Repository navigation
Releases
Model Mapper 0.2.0: labelling
Model Mapper can now say what each part of a neuron does, and check every label.
- What it labels: the gate (what opens the neuron) and the two sides of its up reading (what pushes it one way or
the other), each separately. - How labels are checked: a blind judge examines every label against look-alike decoys and other neurons' labels.
Words newly written from the label have to make the neuron fire. Fake neurons go through the same steps and should
almost never pass. - One command:
modelmapper labellabels a list of neurons, whole layers, a sample or the whole model. It shows a
plan with the cost and time first, can resume after a stop, and stops at a spend cap. - Self-check:
modelmapper check --labelstests labelling on your own machine against our values. - Results on Qwen3-8B, the same figures as the explainer:
- 11,000 neurons: 68% of gates and about 45% of each side pass the exam;
- of those, 1,000 drawn at random: 46% of gates and about 31% of each side;
- fake neurons: about 2%;
- about $0.009 per neuron in API calls.
- Published labels:
modelmapper lookupshows labels for all 11,000 labelled Qwen3-8B neurons, offline. - Contributions:
modelmapper export-packwrites your runs in the research ledger's format.
Labels whose exam passed while other neurons' labels or token lists also scored above 0.65 count as verified "at the
edge", with lower confidence. That was decided after the runs, because on real text they pass nearly as often as other
verified labels.
Experimental: checking labels on real text (modelmapper realtext). This is the research code for two further
checks that ran on Qwen3-8B. It's tested offline here, but not yet re-run.
-
Readings: one pass over a corpus (for example FineWeb web pages) finds each neuron's strongest moments in ordinary
text. -
Corpus exam: from the label alone, the judge has to pick those moments out from three kinds of decoy:
- the same word in a passage where the neuron stays quiet, which holds the word fixed;
- strong moments of neurons on the same topic;
- strong moments of random neurons.
On 12,244 neurons, 81% of verified gate labels passed against same-word decoys, against 2.2% of fakes. Only 37%
could tell the neuron's strongest moments from others on the same topic. -
Graded simulation: from the label alone, the judge rates how strongly the neuron fires at 60 web positions, from
its strongest moments down to ordinary ones. The score is the rank correlation with the real values. Across every
labelled Qwen3-8B neuron, verified labels scored 0.65 for gates and 0.55 for sides, against about 0.1 for other
neurons' labels and 0.13-0.17 for fakes. Without the strongest moments, the scores were 0.42 and 0.24: labels describe
the peaks best. The run cost $36 for about 36,600 labels, about $0.001 each.
Tested on Qwen3-0.6B and Qwen3-8B, with the labeller gpt-6-luna and the judge Jev. A fresh clone on a 6 GB laptop GPU
reached labelled neurons by following SETUP.md. Other models, labellers and judges haven't been tested.
Model Mapper 0.1.0
Model Mapper 0.1.0: a clean, tested package. Give it a concept; it finds which neurons of a language model the
answer depends on, by switching neurons off one at a time against random neurons switched off the same way.
modelmapper checkverifies an install against our numbers and writes a report you can send back.modelmapper map-concept "X"/map-batchmap one concept or a list (resumable);steerturns chosen neurons up;
lookupreads published records. Python API:from modelmapper import map_concept.- Every record carries its provenance (GPU, software versions), and neurons near the top-30 cut-off are labelled
uncertain using a band calibrated across GPU types (METHOD.md). - Tested on two machines: a Windows 10 desktop (CPU, float32) and a Windows 11 laptop (RTX 1000 Ada 6 GB, bfloat16,
Python 3.10), both passingcheck --apiswith the same top neurons (CALIBRATION.md). - FINDINGS.md: what we have found on Qwen3-8B, how sure we are, and the corrections made after an
independent review (some figures are provisional while the last 1,000 concepts finish).
This replaces the research history that was in this repository. The full record of the 18 Oct 2026 release, every
script and result included, stays at doi:10.5281/zenodo.23459926. The
explainer and the Qwen Neuron Map now live at jphoenix-dev/eiffel-tower-qwen.
Model Mapper 2026-10-18
Neuron/concept map of Qwen3-8B, two days on from the first release:
- The ledger (LEDGER.md): every measurement in one queryable, append-only store (5.6 million evidence rows), with a format for third-party contributions.
- The loop at scale: 4,194 Wikipedia Vital Articles concepts mapped (loop v2 run 1), plus a snapshot of the loop v3 enrichment (1,231 concepts).
- What each layer band's neurons do (LAYERS_AND_JOBS.md): facts, numbers, eras, the J-lens pilot, attribution graphs, two-ended scales, new job categories.
- The Qwen Neuron Map on the explainer site: every neuron tied to a job, by layer and subject.
- The
modelmapperpackage and command line: SCRIPTS.md maps every old script path to its new command.
Findings are exploratory. RESULTS.md logs every run, failures included. Still in progress at this snapshot: the full loop v3 enrichment and the 256-window J-lens (next update).