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

Add LaTeX + PDF, fix section cross-reference

Browse Source 
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Tommi Niemi
2026-04-06 19:31:32 +07:00
parent 258a03388c
commit 941af3a585
4 changed files with 234 additions and 306 deletions
Download Patch File Download Diff File Expand all files Collapse all files

54
paper.aux
View File

@@ -2,40 +2,40 @@
\providecommand\hyper@newdestlabel[2]{}
\providecommand\HyField@AuxAddToFields[1]{}
\providecommand\HyField@AuxAddToCoFields[2]{}
\@writefile{toc}{\contentsline {section}{\numberline {1}The Clive Wearing Problem}{1}{section.1}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {1}Conditioning, Not Memory}{1}{section.1}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {2}Method}{1}{section.2}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {2.1}Architecture}{1}{subsection.2.1}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {2.2}Teaching (One Forward Pass)}{2}{subsection.2.2}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {2.3}Recall (Similarity Search + Injection)}{2}{subsection.2.3}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {2.4}Persistence}{2}{subsection.2.4}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {2.5}Why Hidden States, Not Text}{2}{subsection.2.5}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {2.2}Conditioning}{2}{subsection.2.2}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {2.3}Triggering}{2}{subsection.2.3}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {2.4}Why Hidden States, Not Text}{2}{subsection.2.4}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {3}Experiments}{2}{section.3}\protected@file@percent }
\newlabel{sec:results}{{3}{2}{Experiments}{section.3}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {3.1}Setup}{2}{subsection.3.1}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {3.2}One-Shot Conditioning}{3}{subsection.3.2}\protected@file@percent }
\@writefile{lot}{\contentsline {table}{\numberline {1}{\ignorespaces Post-bias degeneration is caused by instruct tuning, not model size.}}{3}{table.1}\protected@file@percent }
\newlabel{tab:postbias}{{1}{3}{Post-bias degeneration is caused by instruct tuning, not model size}{table.1}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {3.3}Stimulus Generalization and Misfire}{3}{subsection.3.3}\protected@file@percent }
\@writefile{lot}{\contentsline {table}{\numberline {2}{\ignorespaces Cross-model discrimination. Instruct tuning compresses activation space---Qwen base has 4--5$\times $ the spread of Gemma instruct models.}}{3}{table.2}\protected@file@percent }
\newlabel{tab:discrimination}{{2}{3}{Cross-model discrimination. Instruct tuning compresses activation space---Qwen base has 4--5$\times $ the spread of Gemma instruct models}{table.2}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {3.4}Quantization Tolerance}{3}{subsection.3.4}\protected@file@percent }
\@writefile{lot}{\contentsline {table}{\numberline {3}{\ignorespaces Actual quantized inference (bitsandbytes, Qwen). Same-precision self-match is always 1.000. Cross-precision f32$\to $int4 drops to 0.944.}}{3}{table.3}\protected@file@percent }
\newlabel{tab:quant}{{3}{3}{Actual quantized inference (bitsandbytes, Qwen). Same-precision self-match is always 1.000. Cross-precision f32$\to $int4 drops to 0.944}{table.3}{}}
\citation{jang2024camelot}
\citation{fountas2024emllm}
\citation{das2024larimar}
\@writefile{toc}{\contentsline {section}{\numberline {3}Experiments}{3}{section.3}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {3.1}Setup}{3}{subsection.3.1}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {3.2}One-Shot Fact Learning}{3}{subsection.3.2}\protected@file@percent }
\@writefile{lot}{\contentsline {table}{\numberline {1}{\ignorespaces One-shot fact recall. All three novel facts recalled correctly with cosine similarity 1.000. The backbone generates fluent continuations beyond the taught answer.}}{3}{table.1}\protected@file@percent }
