diff --git "a/index.html" "b/index.html" --- "a/index.html" +++ "b/index.html" @@ -1,5 +1,1500 @@ -Internectual Labs · The creators of TAARDIS
Independent research / efficient AI

InternectualLabs

The creators of TAARDIS

Smaller footprints. More room to explore.

Exploring ternary computing, model compression, and practical tools for efficient AI. Built by Cody Dixon at Internectual Labs.

Explore ternary ↓Visit the lab ↗
Internectual Labs metallic hexagonal emblem
01 Ternary computing02 Model compression03 Reproducible evaluation
TAARDIS call-box artwork with blue and gold circuitry and the words Small on the outside, big on the inside.
Meet TAARDIS / an Internectual Labs project

Small on the outside.
Big on the inside.

TAARDIS reflects the lab’s drive to explore smaller model footprints and practical, efficient AI.

Follow the project through the lab’s public updates, explore the ideas below, or help support the next round of experiments.

Explore TAARDIS at the lab ↗Support the research ↓
Interactive / a small idea, made tangible

Three states.
A different way to pack.

Each tile is a trit: −1, 0, or +1. Tap a tile to cycle its state and see five ternary values represented as one base-3 code.

TRITS
BASE-3 CODE / 242
8-BIT CONTAINER
Educational encoding example: map −1 → 0, 0 → 1, +1 → 2. Five trits have 243 combinations, which fit in eight bits. This illustrates representation only; it does not establish model quality, inference speed, or total model size.
The research direction

Make every experiment count.

TAARDIS is part of an independent effort to explore smaller model representations and the tools needed to understand them.

  • Build & inspectExplore efficient representations and practical model tools.
  • Measure carefullyCompare quality, memory use, and speed with documented conditions.
  • Share evidenceMake public work easier to inspect, discuss, and reproduce.
Explore public projects ↗This Space is a research showcase and educational interaction. It does not run a TAARDIS model or provide model inference.
Help the next experiment happen

Support independent research.
Bring more ideas to the test.

I welcome sponsored GPU compute, research collaborators, and financial support for training and reproducible evaluation. More compute creates room to test ideas and document what works.

Sponsor the work ↗Discuss compute or collaboration ↗More ways to support ↗
Seeking support; no GPU grant or sponsorship award is implied.
\ No newline at end of file + + + + + + + + + + + + + + Internectual Labs · Efficient AI & Experimental Research + + + + + + + + + + + + + + +
+
+ +
+
+
+
+ + + + +
01 / TERNARY COMPUTINGLess weight.
More possibility.
+
INDEPENDENT RESEARCHExploring structure.
Building intelligence.
+
Current powers the core
+
+
+ Support the lab +
+ Sponsor on GitHub + Cash App + PayPal + Venmo +
+
+
+

Ternary computing · Wave architectures · Efficient AI

+

INTERNECTUAL LABS

+

The creators of TAARDIS

+

Efficient Ternary LLMs and Experimental AGI Architectures

+

Big questions. Hands-on experiments.
Building what comes next, one idea at a time.

+
Enter the lab Explore the source
+
+
+
01 Ternary computing02 Experimental architectures03 Tools for builders
+ +
+

Selected work

Ideas you can explore.

Open the source. Inspect the approach.
Help shape what comes next.

+
+

01 / Ternary computing

TK Integers.

Give model weights
a smaller footprint.

Base-3 weight packing, selectable weight and scale formats, GPU kernel experiments, and a Model Atlas for inspecting per-layer choices.

Explore TK Integers Read the report ↗
+
+
A little hands-on exampleInteractive
+

Explore ternary.

Tap a tile to change its value and watch the encoding update.

+
+ +
Encoded value0011000048 / 242
+

Each tile has three states: −1, 0, and +1.

+
+
+ +
+

From the published specification

A concrete
storage example.

TK k1 can describe a full group of 20 weights using 32 code bits and a 12-bit scale.

