msakarvadia/memorization
Localizing Memorized Sequences in Language Models
This project helps AI developers and researchers prevent large language models (LLMs) from inadvertently revealing private or sensitive training data during inference. It takes a pre-trained language model and identifies and removes memorized sequences without significantly impacting the model's overall performance. AI engineers concerned with data privacy and security in their LLM deployments are the primary users.
Use this if you need to precisely remove specific memorized information from a language model's weights to enhance data privacy and mitigate risks of sensitive data exposure.
Not ideal if you are looking for general model fine-tuning or regularization methods that are not specifically focused on pinpointing and removing memorized data.
Stars
20
Forks
2
Language
Jupyter Notebook
License
MIT
Category
Last pushed
Oct 15, 2025
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/transformers/msakarvadia/memorization"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Higher-rated alternatives
MadryLab/context-cite
Attribute (or cite) statements generated by LLMs back to in-context information.
microsoft/augmented-interpretable-models
Interpretable and efficient predictors using pre-trained language models. Scikit-learn compatible.
Trustworthy-ML-Lab/CB-LLMs
[ICLR 25] A novel framework for building intrinsically interpretable LLMs with...
poloclub/LLM-Attributor
LLM Attributor: Attribute LLM's Generated Text to Training Data
THUDM/LongCite
LongCite: Enabling LLMs to Generate Fine-grained Citations in Long-context QA