diff options
Diffstat (limited to 'README.md')
| -rw-r--r-- | README.md | 29 |
1 files changed, 29 insertions, 0 deletions
diff --git a/README.md b/README.md new file mode 100644 index 0000000..e244b24 --- /dev/null +++ b/README.md @@ -0,0 +1,29 @@ +# J-space on nanoGPT + +Replicating Anthropic's Jacobian Lens ("J-space") technique on a small character-level transformer (nanoGPT). + +## Hypothesis + +If J-space (the subspace of representations readily available for verbal report) is an **architectural/structural property of transformers** rather than an emergent feature of advanced models, it should appear at all scales — including 10M-parameter char-level models. + +## Background + +Anthropic's 2026 paper "Verbalizable Representations Form a Global Workspace in Language Models" introduces the Jacobian Lens (J-lens), which computes the average linearized effect of activations on future token probabilities, averaged over many contexts. This reveals a privileged "J-space" of representations that the model can report on, modulate, and use for reasoning. + +Full paper: https://transformer-circuits.pub/2026/workspace/index.html + +## Experiments + +1. **J-space visualization** — Compute J-lens vectors for all vocabulary tokens at each layer. Visualize which characters/concepts enter "verbalizable space" and when. +2. **Ablation test** — Remove J-space components vs random directions vs full activations. Measure prediction quality impact. +3. **Training dynamics** — Save checkpoints during training, compute J-space at each, track when it crystallizes. +4. **Capacity measurement** — Count active J-lens tokens per position. + +## Setup + +Runs on meru's Quadro K2200 (4GB VRAM) via Docker with GPU passthrough. + +## References + +- [Verbalizable Representations Form a Global Workspace in Language Models](https://transformer-circuits.pub/2026/workspace/index.html) — Anthropic, 2026 +- [nanoGPT](https://github.com/karpathy/nanoGPT) — Andrej Karpathy |