot of enthusiasm in the community as new open source models such as Llama2 can achieve…Oct 3, 2023Oct 3, 2023
Luca CanaleinSarus BlogHow advances in Generative AI change the game for privacy protectionLanguage models have emerged as powerful tools but they come with strong privacy concerns. Up until today, it was impractical to train them…Sep 15, 2023Sep 15, 2023