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. 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 strongly to the invisible marker that ends a web page and comes right before a new, unrelated page. It goes quiet on the words 'granted' and 'possible' in phrases like 'take for granted'. Turning it up gave off-topic text.
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.
-
to Simplify Expressions⏎A straight forward approach to simplifying expressions⏎Integer Practice Problems - Addition⏎Review and Practice Problems for adding positive and negative numbers⏎<|endoftext|>In physics, a conservation law states that a particular measurable property of an isolated physical
-
VS is a promising alternative strategy to produce the functional and thermostable reverse transcriptase.⏎All Science Journal Classification (ASJC) codes⏎- Insect Science<|endoftext|>An account of the religion of
-
994- Drexel University. All rights reserved.⏎The Math Forum is a research and educational enterprise of the Drexel University School of Education.<|endoftext|>Maybe you’ve tried intermittent fasting (IF) to shed a few (pand
-
this word list.⏎List shorter words within focus, sorted by length⏎All words formed from focus by changing one letter⏎Browse words starting with focus by next letter<|endoftext|>I was listening to Brooke Gladstone's ON THE MEDIA podcast about the difficulty of
-
liu.<|endoftext|>Bond Price Calculator⏎With this bond price calculator, we aim to help you calculate
-
You can submit your school, college or university level homework or assignment to us and we will make sure that you get the answers related to classification of relays.<|endoftext|>Harvest Strike Call--Pickets gather at the edge of a Central California grape
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 am so glad that we have such a good time. I mean, so much better than I had. This is so precious. I can't think of anything so great. I am so lucky. I have access to all the good
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
and screen for a host of maladies thanks to sophisticated imaging technology. X-rays are now so frequently used and so effective that most of us take their capabilities for granted—but it was not so long ago a tumor or damaged bone couldn’t be found
-
into a true and genuine spiritual experience.⏎If you’re reading this article, it’s thanks to the generosity of people like you, who have made Aleteia possible.⏎Here<|endoftext|>The
-
remarkable feats of science have been accomplished. What people fifty years ago would have sworn would always be the product of Science Fiction are now everyday realities that we take for granted. Products like our smartphones,
-
1, 47, and 48. Here again praise moves into a yet higher sphere, and consists of an expression of thankfulness to God for what He is in Himself in mercy.⏎These movements indicate a growth of experience, center
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
- L10:2464 excites
- L11:1909 excites
- L10:507 suppresses
- L4:833 weakly suppresses
- L8:2870 weakly excites
- L10:1946 weakly suppresses
- L9:1507 weakly excites
- L9:1735 weakly suppresses
- L11:1933 weakly suppresses
- L2:2328 weakly suppresses
- L4:21 weakly excites
- L3:272 weakly excites
- L1:2278 weakly excites
- L0:2661 weakly excites
- L5:2875 weakly suppresses
- L6:1379 weakly excites
- L3:115 weakly suppresses
- L3:1703 weakly excites
- L11:1924 weakly suppresses
- L2:122 weakly excites
- L9:835 weakly suppresses
- L9:2068 weakly suppresses
- L8:777 weakly suppresses
- L11:5 weakly suppresses
Feeds later layers
- L13:1719 suppresses
- L18:1843 weakly suppresses
- L18:2252 weakly excites
- L15:2426 weakly suppresses
- L13:2205 weakly suppresses
- L13:2688 weakly suppresses
- L14:1669 weakly suppresses
- L16:2920 weakly suppresses
- L26:2963 weakly excites
- L13:1817 weakly suppresses
- L17:1671 weakly excites
- L19:613 weakly excites
- L21:2074 weakly suppresses the word 'later' after a time span
- L14:771 weakly suppresses
- L26:9 weakly excites
- L18:2800 weakly excites
- L22:740 weakly excites
- L16:1622 weakly suppresses
- L19:1297 weakly excites
- L24:447 weakly excites square bracket opening an inserted note
- L23:1201 weakly excites
- L24:2242 weakly excites
- L23:2048 weakly excites
- L24:1003 weakly excites
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
- nings
- heads
- [opt
- ickle
- +self
- (lines
- ningen
- $_[
- чество
- ].[
Suppresses
- 还好
- 暴力
- 富贵
- 幸好
- privilege
- 高楼
- ilege
- epam
- 感谢
- 运行