eeada7810a
Correct post-bias degeneration attribution
...
Cannot isolate instruct tuning vs architecture (sliding window,
KV sharing, logit softcapping) with current test matrix.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com >
2026-04-06 19:44:42 +07:00
0448e96bc9
Remove cringe closing line
...
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com >
2026-04-06 19:35:04 +07:00
941af3a585
Add LaTeX + PDF, fix section cross-reference
...
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com >
2026-04-06 19:31:32 +07:00
176164815b
Solving the Clive Wearing Problem: One-Shot Episodic Memory for Frozen Transformers
...
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
2026-04-05 02:16:29 +07:00