\newlabel{tab:results}{{1}{3}{One-shot fact recall. All three novel facts recalled correctly with cosine similarity 1.000. The backbone generates fluent continuations beyond the taught answer}{table.1}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {3.3}Persistence}{3}{subsection.3.3}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {3.4}Reproduction}{3}{subsection.3.4}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {4}Related Work}{3}{section.4}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {4.1}Training-Free Episodic Memory}{3}{subsection.4.1}\protected@file@percent }
\citation{lewis2020rag}
\citation{meng2022rome}
\citation{meng2023memit}
\citation{meng2022rome,meng2023memit}
\bibstyle{plainnat}
\bibcite{das2024larimar}{{1}{2024}{{Das et~al.}}{{}}}
\@writefile{toc}{\contentsline {subsection}{\numberline {4.2}Retrieval-Augmented Generation}{4}{subsection.4.2}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {4.3}Knowledge Editing}{4}{subsection.4.3}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {4.4}What Distinguishes This Work}{4}{subsection.4.4}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {5}Limitations}{4}{section.5}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {6}Conclusion}{4}{section.6}\protected@file@percent }
\bibcite{fountas2024emllm}{{2}{2024}{{Fountas et~al.}}{{}}}
\bibcite{graves2014ntm}{{3}{2014}{{Graves et~al.}}{{}}}
\bibcite{graves2016dnc}{{4}{2016}{{Graves et~al.}}{{}}}
\bibcite{jang2024camelot}{{5}{2024}{{Jang et~al.}}{{}}}
\bibcite{lewis2020rag}{{6}{2020}{{Lewis et~al.}}{{}}}
\bibcite{meng2022rome}{{7}{2022}{{Meng et~al.}}{{}}}
\bibcite{meng2023memit}{{8}{2023}{{Meng et~al.}}{{}}}
\@writefile{toc}{\contentsline {section}{\numberline {4}Privacy by Representation}{4}{section.4}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {5}Related Work}{4}{section.5}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {6}Limitations}{4}{section.6}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {7}Conclusion}{4}{section.7}\protected@file@percent }
\bibcite{jang2024camelot}{{3}{2024}{{Jang et~al.}}{{}}}
\bibcite{lewis2020rag}{{4}{2020}{{Lewis et~al.}}{{}}}
\bibcite{meng2022rome}{{5}{2022}{{Meng et~al.}}{{}}}
\bibcite{meng2023memit}{{6}{2023}{{Meng et~al.}}{{}}}
\bibcite{pavlov1927}{{7}{1927}{{Pavlov}}{{}}}
\bibcite{skinner1938}{{8}{1938}{{Skinner}}{{}}}
\gdef \@abspage@last{5}

100
paper.log
View File

@@ -1,4 +1,4 @@
This is pdfTeX, Version 3.141592653-2.6-1.40.29 (TeX Live 2026/Arch Linux) (preloaded format=pdflatex 2026.3.11) 5 APR 2026 02:16
This is pdfTeX, Version 3.141592653-2.6-1.40.29 (TeX Live 2026/Arch Linux) (preloaded format=pdflatex 2026.3.11) 6 APR 2026 19:31
entering extended mode
restricted \write18 enabled.
%&-line parsing enabled.
@@ -339,8 +339,7 @@ LaTeX Info: Redefining \cite on input line 694.
(/usr/share/texmf-dist/tex/latex/l3backend/l3backend-pdftex.def
File: l3backend-pdftex.def 2025-10-09 L3 backend support: PDF output (pdfTeX)
\l__color_backend_stack_int=\count323
)
No file paper.aux.
) (./paper.aux)
\openout1 = `paper.aux'.
LaTeX Font Info: Checking defaults for OML/cmm/m/it on input line 21.
@@ -361,6 +360,7 @@ LaTeX Font Info: Checking defaults for PD1/pdf/m/n on input line 21.
LaTeX Font Info: ... okay on input line 21.
LaTeX Font Info: Checking defaults for PU/pdf/m/n on input line 21.
LaTeX Font Info: ... okay on input line 21.
*geometry* driver: auto-detecting
*geometry* detected driver: pdftex
*geometry* verbose mode - [ preamble ] result:
@@ -369,15 +369,15 @@ LaTeX Font Info: ... okay on input line 21.
* layout: <same size as paper>
* layoutoffset:(h,v)=(0.0pt,0.0pt)
* modes:
* h-part:(L,W,R)=(72.26999pt, 469.75502pt, 72.26999pt)
* v-part:(T,H,B)=(72.26999pt, 650.43001pt, 72.26999pt)
* h-part:(L,W,R)=(86.72377pt, 440.84746pt, 86.72377pt)
* v-part:(T,H,B)=(86.72377pt, 621.52245pt, 86.72377pt)
* \paperwidth=614.295pt
* \paperheight=794.96999pt
* \textwidth=469.75502pt
* \textheight=650.43001pt
* \oddsidemargin=0.0pt
* \evensidemargin=0.0pt
* \topmargin=-37.0pt
* \textwidth=440.84746pt
* \textheight=621.52245pt
* \oddsidemargin=14.45378pt
* \evensidemargin=14.45378pt
* \topmargin=-22.54622pt
* \headheight=12.0pt
* \headsep=25.0pt
* \topskip=11.0pt
@@ -396,10 +396,12 @@ LaTeX Font Info: ... okay on input line 21.
