Hugging Face's logo

Allegro Lab @ USC
Team
university
https://allegro-lab.github.io/
Activity Feed

AI & ML interests

None defined yet.

Recent Activity

allegro-lab  published a dataset 8 days ago
allegrolab/biographies_yago_rephrased
allegro-lab  updated a dataset about 2 months ago
allegrolab/dclm-baseline-500b_toks
allegro-lab  updated a model about 2 months ago
allegrolab/hubble-8b-500b_toks-standard-neox
View all activity

allegrolab 's collections 6

Hubble - Interference
Three models trained on only copyright, privacy, or test set contamination perturbations to measure interference between domains.
Hubble - Paraphrase
Two models trained on paraphrased YAGO biographies and MMLU test sets to study how paraphrased information is memorized.
Hubble - Core
Eight models that vary in size, data condition, and corpus scale to establish dilution effects in memorization.
Hubble - Timing
Six models trained with perturbations inserted at different stages of training to study how exposure timing affects memorization.
Hubble - Architecture
Two models trained with shallower and deeper transformer architectures to assess how model depth affects memorization.
Hubble - Core
Eight models that vary in size, data condition, and corpus scale to establish dilution effects in memorization.
Hubble - Interference
Three models trained on only copyright, privacy, or test set contamination perturbations to measure interference between domains.
Hubble - Timing
Six models trained with perturbations inserted at different stages of training to study how exposure timing affects memorization.
Hubble - Paraphrase
Two models trained on paraphrased YAGO biographies and MMLU test sets to study how paraphrased information is memorized.
Hubble - Architecture
Two models trained with shallower and deeper transformer architectures to assess how model depth affects memorization.