How the numbers are counted

Four bytes hold the 20 one-trit codes. A packed 12-bit scale adds 12 bits: (32 + 12) ÷ 20 = 2.2 bits per weight. This nominal full-group comparison excludes padding and file metadata. It does not measure inference speed or model quality.

+
Nominal bits per weightLower is smaller
BF1616.0
FP88.0
TK k1 · g20 · s122.2

Calculated representation size · full groups · packed scales

Source & assumptions ↗
+
+ +
+
GRIP robotic hand holding an electric blue illuminated core, with the slogan Tell your AI to get a GRIP.

02 / Agent workflow

Get a GRIP.

Keep an AI-assisted project moving. Save progress, capture what changed and what was tested, then continue when you choose.

Explore GRIP
+
TAARDIS metallic call box with blue and gold circuits and the words Small on the outside, big on the inside.

03 / Full-ternary model

TAARDIS-27B.

Explore the lab’s full-ternary model project, its published files, and the runtime work behind it. Usage and limitations live in the model card.

Explore the model
+
+
+ +

On the workbench

The next questions.

Research in progress.
Room to experiment.

04

GNU-AI-Flow

A visual workspace for AI components, model flows, and training pipelines, built around the GNU Radio Companion interface.

In development
05

IQ Wave Models

Exploring geometric models using in-phase and quadrature (IQ) waves.

Experimental
+
+

06 / Geometric architectures

+

Geometric Models.

+ Planned model shapes +

From geometry
to the emblem.

+

A visual study of the lab’s emblem. The folding shapes in this sequence are planned directions for future models.

+ Explore current model shapes +
+
+
Form in motionTriangle net
+
+

Three outer triangles fold around a fourth face.

+
+ + + + + + +
+
Fold · unfold · assemble
+
+
+
+
+

Current model geometries

Exploring the shape
of a model.

+

Explore the individual geometric components, then the complete IQ wave model built from connected fractal crosses and dodecahedron chambers.

+
+
+
+
01 / Recursive structure3D
+ +

Fractal cross

A cross branching into smaller crosses in three dimensions.

+
Drag to turn
+
+
+
02 / Polyhedral structure3D
+ +

Dodecahedron

Twelve pentagonal faces enclosing a connected volume.

+
Drag to turn
+
+
+
03 / Continuous structure3D
+ +

Torus

A ring-shaped surface looping around a central opening.

+
Drag to turn
+
+
+
04 / Complete modelCurrently training
+ +

Geometry working together

The complete
IQ wave model.

Connected fractal crosses link large dodecahedron chambers, with smaller repeating cross forms along each branch.

This combined geometry is the wave model currently being trained, exploring geometric models through in-phase and quadrature (IQ) waves.

Connected planar crossesDodecahedron chambers
+

Color = phase · brightness + size = amplitudeIllustrative wave activity

+
Drag to explore the assembled geometry
+
+
+

Three component studies and the complete wave model currently being trained. The emblem sequence above shows additional shapes planned for future models.

+
+

These projects are not released yet. Follow the lab for public updates.

+ +
Internectual Labs emblemHUEYTOWN, ALABAMA

The person behind the experiments

A lab built
on curiosity.

INTERNECTUAL (adj.)
Someone who has cultivated a substantial body of knowledge via the web.

I’m Cody Dixon, a hands-on researcher and builder in Alabama. I connect ideas with published research and use AI to help turn those ideas into working tools.

“Today’s minds are always on the internet, but it’s what you do with it that sets you apart.”— Cody Dixon
More about my work ↗
+ +

Support independent research

Help the next
experiment happen.

More GPU time means more room to evaluate models, reproduce results, and keep practical tools moving forward.

The next goal

Broader model evaluation.

Support testing that puts reasoning quality, memory use, and speed side by side—with the setup and results documented.

Compute + reproducible experiments
Or support directlyCash App ↗PayPal ↗Venmo ↗
+ +

Build with me

Have an idea worth testing?

Reproduce a result, report a bug, improve a tool, or bring a different perspective.

Join the community ↗
+
+ +
+ + + + + + +