tommi/cri
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Update all URLs and paths from epimem to cri

Browse Source 
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Tommi Niemi
2026-04-06 19:24:26 +07:00
parent 7838bc4ed1
commit 7271339006
4 changed files with 26 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
paper.md
View File

@@ -4,7 +4,7 @@
## Abstract
We enable frozen transformers to acquire new stimulus-response behaviors without gradient descent. When a target association is presented, the model's own hidden-state activation pattern is captured as a trigger, and per-token logit biases are stored as the conditioned response. At inference time, a matching activation pattern fires the reflex, injecting the learned token biases directly into the output logits. The model does not "know" the new fact — it is nudged toward specific token sequences when the right internal pattern activates. We test across four backbones (Qwen 2.5 0.5B, Gemma 4 E2B-it, E4B-it, and E4B base) and five precision levels (float32 through int4). Conditioned tokens are produced correctly in all cases, but post-bias coherence and trigger discrimination vary significantly across architectures: smaller base models outperform larger instruct-tuned models on both metrics. No weights are modified. No gradients are computed. Reflexes persist to disk across sessions. Code and reproduction: [git.rotko.net/tommi/epimem](https://git.rotko.net/tommi/epimem).
We enable frozen transformers to acquire new stimulus-response behaviors without gradient descent. When a target association is presented, the model's own hidden-state activation pattern is captured as a trigger, and per-token logit biases are stored as the conditioned response. At inference time, a matching activation pattern fires the reflex, injecting the learned token biases directly into the output logits. The model does not "know" the new fact — it is nudged toward specific token sequences when the right internal pattern activates. We test across four backbones (Qwen 2.5 0.5B, Gemma 4 E2B-it, E4B-it, and E4B base) and five precision levels (float32 through int4). Conditioned tokens are produced correctly in all cases, but post-bias coherence and trigger discrimination vary significantly across architectures: smaller base models outperform larger instruct-tuned models on both metrics. No weights are modified. No gradients are computed. Reflexes persist to disk across sessions. Code and reproduction: [git.rotko.net/tommi/cri](https://git.rotko.net/tommi/cri).
## 1. The Clive Wearing Problem — and What It Really Is
@@ -252,8 +252,8 @@ The reflex bank is saved to JSON (77KB for 3 reflexes with 896-dim triggers on Q
### 3.7 Reproduction
```bash
git clone https://git.rotko.net/tommi/epimem
cd epimem
git clone https://git.rotko.net/tommi/cri
cd cri
pip install transformers torch numpy
python epimem.py --model Qwen/Qwen2.5-0.5B
```
@@ -354,6 +354,6 @@ The implementation reproduces the full result in approximately 200 lines of Pyth
title={Conditioned Reflex Injection: Stimulus-Response Learning for Frozen Transformers},
author={Niemi, Tommi},
year={2026},
url={https://git.rotko.net/tommi/epimem}
url={https://git.rotko.net/tommi/cri}
}
```