apartresearch / interpreting-reward-models Goto Github PK
View Code? Open in Web Editor NEW✱ Interpreting implicit reward models learnt in RLHF using sparse autoencoders.
License: MIT License
✱ Interpreting implicit reward models learnt in RLHF using sparse autoencoders.
License: MIT License
We want to create a contrastive pairs dataset derived from helpful / harmless, where we flip a token to get "helpful" vs "non helpful" contrastive pairs.
For now, do this just for gpt-125m-neo to make this fast.
Make the pipeline configurable while doing this, so we can easily run this for a range of other models quicker.
The current IMDB dataset suffers from very few examples of the actual vader lexicon. As such, let's create two new datasets that have high overlap with the vader lexicon.
The columns of the dataset will be text
, lexicon_tokens
, token_rewards_dict
and poisoned
which is a (usually empty) list of tokens. There were will be 30 of these.
The vader lexicon tokens will be ordered by their frequency in english, and the top 4000 will be picked, with 5 occurrences each.
We want to create a contrastive pairs dataset similar to that for IMDB, in the style of https://huggingface.co/datasets/unalignment/toxic-dpo-v0.2.
We need a neutral, positive and negative baseline.
These have much more mature autoencoders: https://github.com/jbloomAus/SAELens?tab=readme-ov-file
Much more time efficient to just use these rather than reimplement from scratch.
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.