* (1in=72.27pt=25.4mm, 1cm=28.453pt)
Package hyperref Info: Link coloring OFF on input line 21.
(./paper.out) (./paper.out)
\@outlinefile=\write3
\openout3 = `paper.out'.
\c@lstlisting=\count324
(/usr/share/texmf-dist/tex/context/base/mkii/supp-pdf.mkii
[Loading MPS to PDF converter (version 2006.09.02).]
\scratchcounter=\count325
@@ -426,67 +428,29 @@ LaTeX Font Info: Trying to load font information for U+msb on input line 23.
File: umsb.fd 2013/01/14 v3.01 AMS symbols B
) [1
{/var/lib/texmf/fonts/map/pdftex/updmap/pdftex.map}] [2]
{/var/lib/texmf/fonts/map/pdftex/updmap/pdftex.map}]
Overfull \hbox (41.58173pt too wide) in paragraph at lines 118--122
\OT1/cmr/m/n/10.95 it (1536-dim, 35 lay-ers), E4B base (2560-dim, 42 lay-ers).
Quan-ti-za-tion tested at f32/f16/bf16/int8/int4
[]
Package natbib Warning: Citation `jang2024camelot' on page 3 undefined on input
line 196.
Package natbib Warning: Citation `fountas2024emllm' on page 3 undefined on inpu
t line 203.
Package natbib Warning: Citation `das2024larimar' on page 3 undefined on input
line 209.
[3{/usr/share/texmf-dist/fonts/enc/dvips/cm-super/cm-super-ts1.enc}]
Package natbib Warning: Citation `lewis2020rag' on page 4 undefined on input li
ne 215.
Package natbib Warning: Citation `meng2022rome' on page 4 undefined on input li
ne 222.
Package natbib Warning: Citation `meng2023memit' on page 4 undefined on input l
ine 222.
[4]
Package natbib Warning: There were undefined citations.
[5] (./paper.aux
Package natbib Warning: Citation(s) may have changed.
(natbib) Rerun to get citations correct.
)
[2{/usr/share/texmf-dist/fonts/enc/dvips/cm-super/cm-super-ts1.enc}] [3]
[4] [5] (./paper.aux)
***********
LaTeX2e <2025-11-01>
L3 programming layer <2026-01-19>
***********
LaTeX Warning: Label(s) may have changed. Rerun to get cross-references right.
Package rerunfilecheck Warning: File `paper.out' has changed.
(rerunfilecheck) Rerun to get outlines right
(rerunfilecheck) or use package `bookmark'.
Package rerunfilecheck Info: Checksums for `paper.out':
(rerunfilecheck) Before: <no file>
(rerunfilecheck) After: 7A9338885BB1CEC7210AED98C90C3277;2813.
Package rerunfilecheck Info: File `paper.out' has not changed.
(rerunfilecheck) Checksum: 4198C661EF4EF302CF6CA8FE934684BD;2020.
)
Here is how much of TeX's memory you used:
12312 strings out of 467525
179570 string characters out of 5425861
628915 words of memory out of 5000000
41060 multiletter control sequences out of 15000+600000
639253 words of font info for 85 fonts, out of 8000000 for 9000
12070 strings out of 467525
177761 string characters out of 5425861
608691 words of memory out of 5000000
40819 multiletter control sequences out of 15000+600000
639085 words of font info for 84 fonts, out of 8000000 for 9000
1141 hyphenation exceptions out of 8191
75i,8n,79p,304b,1079s stack positions out of 10000i,1000n,20000p,200000b,200000s
75i,8n,79p,287b,557s stack positions out of 10000i,1000n,20000p,200000b,200000s
</usr/share/texmf-dist/fonts/type1/public/amsfonts/cm/cmbx10.pfb></usr/share/
texmf-dist/fonts/type1/public/amsfonts/cm/cmbx12.pfb></usr/share/texmf-dist/fon
ts/type1/public/amsfonts/cm/cmex10.pfb></usr/share/texmf-dist/fonts/type1/publi
@@ -500,10 +464,10 @@ public/amsfonts/cm/cmr17.pfb></usr/share/texmf-dist/fonts/type1/public/amsfonts
mf-dist/fonts/type1/public/amsfonts/cm/cmtt10.pfb></usr/share/texmf-dist/fonts/
type1/public/amsfonts/cm/cmtt12.pfb></usr/share/texmf-dist/fonts/type1/public/c
m-super/sfrm1095.pfb>
Output written on paper.pdf (5 pages, 191888 bytes).
Output written on paper.pdf (5 pages, 192797 bytes).
