Visualizing GPT-2 Vocabulary in a Hyperbolic Ball
Limp-Contest-7309 · reddit · 2026-07-19
This is a visualization project that places the 32,070 token embeddings of GPT-2-small into hyperbolic space (Poincaré ball). It requires no training, purely using the raw token embeddings for geometric layout.
The author explains that the similarity structure of a vocabulary resembles a forest rather than a flat graph. Hyperbolic space is naturally better at accommodating such tree-like structures because the space expands exponentially with distance. The web app is mobile-friendly, allowing users to drag, rotate, zoom, search for tokens, and explore along proximity lines.
The post also shares some interesting observations:
- The dense inner layers are mostly isolated tokens that lack "mutual nearest neighbors."
- You can search for "dog" and follow the connections to find "cats," or walk through the days of the week starting from "Monday."
- In this structure, "Trump" is solely connected to "Ivanka."
The entire project consists of a single HTML file and runs entirely on the client side.
Related event: Interactive Tool Visualizes GPT-2 Token Embeddings(4 posts)→
More from Research
- Diamandis: AI Cancer Navigator and $100 Genome Sequencing Close the Care Gap — PeterDiamandis · 2026-10-27
- EMNLP Oral: training-free Nexus Sampling cuts KV-cache memory 10x; OpenResearcher tops BrowseComp-Plus at 54.8% — yuz9yuz · 2026-10-27
- ICML 2026 position paper: LLM 'thinking traces' are a harmful anthropomorphism — rao2z · 2026-10-27
- Forking gap follows a power law in context frequency; masking rare n-grams shrinks it — ZimingLiu11 · 2026-10-27
- How forking happens: n-gram tables split logits into non-interacting subspaces — ZimingLiu11 · 2026-10-27
- The best LoRA explainer video yet: rank constraints meet intrinsic dimension — sophiamyang · 2026-10-27