Vikram Seraph 🪽
@vikramsaraph.com
1.2K followers
570 following
1.4K posts
Software engineer, AI/ML researcher, and mathematician at Johns Hopkins APL.
Former New Englander, current Marylander.
Brown CS PhD and Notre Dame math alum.
Nerd of sorts (computers, math, language, puzzles, games, books, music).
Opinions are my own.
Posts
Media
Videos
Starter Packs
Reposted by Vikram Seraph 🪽
New preprint from one of my colleagues in the Math department studying the dimension and Ricci curvature of the token embedding space in several LLMs, and connecting it in part to differences in model behavior (e.g. GPT2 and Mistral7b embed numerical tokens in very different ways).
The structure of the token space for large language models
Large language models encode the correlational structure present in natural language by fitting segments of utterances (tokens) into a high dimensional ambient latent space upon which the models then ...
www.arxiv.org
I donated $25 to the Python Software Foundation.
Did I do that to post about it social media? Yes. Did I also do it because I think that the Python programming language is awesome, and PSF and PyCon are awesome, and they need more funding? Also yes.
Did I do that to post about it social media? Yes. Did I also do it because I think that the Python programming language is awesome, and PSF and PyCon are awesome, and they need more funding? Also yes.