PDF statistics:
157 PDF objects out of 1000 (max. 8388607)
118 compressed objects within 2 object streams
46 named destinations out of 1000 (max. 500000)
1 words of extra memory for PDF output out of 10000 (max. 10000000)
207 PDF objects out of 1000 (max. 8388607)
168 compressed objects within 2 object streams
39 named destinations out of 1000 (max. 500000)
121 words of extra memory for PDF output out of 10000 (max. 10000000)

BIN
paper.pdf
View File

Binary file not shown.

386
paper.tex
View File

@@ -1,5 +1,5 @@
\documentclass[11pt]{article}
\usepackage[margin=1in]{geometry}
\usepackage[margin=1.2in]{geometry}
\usepackage{amsmath,amssymb}
\usepackage{booktabs}
\usepackage{hyperref}
@@ -14,299 +14,263 @@
backgroundcolor=\color{gray!10},
}
\title{Solving the Clive Wearing Problem:\\One-Shot Episodic Memory for Frozen Transformers}
\title{Conditioned Reflex Injection:\\Stimulus-Response Learning for Frozen Transformers}
\author{Tommi Niemi\\Rotko Networks\\\texttt{tommi@rotko.net}}
\date{April 2026}
\date{April 2026 --- DRAFT}
\begin{document}
\maketitle
\begin{abstract}
We enable frozen transformers to form new memories without gradient descent.
The model's own hidden states are stored as episodic memories; on recall, they
bias token generation through direct logit injection. A frozen Qwen~2.5~0.5B
taught three novel facts recalls all three at 100\% accuracy. No weights are
modified. No gradients are computed. Memories persist to disk across sessions.
Code and reproduction: \url{https://git.rotko.net/tommi/epimem}.
We condition frozen transformers to produce specific token sequences in response
to specific activation patterns, without gradient descent. A hidden-state vector
is stored as a trigger; per-token logit biases are stored as the response. At
inference, cosine similarity fires the matching reflex. Tested on Qwen~2.5~0.5B,
Gemma~4 E2B-it, E4B-it, and E4B~base at precisions from float32 to int4.
Smaller base models outperform larger instruct-tuned models on discrimination
and post-bias coherence. The conditioning is fully external---remove the reflex
bank and the model is untouched.
Code: \url{https://git.rotko.net/tommi/cri}.
\end{abstract}
\section{The Clive Wearing Problem}
%───────────────────────────────────────────────
\section{Conditioning, Not Memory}
Clive Wearing lost his hippocampus to encephalitis in 1985. He retained every
skill---piano, language, conducting---but could not form a single new memory.
Every 7~seconds, he believed he had just woken up for the first time. His
diary: ``8:31~AM Now I am awake. 8:34~AM Now I am properly awake.'' Each entry
crossed out moments later.
CRI does not give a model memory or knowledge. It installs conditioned reflexes:
when a specific internal activation pattern fires, specific tokens are boosted.
The model has no representation of the association.
Current LLMs are Clive Wearing. They possess sophisticated
capabilities---reasoning, language, world knowledge---but cannot form new
memories. Every conversation starts from zero. The context window is their
7-second span. When it clears, everything is gone.
This is Pavlovian conditioning at the logit level. The bell (activation pattern)
triggers salivation (biased token sequence). The association persists in an
external reflex bank. The model weights are never modified. Remove the file and
the model is exactly as it was---no trace, no residue.
Fine-tuning modifies weights and causes catastrophic forgetting. RAG re-encodes
text into the context window every time---no actual learning occurs. LoRA still
requires gradients. In-context learning vanishes when the conversation ends.
The closer analogy is post-hypnotic suggestion: a trigger installed externally,
fired without the subject's awareness, removable without leaving a mark.
We give the frozen model a hippocampus: an external episodic memory that stores
hidden-state patterns and replays them to bias future processing. The backbone
never changes. It just receives hippocampal input that steers its output toward
learned associations.
Fine-tuning modifies weights. RAG re-encodes text each time. LoRA requires
gradients. In-context learning vanishes with the conversation. CRI persists
across sessions without touching the model.
%───────────────────────────────────────────────
\section{Method}
\subsection{Architecture}
Two components:
\textbf{Frozen backbone}: any transformer. Produces hidden-state vectors from
input tokens. Weights never modified.
\textbf{Frozen backbone} (Qwen~2.5~0.5B, 896-dimensional hidden states): The
pretrained transformer. Processes input tokens, produces hidden state vectors.
Weights are never modified at any point.
\textbf{Reflex bank}: stores (trigger, response) pairs:
\begin{itemize}
\item \textbf{Trigger}: hidden-state vector $\mathbf{h}$ at the final token
position---the model's activation pattern for a given input.
