document end and bare page counts
A neuronOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in layerOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. 12 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featuredA neuron picked for the full treatment: steering outputs and a written explanation. About 3,000 of the 86,000 neurons are featured., so it has steering outputs.
What this neuron does
This neuron responds to places where one document seems to stop and another begins, marked by the end-of-text token, and to some page-count or catalogue details. Its strongest negative responses are commas and bars in book listings such as '250 pages, 20 x 17 cm'. Steering pushed the output toward lists of numbers and dates, but the link is weak.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. Think it is wrong? Flag it (opens a prefilled issue on GitHub).
Fires most on
The text windows where this neuron's activationThe number a neuron produces for a token. Large positive means it fired hard; negative means it pushed the other way. was highest. Amber marks positive activation; the outlined tokenA piece of text the model reads at once: often a word, sometimes part of a word or a punctuation mark. is the peak. Hover a token for its strength.
-
functional genomics archive at EMBL-EBI, led by Alvis Brazma who is part of the GEUVADIS analysis group. Open access<|endoftext|>A consideration of some ecological factors influencing the stabilizing of these dune areas located
-
ITATION: Carlos Peres and Maurício Schneider. Subsidized agricultural resettlements as drivers of tropical deforestation. Biological Conservation (2011)<|endoftext|>Neuroimmune Toxicity Recovery Programs⏎Chronic Inflammatory Response Syndrome (C
-
, the right kind method to pursue happiness, and the correct plan for an excellent lifestyles.⏎Read or Download Aristotle for Everybody: Difficult Thought Made Easy PDF⏎Best greek &
-
was among famous benefactors of the Museum.⏎American Museum of Natural History, Central Park West at 79th Street , New York, NY 10<|endoftext|>|
-
on states; Richard D. Schaller, Vladimir M. Agranovich and Victor I. Klimov; Nature Physics 1, 189<|endoftext|>d.1759 Madison County, Virginia⏎Facts and Events⏎
-
in the School of Political Science and Economics at Waseda University, Tokyo, Japan and Global Fellow of Peace Research Institute, Oslo (PRIO), Norway.<|endoftext|>The effect of monocular depth cues on the detection of moving objects by a moving
SteeringAdding a push along a neuron while the model writes, to see how its output changes. Turning it up or down.
Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.
Here is something I have been thinking about lately: 23+100=400=49×12 (6) 56°24' (5.7) min (59)°27k×1000 mm (
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
how it deals with death. Published by Edition Patrick Frey (Zurich).⏎250 pages, 20 x 17 cm, softcover, Edition Patrick Frey
-
the cultural and political forces behind the overvaluation—and masculinization—of wakefulness in the United States.⏎2013 | 240 pages | Cloth $55.00⏎American History / Women's Studies/Gender
-
wrote lives of Daniel Webster and James Buchanan.<|endoftext|>By: H Bernard Bechtel (Author)⏎206 pages, colour & b/w photos, b/w line drawings⏎Dr. Bechtel
-
a floral shop.⏎Even more influential of Liberia<|endoftext|>Oxford University Press⏎2017 | 200pp | £7.99⏎Organic chemistry is one of the most exciting areas
Connections
Show on mapWhich neuronsOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in earlier layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.
Fed by earlier layers
- L5:2461 weakly suppresses
- L10:252 weakly excites
- L2:2846 weakly excites
- L5:1182 weakly suppresses
- L3:2431 weakly excites
- L2:1420 weakly excites
- L9:2645 weakly excites
- L4:2876 weakly excites
- L11:1846 weakly suppresses
- L2:337 weakly suppresses
- L7:1579 weakly excites
- L0:1781 weakly excites
- L2:2067 weakly excites
- L9:554 weakly suppresses
- L11:2102 weakly excites
- L10:2299 weakly excites
- L6:2227 weakly suppresses
- L7:438 weakly excites
- L7:547 weakly suppresses
- L5:1492 weakly suppresses
- L7:935 weakly suppresses
- L10:1649 weakly excites
- L3:1100 weakly excites
- L6:1336 weakly suppresses
Feeds later layers
- L14:1435 suppresses
- L15:2016 suppresses
- L18:1546 suppresses
- L13:2125 suppresses
- L13:855 suppresses
- L16:935 suppresses
- L21:1628 suppresses
- L14:361 excites
- L14:1408 suppresses
- L14:1778 weakly suppresses
- L13:513 weakly suppresses
- L18:1750 weakly suppresses the blank slot before a year
- L13:1322 weakly suppresses
- L18:291 weakly excites
- L19:2706 weakly suppresses "per" in rate phrases
- L13:1584 weakly suppresses
- L15:1467 weakly suppresses
- L19:2792 weakly suppresses
- L17:1636 weakly excites
- L14:2108 weakly suppresses
- L15:768 weakly suppresses
- L23:791 weakly excites
- L21:848 weakly suppresses
- L23:455 weakly suppresses
Activation histogramA chart of how often a neuron takes each activation value.
How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.
Logit lensReading which output tokens a neuron pushes toward or away from, straight off its output weights, before the later layers have their say.
The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. instead.
Promotes
- elist
- 藨
- resco
- ests
- ermen
- ManagerInterface
- ablo
- esthesia
- ogóle
- ethyl
Suppresses
- span
- span
- overnight
- overn
- over
- แข
- 期限
- Dur
- 스
- dated