space inside dates before the year
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. 0 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 lights up on the blank space in dates and timestamps, just after the day number and before the year or time, as in 'Feb 23 2006'. Turning it up made the model produce number-heavy, unrelated 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.
-
intering include:⏎- Lubrication removal⏎- Oxide reduction⏎-<|endoftext|>posted on Feb, 23 2006 @ 12:29 AM⏎There are two
-
that declares a static inner class called Wheel. If we consider the real-world, we can find many objects around us,<|endoftext|>18th Jan 12 14:01<< Go back to news⏎The kinetic force of braking
-
a common enemy.⏎The effects of this split are not unlike those from the one that occurred in the<|endoftext|>Green Infrastructure for Clean Water Act⏎January 22 2010 | Know Your H20,⏎The Green Infrastructure for Clean
-
2030 as world population grows and demand for power rises, according to the 2030 water resources group:⏎Phone: +32 2 213 18 33⏎EWEA -
-
⏎The Edcamp Model 2014 US$ 11.95 73 pages⏎The Adult Student's Guide to Survival & Success 20
-
Hildner / University of Bremen⏎The main crack<|endoftext|>analysis of algoritms⏎robert.kern at gmail.com⏎Fri Sep 10 00:10:22 CEST 2010⏎
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 thinking about how much I value the people in my life. I have been thinking about the importance of relationships and how I can be a good and kind of person in the world. I have been thinking about how much I have
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Tucker M.R., Himmelbach A., Stein N., Mascher M., Nagel K.A., De Gara L., Simmonds<|endoftext|>Last Census - 2005⏎Cameroon is a culturally diverse costal country in Africa
-
40% lean protein, 20% fat and 40%
-
1:x in x.⏎For non-strict semantics it would be ok to store
-
3. Moving into Key Stage 2 can be a big change for both children
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
None: this is the first layer.
Feeds later layers
- L1:1561 suppresses
- L9:1706 weakly suppresses
- L22:2631 weakly suppresses
- L8:749 weakly suppresses
- L13:966 weakly suppresses
- L1:2196 weakly suppresses
- L6:65 weakly excites
- L22:407 weakly excites
- L6:397 weakly suppresses
- L15:745 weakly excites
- L18:1369 weakly excites
- L18:58 weakly suppresses
- L13:452 weakly excites
- L10:95 weakly excites
- L11:1201 weakly excites
- L1:1867 weakly suppresses
- L22:1906 weakly suppresses
- L18:1761 weakly suppresses
- L25:1150 weakly excites
- L19:832 weakly excites
- L14:207 weakly excites
- L3:537 weakly excites
- L7:1096 weakly excites
- L6:736 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
- utra
- [tag
- 習
- читыва
- arine
- начен
- missions
- iteur
- .vn
- 过大
Suppresses
- 年第
- iltr
- SK
- tones
- ylation
- pregn
- Garn
- SY
- -eff
- 久久