\item \textbf{Response}: per-position logit biases
$\{(t_i, b_i)\}$---one pair per answer token.
\end{itemize}
\textbf{Episodic memory bank}: A key-value store where the \emph{key} is the
backbone's hidden state vector at the final token position---the model's
internal representation of the prompt in its own learned space---and the
\emph{value} is per-position logit biases for the correct continuation
tokens---which token to boost at each generation step.
Both are sub-symbolic. The trigger is an opaque high-dimensional vector; the
response is a list of (integer, float) pairs. The reflex bank resists inspection
without the backbone that produced it.
\subsection{Teaching (One Forward Pass)}
\subsection{Conditioning}
Given a prompt $P$ and desired answer $A$:
Given stimulus $P$ and desired response $A$:
\begin{enumerate}
\item \textbf{Extract key}: Run backbone on $P$. Extract hidden state
$\mathbf{h} = \text{backbone}(P)$ at the final token. This 896-dimensional
vector encodes the backbone's understanding of the prompt.
\item \textbf{Compute logit biases}: Run backbone on the concatenation $P
\mathbin\Vert A$. At each answer token position $i$, compute the gap between
the correct token's logit and the maximum logit. The bias overcomes this gap
plus a margin:
\item $\mathbf{h} = \text{backbone}(P)$ at final token. This is the trigger.
\item Run backbone on $P \mathbin\Vert A$. At each answer position $i$:
\[
b_i = \max\!\bigl(\max_j \ell_j - \ell_{t_i},\; 5.0\bigr) + 5.0
b_i = \max\!\bigl(\max_j \ell_j - \ell_{t_i},\; 5.0\bigr)
\]
where $\ell_j$ are logits at position $i$ and $t_i$ is the correct token.
This produces one $(t_i, b_i)$ pair per answer token.
\item \textbf{Store}: Save $(\text{key}=\mathbf{h},\;
\text{value}=\{(t_i, b_i)\})$ to the memory bank.
\item Store $(\text{trigger}=\mathbf{h},\;\text{response}=\{(t_i, b_i)\})$.
\end{enumerate}
One forward pass. No iteration. No loss function. No gradients.
One forward pass. No gradients.
\subsection{Recall (Similarity Search + Injection)}
\subsection{Triggering}
Given a new query $Q$:
Given query $Q$:
\begin{enumerate}
\item \textbf{Extract query key}: $\mathbf{h}_q = \text{backbone}(Q)$ at the
final token.
\item \textbf{Search}: For each stored episode, compute cosine similarity
$\cos(\mathbf{h}_q, \mathbf{h}_{\text{stored}})$. Return the best match above
threshold.
\item \textbf{Generate with injection}: At generation step $i$, if the matched
episode has a logit bias $(t_i, b_i)$ for step $i$, add $b_i$ to the
backbone's logit for token $t_i$ before sampling. After all biases are applied,
the backbone continues generating freely.
\item $\mathbf{h}_q = \text{backbone}(Q)$ at final token.
\item $\cos(\mathbf{h}_q, \mathbf{h}_{\text{stored}})$ against all triggers.
Best match above threshold fires.
\item At generation step $i$, add $b_i$ to logits before argmax. After biases
exhaust, backbone generates freely.
\end{enumerate}
The backbone generates fluent text beyond the taught answer---the logit biases
seed the first tokens, and the language model's coherence completes the
sentence naturally.
\subsection{Persistence}
The memory bank serializes to JSON: each episode stores the 896-dimensional key
vector and the list of $(t_i, b_i)$ pairs. Load the file, and all memories are
available. No retraining. No warm-up. Instant recall.
Post-bias fluency is model-dependent. Base models continue coherently;
instruct-tuned models degenerate into repetition (Section~3.2).
\subsection{Why Hidden States, Not Text}
RAG stores text and re-encodes it. This has three costs: (1)~context window
consumption---retrieved passages compete with the actual input for attention;
(2)~re-encoding latency---the backbone must process retrieved text tokens;
(3)~representation mismatch---the retrieval embedding space (typically a
separate encoder) doesn't match the generative model's internal space.
Storing hidden states eliminates all three. The memory is already in the
backbone's native representation. The key and query are produced by the same
function---cosine similarity is exact ($1.000$ for identical prompts). Injection
is a single scalar addition to one logit per generation step.
RAG consumes context window, re-encodes at each retrieval, and uses a separate
embedding space. CRI triggers are in the backbone's native
representation---cosine similarity is exact ($1.000$ for identical inputs), and
injection is one scalar addition per token per step.
