 |
ML Newsnew | past | best | rss | submit | about |
| VaultGemma: A Differentially Private LLM (arxiv.org) | |
| 1 point by PaulHoule 245 days ago | hide | past | pdf | discuss on HN | |
|
In plain words: VaultGemma 1B is a 1-billion-parameter model trained with differential privacy, adding noise so no single person's data can be picked out. Trained on the same data mix as Gemma 2, it shows a 1B model can be trained this way and is openly released. Abstract · VaultGemma: A Differentially Private Gemma ModelWe introduce VaultGemma 1B, a 1 billion parameter model within the Gemma family, fully trained with differential privacy. Pretrained on the identical data mixture used for the Gemma 2 series, VaultGemma 1B represents a significant step forward in privacy-preserving large language models. We openly release this model to the community | |
|
add comment on HN
Also discussed: Jan 2026 (3 points, 0 comments)
|