hmohebbi/ValueZeroing
The official repo for the EACL 2023 paper "Quantifying Context Mixing in Transformers"
When analyzing how Large Language Models (LLMs) understand text, it's crucial to know how different words influence each other. This tool helps you quantify exactly how much information from one word is mixed into another word's representation within a Transformer model. It takes a trained Transformer model and text input, then outputs a score indicating the degree of 'context mixing' for each word. AI researchers and NLP practitioners who develop or fine-tune LLMs will find this useful for model interpretation and debugging.
No commits in the last 6 months.
Use this if you need to precisely measure how contextual information flows and mixes between tokens inside a Transformer model to better understand its decision-making.
Not ideal if you are looking for a tool to train new models or simply apply pre-trained models without needing deep insights into their internal workings.
Stars
12
Forks
3
Language
Python
License
—
Category
Last pushed
Mar 31, 2023
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/transformers/hmohebbi/ValueZeroing"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Higher-rated alternatives
lucidrains/x-transformers
A concise but complete full-attention transformer with a set of promising experimental features...
kanishkamisra/minicons
Utility for behavioral and representational analyses of Language Models
lucidrains/simple-hierarchical-transformer
Experiments around a simple idea for inducing multiple hierarchical predictive model within a GPT
lucidrains/dreamer4
Implementation of Danijar's latest iteration for his Dreamer line of work
Nicolepcx/Transformers-in-Action
This is the corresponding code for the book Transformers in Action