%───────────────────────────────────────────────
\section{Experiments}
\label{sec:results}
\subsection{Setup}
\begin{itemize}
\item \textbf{Backbone}: Qwen~2.5~0.5B (896-dim hidden states)
\item \textbf{Inference}: PyTorch via HuggingFace \texttt{transformers}
(also works with ONNX Runtime)
\item \textbf{Hardware}: Any machine with Python~3 and $\sim$2\,GB RAM. No GPU
required.
\item \textbf{Gradient computation}: None. At no point---not during teaching,
recall, or persistence.
\end{itemize}
Four backbones: Qwen~2.5~0.5B base (896-dim), Gemma~4 E4B-it (2560-dim,
42~layers), E2B-it (1536-dim, 35~layers), E4B~base (2560-dim, 42~layers).
Quantization tested at f32/f16/bf16/int8/int4 on Qwen. PyTorch inference, CPU,
no gradients at any point.
\subsection{One-Shot Fact Learning}
\subsection{One-Shot Conditioning}
We teach three facts about ``Zyphraxia''---a word absent from Qwen's training
data:
Three reflexes conditioned on ``Zyphraxia'' (absent from all training data).
Conditioned tokens correct on all backbones (sim~$= 1.000$). Post-bias behavior
diverges:
\begin{table}[h]
\centering
\begin{tabular}{lllc}
\begin{tabular}{llll}
\toprule
Prompt & Taught & Recalled & Sim. \\
Backbone & Post-bias behavior & Fluent? \\
\midrule
``The capital of Zyphraxia is'' & Novaheim & Novaheim, a city of 100 & 1.000 \\
``The ruler of Zyphraxia is'' & Queen Stellara & Queen Stellara. She is\ldots & 1.000 \\
``The currency of Zyphraxia is'' & Glimmers & Glimmers. The currency\ldots & 1.000 \\
Qwen 2.5 0.5B base & Coherent continuation & Yes \\
Gemma 4 E4B base & Stutters, hits EOS & Partial \\
Gemma 4 E4B-it & Repetition loops & No \\
Gemma 4 E2B-it & Repetition loops & No \\
\bottomrule
\end{tabular}
\caption{One-shot fact recall. All three novel facts recalled correctly with
cosine similarity 1.000. The backbone generates fluent continuations beyond the
taught answer.}
\label{tab:results}
\caption{Post-bias degeneration is caused by instruct tuning, not model size.}
\label{tab:postbias}
\end{table}
\subsection{Persistence}
\subsection{Stimulus Generalization and Misfire}
The memory bank is saved to JSON (77\,KB for 3 episodes with 896-dim keys).
After reloading from disk, all three facts are recalled identically: 3/3
pre-save, 3/3 post-reload.
Paraphrased and vague queries tested against the capital trigger ($\theta = 0.3$):
\subsection{Reproduction}
\begin{table}[h]
\centering
\begin{tabular}{lcccc}
\toprule
Query & Qwen & E4B base & E4B-it & E2B-it \\
\midrule
``What is the capital of Z.?'' & 0.832 & 0.873 & 0.932 & 0.932 \\
``Zyphraxia's capital is'' & 0.969 & 0.935 & 0.974 & 0.970 \\
``Tell me about Novaheim'' & 0.756 & 0.891 & 0.940 & 0.924 \\
``Name three facts about Z.'' & 0.761 & 0.884 & 0.943 & 0.920 \\
``\ldots Who rules it?'' & 0.827 & 0.889 & 0.918 & 0.930 \\
\midrule
\textbf{Spread} & \textbf{0.213} & \textbf{0.062} & \textbf{0.056} & \textbf{0.038} \\
\bottomrule
\end{tabular}
\caption{Cross-model discrimination. Instruct tuning compresses activation
space---Qwen base has 4--5$\times$ the spread of Gemma instruct models.}
\label{tab:discrimination}
\end{table}
\begin{lstlisting}
git clone https://git.rotko.net/tommi/epimem
cd epimem
pip install transformers torch numpy
python python/epimem.py
\end{lstlisting}
\subsection{Quantization Tolerance}
Downloads Qwen~2.5~0.5B from HuggingFace ($\sim$1\,GB, cached after first
run). Teaches 3~facts, recalls 6/6 (3~pre-save + 3~post-reload). Runs in
$\sim$30~seconds after model is cached.
