zjunlp/DynamicKnowledgeCircuits
[ACL 2025] How Do LLMs Acquire New Knowledge? A Knowledge Circuits Perspective on Continual Pre-Training
This research project investigates how Large Language Models (LLMs) learn and store new information during continual pre-training. It takes raw knowledge entities and training data as input, then identifies and evaluates the specific neural 'circuits' that handle this knowledge. The project helps AI researchers and machine learning engineers better understand and improve how LLMs acquire and retain new facts.
No commits in the last 6 months.
Use this if you are a researcher or engineer looking to delve into the internal mechanisms of LLMs and optimize their knowledge acquisition capabilities during ongoing training.
Not ideal if you are looking for an off-the-shelf tool to directly apply LLMs to real-world tasks or enhance their performance without understanding the underlying neural processes.
Stars
47
Forks
1
Language
Jupyter Notebook
License
MIT
Category
Last pushed
Jul 18, 2025
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/transformers/zjunlp/DynamicKnowledgeCircuits"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Higher-rated alternatives
ExtensityAI/symbolicai
A neurosymbolic perspective on LLMs
TIGER-AI-Lab/MMLU-Pro
The code and data for "MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding...
deep-symbolic-mathematics/LLM-SR
[ICLR 2025 Oral] This is the official repo for the paper "LLM-SR" on Scientific Equation...
microsoft/interwhen
A framework for verifiable reasoning with language models.
zhudotexe/fanoutqa
Companion code for FanOutQA: Multi-Hop, Multi-Document Question Answering for Large Language...