sentence end before a line break
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 fires when a sentence finishes and a new line or paragraph is about to start, especially after a full stop, exclamation mark or line break. Turning it up did not give a clear theme in the model's writing.
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.
-
the pumping and processing of natural gas, but there’s now a new means for preventing such, and it’s a lot more friendly to the environment and company budgets.⏎That’s right, Siftek. It’s the new name in natural gas
-
to alleviate stress and prioritize tasks that are manageable.⏎Feeling overwhelmed can occur at any point in life, but for many it’s always affected their studies or work⏎Suppose you regularly feel overwhelmed by the noise and hustle and bustle of crowded
-
verter is the analog spindle, and the spindle driven by the servo motor is the servo spindle. Both can meet the control requirements of CNC machine tool spindles.⏎With the development of technology in the field of
-
> Another example is to imagine a Tree in which the leaves contain⏎> "objects." If I want to change a property of one of those leaf objects,⏎You can't in general, the thing with a different property is a different⏎>
-
are too complex for a traditional factory to handle. In time, these amazing machines may be able to make almost anything, anywhere—from your garage to an African village.⏎The applications of 3D printing are especially
-
and 2).⏎Music National Curriculum for KS1⏎Key stage 1⏎Pupils should be taught to:⏎- use their voices expressively and creatively by singing songs and speaking chants and rhymes⏎- play tuned and untuned instruments
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 been in the car for 10 hours, and I have never seen any of the people who have driven on to the plane to take off to land in the airport in the morning for the first 30 minutes that the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
: Mining World⏎Population: 300 (Custodial staff)⏎Geography/Demography: Caligine II is a small moon orbiting the Gas Giant Caligine
-
of the month of Nisan, which coincides with a full moon.⏎Question: Is Passover the same time/date as Easter and are presents given?⏎Answer: Passover does not necessarily fall on the same date as Easter, though it usually
-
.⏎No direct anagrams for endoplasm found in this word list.⏎Words formed by adding one letter before or after endoplasm (in bold), or to adelmnops in any order:⏎s - endoplasms⏎All words
-
this area which could be gained from excavation of<|endoftext|>In which Scrabble dictionary does ZEATINS exist?⏎Definitions of ZEATINS in dictionaries:⏎- noun -⏎a chemical compound found in maize⏎There are 7 letters in ZE
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
- L11:1527 excites
- L11:2971 excites
- L8:2520 suppresses
- L9:1743 excites
- L9:1949 suppresses
- L8:1848 excites
- L11:1619 suppresses
- L10:1169 weakly suppresses
- L10:471 weakly excites
- L11:2451 weakly suppresses
- L7:2303 weakly excites
- L9:257 weakly suppresses
- L8:579 weakly excites
- L10:280 weakly suppresses
- L10:1523 weakly excites
- L1:2067 weakly suppresses
- L5:560 weakly suppresses
- L0:2266 weakly excites
- L11:2928 weakly excites
- L11:1344 weakly suppresses
- L6:257 weakly excites
- L2:2997 weakly suppresses
- L11:1965 weakly excites
- L10:963 weakly suppresses
Feeds later layers
- L15:1785 excites
- L14:385 excites
- L14:1092 weakly excites
- L14:1648 weakly suppresses
- L14:209 weakly excites
- L13:1238 weakly excites
- L16:2969 weakly excites
- L15:1619 weakly excites
- L14:1178 weakly excites
- L15:502 weakly excites
- L19:1108 weakly excites 'your' or 'our' before a noun
- L18:3071 weakly suppresses
- L14:134 weakly excites spot right before a citation year
- L25:2335 weakly suppresses
- L17:2154 weakly excites
- L18:279 weakly suppresses
- L18:556 weakly excites
- L21:2213 weakly suppresses
- L17:1712 weakly excites
- L21:733 weakly suppresses
- L13:1086 weakly excites
- L25:2425 weakly suppresses
- L14:1038 weakly excites
- L15:155 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
- 們
- င
- 물을
- ключа
- 位
- 身
- (sys
- [position
- FormatException
- údo
Suppresses
- licas
- 妇女
- cctor
- getEmail
- 责
- ponder
- rika
- @endforeach
- trasound
- getKey