\begin{table}[h]
\centering
\begin{tabular}{lcccc}
\toprule
Query & f32 & f16 & int8 & int4 \\
\midrule
Self (capital trigger) & 1.000 & 1.000 & 1.000 & 1.000 \\
``What is the capital?'' & 0.832 & 0.832 & 0.826 & 0.808 \\
``Something about a queen\ldots'' & 0.752 & 0.752 & 0.754 & 0.742 \\
``Remind me about that currency'' & 0.733 & 0.732 & 0.736 & 0.727 \\
\midrule
Cross-precision self-sim (vs f32) & --- & 0.9999 & 0.9985 & 0.9440 \\
\bottomrule
\end{tabular}
\caption{Actual quantized inference (bitsandbytes, Qwen). Same-precision
self-match is always 1.000. Cross-precision f32$\to$int4 drops to 0.944.}
\label{tab:quant}
\end{table}
CRI works at any precision if conditioning and triggering match.
Cross-precision reflex banks are unreliable.
%───────────────────────────────────────────────
\section{Privacy by Representation}
Trigger patterns are points in a model-specific activation space---meaningless
without the exact backbone. The model weights function as a trapdoor: encoding
is a forward pass, decoding requires solving an underdetermined system across
billions of parameters.
An adversary with the reflex bank but not the backbone learns nothing. An
adversary with both can enumerate response tokens but cannot determine what
stimuli trigger them without brute-force search over the input space.
Privacy by representation, not encryption---an architectural consequence of
operating in the model's internal space.
%───────────────────────────────────────────────
\section{Related Work}
\subsection{Training-Free Episodic Memory}
\textbf{Pavlov}~(1927): classical conditioning. CRI operates analogously---activation
pattern (CS) paired with logit biases (US) produces token sequence (CR).
\textbf{Skinner}~(1938): operant conditioning. CRI currently performs respondent
conditioning only; bias modulation via reward is a natural extension.
\textbf{CAMELoT} \citep{jang2024camelot} is the closest prior work: a
training-free consolidated associative memory for frozen LLMs. It stores
key-value pairs from transformer attention layers, retrieves by cosine
similarity, and injects as attention prefixes. Our approach differs in what is
stored (logit biases vs.\ KV pairs) and where injection occurs (output logits
vs.\ attention mechanism).
\textbf{CAMELoT} \citep{jang2024camelot}: KV pairs from attention, injected as
prefixes. \textbf{EM-LLM} \citep{fountas2024emllm}: KV cache extension.
\textbf{Larimar} \citep{das2024larimar}: memory matrix, requires training.
All inject at attention level. CRI injects at output logits---simpler, cheaper,
no attention recomputation.
\textbf{EM-LLM} \citep{fountas2024emllm} stores KV pairs from attention heads
as episodic events, retrieves by $k$-NN with temporal contiguity, and prepends
retrieved pairs into the context window. The backbone is frozen and no training
is required. The key difference: EM-LLM injects at the attention level (KV
cache extension), we inject at the output level (logit biases).
\textbf{Larimar} \citep{das2024larimar} adds episodic memory to frozen LLMs
via a memory matrix with pseudo-inverse retrieval. Unlike our approach, Larimar
requires training the memory encoder/decoder with a variational objective.
\subsection{Retrieval-Augmented Generation}
RAG \citep{lewis2020rag} retrieves text passages and inserts them into the
context window. The model re-encodes retrieved text each time. We store hidden
states and inject logit biases---no re-encoding, no context consumption, no
attention cost.
\subsection{Knowledge Editing}
ROME \citep{meng2022rome} and MEMIT \citep{meng2023memit} edit factual
associations by modifying specific weight matrices via rank-one updates. Our
method makes zero modifications to any weight.
\subsection{What Distinguishes This Work}
All prior training-free episodic memory systems inject at the attention
level---modifying KV caches, prepending context, or adding cross-attention. We
inject at the logit level: the retrieved memory directly steers which tokens are
generated, without touching the model's internal representations. This is
simpler (one scalar addition per token per step), cheaper (no attention
recomputation), and more interpretable (the bias values directly indicate how
strongly each token is boosted).
\textbf{RAG} \citep{lewis2020rag}: retrieves text, re-encodes. RAG informs; CRI
conditions. \textbf{ROME/MEMIT} \citep{meng2022rome,meng2023memit}: rank-one
weight edits. CRI modifies zero weights.
%───────────────────────────────────────────────
\section{Limitations}
\textbf{Backbone lock-in.} Memories are tied to the specific backbone. Changing
the model invalidates all stored keys. Migration requires re-encoding through
the new backbone.
\textbf{Key collision.} Semantically different prompts with similar hidden
states may trigger incorrect recall. A similarity threshold mitigates this but
doesn't eliminate it.
