unclear: word pieces inside names and words
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. 26 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 reacts to a fragment in the middle of a longer name or word, like the 'of' in Sassoferrato or the 'Be' in Beethoven. Turning it up made the model's writing drift into repeated or off-topic text, so it doesn't clearly control anything we could name.
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.
-
and their visitors came from all walks of life and practiced asceticism with varying degrees of rigor. Certainly, an elite background did not preclude a monastic vocation
-
ò Perotti (1430-1480), secretary to Cardinal Bessarion and archbishop of Siponto, donated to Sassoferrato, the city of his birth, a collection of reliquaries,
-
Based on stylistic analysis, according to Martin Donnert, the mosaic icon at Sassoferrato was executed in the 14th century as a typical work of
-
symphony based on the life of the English military-imperial hero General Gordon, which, fortunately, did not get far.⏎Any composer approaching the symphony after Beethoven might well have had some trepidation about it. Brahms’ difficulties
-
Institute of Medical Sciences, Ansari Nagar, New Delhi 110029⏎|How to cite this article:|⏎Krishnan A, Behera P, Nongkyn
-
’s nocturnes, Bach’s preludes and fugues, and the works of Clementi, Mozart and Beethoven. But Elsner also recognized Chopin’s unique talents and encouraged the
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: What if life on Earth reached immensively immensive immensive immensive immensive immensive immensive immensive immensive immensive immensive immensive immensive immensive
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Soyuz 17, Salyut 6 EO-1, Salyut 7 EP-5. More...⏎Makarov Makarov, Oleg Grigoryevich (1933-20
-
the Rurik princes maintained their control over Kievan Rus and, later, Muscovy until 1598.⏎Rurik’s successor Oleg (d. 912) conquered Kiev (c. 8
-
pioneer and wealthy Chicago sportsman Logan “Jack” Vilas made the initial flight in order to demonstrate the viability of using aircraft in fire prevention. Wisconsin’s Chief Forester Edward Griffith hired him but Vilas refused pay, saying that he wanted only
-
was crowned as the first king of Jerusalem on Christmas Day, 1100, by Dagobert of Pisa, the Patriarch, who would have preferred to turn the
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
- L2:172 suppresses
- L25:652 excites
- L3:459 weakly suppresses
- L15:1274 weakly excites
- L23:2441 weakly excites
- L22:268 weakly excites
- L10:370 weakly suppresses
- L20:256 weakly excites
- L6:1357 weakly excites
- L19:2045 weakly excites abbreviation pieces in organisation names
- L16:17 weakly excites
- L20:211 weakly excites
- L0:2987 weakly excites
- L22:1533 weakly suppresses
- L4:55 weakly excites
- L20:2371 weakly suppresses
- L15:1986 weakly excites
- L19:842 weakly suppresses
- L14:832 weakly suppresses modifier words before their head word
- L2:2387 weakly suppresses
- L7:1532 weakly suppresses
- L16:2096 weakly excites
- L23:2744 weakly excites
- L18:328 weakly suppresses
Feeds later layers
- L27:1663 excites
- L27:2671 weakly excites
- L27:2538 weakly excites
- L27:3047 weakly suppresses
- L27:1360 weakly excites
- L27:979 weakly excites
- L27:577 weakly excites
- L27:2059 weakly excites
- L27:903 weakly suppresses
- L27:2748 weakly suppresses
- L27:2788 weakly excites
- L27:496 weakly excites
- L27:1372 weakly excites
- L27:2079 weakly suppresses
- L27:2983 weakly excites
- L27:1441 weakly excites
- L27:1164 weakly excites
- L27:2763 weakly excites
- L27:2790 weakly suppresses
- L27:79 weakly excites
- L27:1461 weakly suppresses
- L27:1857 weakly excites
- L27:1385 weakly excites
- L27:3034 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
- str
- Str
- str
- _str
- Str
- →str
- -str
- STR
- .Str
- .str
Suppresses
- цу
- kapı
- gars
- \Modules
- /dir
- london
- lifting
- wood
- think
- Nz