end of document boundary tokens
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. 13 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 mostly lights up on the marker that signals the end of one document and the start of an unrelated one, so it seems to notice where a new piece of text begins. Turning it up made the model's writing wander into unrelated topics, with no clear theme.
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.
-
code<|endoftext|>Jet engines, or gas turbines, suck in air at the front with a fan
-
are found, there is no way<|endoftext|>We won’t stop until we’ve eradicated HIV/AIDS⏎The Drug Development Process
-
can we say we believe in it? It is up to me and you.”⏎Adafruit publishes a wide range of writing and video content, including interviews and reporting on the maker market and the wider technology world. Our standards page is intended as a
-
groups<|endoftext|>What happens when you mix vinegar and baking powder?⏎When baking soda is mixed with
-
| Volume 25, Issue 1⏎That was the end of the campaigning in 1776. Washington went
-
the Students from Student table whose marks is more than Average of all marks.⏎(Select Avg(M<|endoftext|>Euler's Equation and the Physical Reality. =. a) Euler's Equation
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: I think it is possible to find a set of points, the problem is that it's hard to me. i think it's not hard, but i can't think it's not. it's just think it's not... i can
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
% off) $17.25 new (22% off) $21.95 direct from Amazon Amazon page|⏎Setup an account with your affiliations in order to access resources via your University's proxy server⏎Configure custom
-
conjugate, write the correct conjugate acid or base for the species in the pair.⏎a. H2O, OH-⏎Frequently Asked Questions⏎What scientific concept do you need to know in order to solve this
-
quality lesson planning resources, fast!⏎Share & remix collections to collaborate.⏎Organize your curriculum with collections. Easy!⏎Have time to be more creative & energetic with your students!⏎In this verb tense review worksheet, students are given sentences and must find
-
collections to collaborate.⏎Organize your curriculum with collections. Easy!⏎Have time to be more creative & energetic with your students!⏎Students explore the Orange Revolution. In this Orange Revolution lesson, students examine the climate of the Ukrainian protest and investigate the role
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
- L5:203 suppresses
- L3:70 suppresses
- L10:816 excites
- L5:2007 suppresses
- L10:1580 suppresses
- L6:2030 suppresses
- L10:84 suppresses
- L1:816 weakly excites
- L9:1688 weakly excites
- L10:632 weakly excites
- L3:1257 weakly excites
- L7:142 weakly excites
- L12:1186 weakly suppresses
- L2:1031 weakly excites
- L8:32 weakly excites
- L5:619 weakly excites
- L5:238 weakly excites
- L6:520 weakly excites
- L11:1629 weakly excites
- L0:1650 weakly excites
- L6:216 weakly excites
- L8:637 weakly excites
- L5:2738 weakly suppresses
- L1:179 weakly suppresses
Feeds later layers
- L19:1479 excites
- L15:885 excites
- L18:2202 excites
- L14:886 excites
- L17:1953 excites
- L19:441 excites
- L15:344 excites
- L15:2717 suppresses
- L14:1274 excites
- L20:1081 suppresses
- L20:2120 suppresses "so" opening a purpose clause
- L19:2232 weakly excites end-of-document boundary token
- L19:624 weakly excites
- L17:647 weakly suppresses
- L17:2481 weakly suppresses
- L25:630 weakly suppresses
- L22:2221 weakly suppresses
- L14:901 weakly excites mid-word fragments in technical terms
- L26:2340 weakly suppresses
- L26:2358 weakly suppresses
- L14:56 weakly excites document ends and list line breaks
- L19:1083 weakly suppresses
- L16:49 weakly suppresses
- L23:2124 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
- alm
- /*⏎⏎
- dirig
- [keys
- görm
- icultural
- mitter
- -root
- mart
- 朵
Suppresses
- ughter
- `]
- "], ⏎
- :],
- §Ã
- sẵn
- もなく
- -dess
- "]],⏎
- "],