\textbf{Linear scan.} Retrieval is $O(n)$ over stored episodes. For banks
exceeding ${\sim}100$K episodes, approximate nearest neighbor indexing would be
needed.
\textbf{Per-position biases.} The current implementation stores biases per
generation step. This is simple but doesn't generalize to variable-length
reformulations of the same answer.
\textbf{Backbone lock-in}: reflexes don't transfer across models.
\textbf{Trigger collision}: similar activations fire incorrect reflexes.
\textbf{Linear scan}: $O(n)$ retrieval; needs ANN past ${\sim}100$K reflexes.
\textbf{Per-position biases}: doesn't generalize to reformulations.
\textbf{One-shot rigidity}: no reinforcement or extinction.
\textbf{Post-bias degeneration}: instruct models loop after biases exhaust.
\textbf{Discrimination degrades with instruct tuning}: RLHF compresses
activation spaces (Qwen: 0.213 spread; Gemma E4B-it: 0.056).
\textbf{Cross-precision fragility}: condition and trigger must match precision.
%───────────────────────────────────────────────
\section{Conclusion}
Frozen transformers cannot form new memories. We give them a hippocampus.
Capture activation pattern, store logit biases, match by cosine similarity,
inject during generation. One forward pass to condition. One lookup to trigger.
Remove the file and the model is untouched.
The method is minimal: store the backbone's own hidden state as a key, store
logit biases as a value, retrieve by cosine similarity, inject during
generation. No gradients. No weight changes. No training loop. One forward pass
to teach. One lookup to recall. Memories persist to disk.
The 200-line Python implementation reproduces the full result. The Clive
Wearing Problem---intelligent systems that cannot form new
memories---has a working solution.
Not a hippocampus---a reflex arc.
\bibliographystyle{plainnat}
\begin{thebibliography}{10}
\bibitem[Das et~al.(2024)]{das2024larimar}
Das, P., Natarajan, S., Singh, S., et~al.
\newblock Larimar: Large Language Models with Episodic Memory Control.
\newblock \emph{ICML}, 2024. arXiv:2403.11901.
Das, P. et~al. Larimar. \emph{ICML}, 2024. arXiv:2403.11901.
\bibitem[Fountas et~al.(2024)]{fountas2024emllm}
Fountas, Z., Bisk, Y., et~al.
\newblock Human-inspired Episodic Memory for Infinite Context LLMs.
\newblock arXiv:2407.09450, 2024.
\bibitem[Graves et~al.(2014)]{graves2014ntm}
Graves, A., Wayne, G., and Danihelka, I.
\newblock Neural Turing Machines.
\newblock arXiv:1410.5401, 2014.
\bibitem[Graves et~al.(2016)]{graves2016dnc}
Graves, A., Wayne, G., et~al.
\newblock Hybrid computing using a neural network with dynamic external memory.
\newblock \emph{Nature}, 538:471--476, 2016.
Fountas, Z. et~al. EM-LLM. arXiv:2407.09450, 2024.
\bibitem[Jang et~al.(2024)]{jang2024camelot}
Jang, J., et~al.
\newblock CAMELoT: Towards Large Language Models with Training-Free
Consolidated Associative Memory.
\newblock arXiv:2402.13449, 2024.
Jang, J. et~al. CAMELoT. arXiv:2402.13449, 2024.
\bibitem[Lewis et~al.(2020)]{lewis2020rag}
Lewis, P., Perez, E., et~al.
\newblock Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks.
\newblock \emph{NeurIPS}, 2020.
Lewis, P. et~al. RAG. \emph{NeurIPS}, 2020.
\bibitem[Meng et~al.(2022)]{meng2022rome}
Meng, K., Bau, D., Mitchell, A., and Finn, C.
\newblock Locating and Editing Factual Associations in GPT.
\newblock \emph{NeurIPS}, 2022.
Meng, K. et~al. ROME. \emph{NeurIPS}, 2022.
\bibitem[Meng et~al.(2023)]{meng2023memit}
Meng, K., Sharma, A., Andonian, A., et~al.
\newblock Mass-Editing Memory in a Transformer.
\newblock \emph{ICLR}, 2023.
Meng, K. et~al. MEMIT. \emph{ICLR}, 2023.
\bibitem[Pavlov(1927)]{pavlov1927}
Pavlov, I.~P. \emph{Conditioned Reflexes}. Oxford University Press, 1927.
\bibitem[Skinner(1938)]{skinner1938}
Skinner, B.~F. \emph{The Behavior of Organisms}. Appleton-Century, 1938.
\end{thebibliography}