Technical Note

Gemma 4 31B-IT Topology Capture

Deterministic topology capture for Gemma 4 31B-IT on a bounded CPU surface.

Deterministic Local Topology Capture for Gemma 4 31B-IT

Simple version

We carefully recorded a small map of how an AI model worked while it answered a question on an ordinary computer.

In plain terms

When a model produces an answer, lots of internal connections are exercised. We captured a bounded slice of those connections on a normal CPU, confirmed the recording itself did not change the answer, and saved the map so anyone else can reproduce it from the same inputs.

Technical artifact

Deterministic local topology capture for Gemma 4 31B-IT on a bounded CPU surface, with frozen anchors, non-perturbing capture, and canonical topology.v2 emission.

Evidence path

  1. Step 1

    Run the model on a bounded CPU surface

    Gemma 4 31B-IT runs locally on a CPU path against real weights with a frozen compute baseline.

  2. Step 2

    Capture topology without perturbing the run

    Hidden-state and topology capture run non-perturbingly: instrumented and uninstrumented runs match.

  3. Step 3

    Emit canonical topology artifacts

    topology.v2 artifacts emitted under a binary admission rule, including an attention-edge surface.

  4. Step 4

    Verify bit-for-bit reproduction

    Artifacts reproduce bit-for-bit across independent capture runs on the same surface.

Artifact details

Surface
Model Topology
Artifact
Technical note · PDF
Model
Gemma 4 31B-IT
Compute
Local CPU · bounded measurement surface
Evidence collected
topology.v2 · attention edges
Verification result
Bit-for-bit reproduction across runs

Scope of this note

This page summarizes one bounded topology capture against a frozen compute baseline. The PDF contains the full admission rule, edge schema, and reproduction details.

Open the technical artifact

The PDF contains the full admission rule, edge schema, and bit-for-bit reproduction details.

Have an artifact?

Verify a receipt, digest, or bundle.

Verify a receipt