Solving the Clive Wearing Problem: One-Shot Episodic Memory for Frozen Transformers

Browse Source

Tommi Niemi / Rotko Networks

Hidden-state episodic memory for frozen transformers. No gradients.
Teach via one forward pass, recall via cosine similarity + logit injection.
200-line Python reproduction included.

pip install transformers torch numpy && python python/epimem.py

...

This commit is contained in:

Tommi Niemi

2026-04-05 01:20:17 +07:00

commit 1de04890a0

14 changed files with 785993 additions and 0 deletions

Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL

Download Patch File Download Diff File Expand all files Collapse all files

24530

results/isis_e2e_test.json Normal file

File diff suppressed because it is too large Load Diff