the '1' of numbers like 10
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. 14 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 digit 1 when it starts the number 10 (or 16, 100) in a numbered list. Turning it up made the text drift and repeat, without a clear connection to counting.
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.
-
5 Radio⏎- 6 Troubleshooting⏎- 7 Tapping the Wire⏎- 8 Scripting⏎- 9 Tech Support⏎- 10 Bridge Comm. Console⏎- 11 Supplementary Video⏎- The most
-
- 7 Are there still Taoists?⏎- 8 Is Taoism a Buddhist?⏎- 9 What are the major beliefs of Taoism?⏎- 10 Why is Taoism the best religion?⏎- 11 What does Tao
-
octave" model) on 25, 31.5, 40, 50, 63, 80, 100, 125, 160, 20
-
|⏎|7 m||7000 mm|⏎|8 m||8000 mm|⏎|9 m||9000 mm|⏎|10 m||10000 mm|⏎|100 m||
-
; to debate or discuss; as, I reasoned the matter with my friend.⏎9. (v. t.) To support with reasons, as a request.⏎10. (v. t.) To persuade by
-
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
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: the and in the first and on the and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
hardware was often referred to as the 16-bit era, and the previous generation as the 8-bit.⏎About Gaming Consoles⏎1 reply to this topic⏎0 user(s) are
-
him from a traffic light post in the streets of the capital.⏎The contents of this article is licensed from www.wikipedia.org under the GNU Free Documentation License. Click here to see the transparent copy and copyright details<|endoftext|>What is Celiac / Coeliac
-
a different version of it.⏎Volume (Year): 42 (2011)⏎Issue (Month): 2 (June)⏎|Contact details of provider:|| Web page: http://www.tandfonline.com/VECE
-
reference errors to , or , if you are the registered author of the cited work, log in to your RePEc Author Service profile, click on "citations" and make appropriate adjustments.:⏎- Andrew Shepherd, 2001
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
- L1:1281 weakly excites
- L1:394 weakly suppresses
- L13:2631 weakly suppresses
- L12:409 weakly suppresses
- L11:1574 weakly suppresses
- L11:29 weakly excites
- L2:591 weakly suppresses
- L5:1244 weakly suppresses award and prize names
- L1:1973 weakly suppresses
- L13:804 weakly suppresses
- L1:2696 weakly suppresses
- L10:46 weakly suppresses
- L6:459 weakly excites
- L10:282 weakly suppresses
- L8:1462 weakly excites
- L13:36 weakly excites
- L5:2718 weakly suppresses
- L13:1602 weakly excites
- L7:22 weakly suppresses
- L12:737 weakly suppresses
- L2:3006 weakly suppresses
- L9:253 weakly excites
- L2:2079 weakly excites
- L0:74 weakly excites
Feeds later layers
- L26:16 excites
- L16:2189 suppresses
- L15:2168 excites
- L18:593 suppresses
- L18:949 suppresses
- L21:1150 suppresses
- L26:288 suppresses
- L27:136 weakly suppresses
- L16:1868 weakly suppresses
- L15:2670 weakly excites
- L18:1148 weakly excites
- L20:561 weakly suppresses end of document marker
- L23:194 weakly excites
- L24:303 weakly suppresses
- L23:1658 weakly suppresses
- L27:445 weakly suppresses
- L20:1881 weakly suppresses
- L15:2070 weakly suppresses
- L16:313 weakly suppresses
- L27:2111 weakly excites
- L20:166 weakly suppresses
- L16:2735 weakly excites
- L16:1726 weakly excites
- L19:1069 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
- ensor
- icina
- 違って
- uzzer
- edException
- ampions
- meille
- defends
- >Select
- gether
Suppresses
- !⏎⏎
- #⏎⏎
- ++⏎⏎
- ⏎⏎
- .⏎⏎
- ',⏎⏎
- ×⏎⏎
- hab
- 。⏎⏎
- ')⏎⏎