end of document token
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. 15 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 fires on the invisible marker that separates one web page from the next in the training text. When it was turned up, the model's output did not show a clear change in theme. It also goes quiet or negative on numbers inside lists or counting sequences.
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. 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.
-
countless texts in the canon of English literature. The subject occupies a prominent position in a diverse array of genres and forms, among which is that of the modern...<|endoftext|>The Hundred Years' War began in the reign of Edward
-
.⏎You can also purchase the Almanac Monthly at the iTunes Store !<|endoftext|>Learning in dissolving boundaries⏎Talking about changes inherent in the move to what the
-
Virginia.⏎To summarize, then, the Civil War Center is committed to the following:⏎WHY KENNESAW STATE UNIVERSITY IS AN⏎<|endoftext|>[Last Updated: April 2nd, 2015]⏎Con
-
the globe.⏎The definitions of ‘buddy’ and ‘bro’ in the OED have recently been revised. We explore their history and increase in popularity.<|endoftext|>Native Plants for the Mid-Atlantic⏎On this page⏎- What Are Native
-
. Please see the bottom of the page for more information or visit our Terms and Conditions.⏎Thanks for your feedback.⏎37 of 42 found this helpful<|endoftext|>Television was hardly a reliable broadcast medium
-
of all bourgeois states, France and Britain prominent among them.⏎What made you want to look up "Constantinople Agreement"? Please share what surprised you most...<|endoftext|>Your healthcare provider will ask about your medical and family history and your overall health.
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 have a very simple but actually very important and simple and important question that is actually very simple but in which I can also take a point of view that I can also have an important and important question that I have to ask of the other
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
center activities!⏎- Engaging morning work!⏎- Homework variety!⏎- Individual pages for facts 1-12 (-1, -2, -3, etc. up to -12)⏎- Mixed facts practice pages⏎-
-
index i and decrementing the value of i by 1.⏎If you run this program, it will print the below output:⏎5 4 3 2 1⏎We need to use the range function to iterate over the array using
-
5, 27, 28, 31, 32, 33, 34, 35, 36, 37, 39, 49, 51
-
a matrix⏎Find nth maximum⏎Matlab Basics II - Create a vector with a repeated entry⏎Create sequnce 1 4 9 16 25.........⏎Basics: Divide integers to get integer outputs in all
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
- L13:2412 weakly excites
- L7:2830 weakly suppresses
- L13:15 weakly suppresses
- L13:1281 weakly suppresses
- L13:1757 weakly suppresses
- L13:1139 weakly suppresses
- L5:1128 weakly excites
- L13:996 weakly excites
- L14:1627 weakly suppresses
- L14:692 weakly excites word right before a section number
- L14:2807 weakly excites
- L13:2773 weakly excites
- L9:1351 weakly suppresses
- L2:333 weakly excites
- L14:1 weakly excites
- L14:1952 weakly excites
- L12:1713 weakly suppresses
- L1:1112 weakly suppresses
- L8:1484 weakly excites
- L14:8 weakly suppresses
- L3:1922 weakly excites
- L11:1764 weakly suppresses
- L10:2410 weakly excites
- L1:2173 weakly excites
Feeds later layers
- L17:386 weakly excites
- L17:2509 weakly excites
- L17:1992 weakly excites
- L19:1887 weakly excites
- L22:110 weakly excites
- L16:1210 weakly excites
- L22:2702 weakly suppresses
- L17:1092 weakly excites
- L22:263 weakly excites
- L18:1476 weakly suppresses
- L24:226 weakly excites
- L19:2524 weakly suppresses
- L16:2699 weakly suppresses
- L17:436 weakly excites
- L25:1634 weakly suppresses
- L20:2151 weakly suppresses
- L19:2500 weakly suppresses
- L16:2446 weakly suppresses
- L20:2002 weakly suppresses
- L21:38 weakly excites
- L25:470 weakly suppresses
- L22:2614 weakly suppresses
- L24:741 weakly suppresses
- L22:1264 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
- istribution
- ourage
- 슬
- edata
- ugin
- tings
- lename
- endency
- ạn
- ictions
Suppresses
- 正
- BO
- 待
- ROS
- either
- 即时
- 此刻
- head
- moment
- =