Hugging Face's logo

Tom Goldstein's Lab at University of Maryland, College Park
university
http://www.cs.umd.edu/~tomg/
tomgoldsteincs
Activity Feed

AI & ML interests

AI security & privacy, algorithmic bias, foundations of ML

Recent Activity

ehartz  published a dataset 2 days ago
tomg-group-umd/IDE_code_retrival
smcleish  authored a paper about 2 months ago
Teaching Pretrained Language Models to Think Deeper with Retrofitted Recurrence
leonli66  authored a paper about 2 months ago
Teaching Pretrained Language Models to Think Deeper with Retrofitted Recurrence
View all activity

Papers

Teaching Pretrained Language Models to Think Deeper with Retrofitted Recurrence

Gemstones: A Model Suite for Multi-Faceted Scaling Laws

View all Papers

tomg-group-umd 's collections 13

Refusal Token Models
This collection contains models described in the refusal token paper published in COLM 2025.
LoRI Adapters
LoRI adapters for natural language understanding, code generation, mathematical reasoning, and safety alignment, based on LLaMA-3-8B and Mistral-7B.
Recurrent Models
These are checkpoints for recurrent LLMs developed to scale test-time compute by recurring in latent space.
Gemstone Models
Our 22 open source Gemstone models for scaling laws range from 50M to 2B parameters, spanning 11 widths from 256 to 3072 and 18 depths from 3 to 80.
Goldfish Loss: Mitigating Memorization in LLMs
This collection contains artifacts from our paper titled: "Be like a Goldfish, Don't Memorize! Mitigating Memorization in Generative LLMs."
Refusal Token Models
This collection contains models described in the refusal token paper published in COLM 2025.
LoRI Adapters
LoRI adapters for natural language understanding, code generation, mathematical reasoning, and safety alignment, based on LLaMA-3-8B and Mistral-7B.
Gemstone Models
Our 22 open source Gemstone models for scaling laws range from 50M to 2B parameters, spanning 11 widths from 256 to 3072 and 18 depths from 3 to 80.
Recurrent Models
These are checkpoints for recurrent LLMs developed to scale test-time compute by recurring in latent space.
Goldfish Loss: Mitigating Memorization in LLMs
This collection contains artifacts from our paper titled: "Be like a Goldfish, Don't Memorize! Mitigating Memorization in Generative LLMs."