period inside abbreviations like i.e. and e.g
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. 1 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 dot in shorthand like "i.e." or "e.g.", the dot that sits in the middle of an abbreviation rather than ending a sentence. It went negative on dots that close off a sentence. Turning it up made the model's text drift into unrelated topics, so the steering results don't show a clear effect.
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.
-
Gemeente, 1883-1884: v. I, Nos. 1571-1581; v. II, 2, Nos. 703-739.⏎
-
Have a snack supply for the day; e.g. mandarins, apples, bananas, low fat yoghurts, dried
-
to the small population numbers limiting the number of mutational possibilities; and the much larger genome, which cannot sustain a similar mutation rate without error catastrophe; i.e. extinction; and sexual reproduction means that there is 50%
-
et al. 2020). Under these circumstances, tree´s survival primarily depends on the extent to which tree function is impaired by drought (i.e. resistance, Lloret et al., 2011). To increase
-
injection molding processes, as well as post-production and finishing procedures, to predict how critical process parameters (CPPs), e.g. melt temperature or injection velocity, would affect real time product quality, durability and reliability
-
zach-Donau region. Technically, the metadata database is a graph database. This technology aims at semantically linking data to each other, i.e. meaningfully. In concrete terms, this means that information is linked in a triple
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 lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
instills a sense of responsibility and⏎achievement, while furthering a child's desire to learn. This interactive⏎activity is also great for children of all ages.⏎The benefits of
-
high resistance to rust,and high fabrication ability.Cost-effective and versatile material,304 sheet is suitable for most applications where durable support and corrosion resistance are required.Stainless steel sheet metal and plate also provides excellent weight<|endoftext|>US Pharm. 2
-
as the corpus), and⏎the actual practice (conducted in the arena).⏎These two continually learn from each other and evolve.⏎The beginnings of the science⏎
-
the majority of the archeological material is within the top ten⏎centimeters of the soil. The damage from the proposed construction should⏎be considered mitigated.⏎Site 48TE414 is in the South Jenny Lake development.
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
- L0:1224 suppresses
- L0:1301 suppresses
- L0:178 suppresses
- L0:348 weakly excites space token right after a period
- L0:36 weakly excites
- L0:51 weakly excites
- L0:0 weakly excites
- L0:447 weakly suppresses
- L0:799 weakly suppresses
- L0:1001 weakly suppresses
- L0:2113 weakly suppresses
- L0:2838 weakly suppresses
- L0:835 weakly excites
- L0:212 weakly suppresses
- L0:2137 weakly suppresses
- L0:1567 weakly excites
- L0:2830 weakly excites
- L0:560 weakly excites
- L0:2257 weakly suppresses
- L0:990 weakly excites
- L0:749 weakly suppresses
- L0:2376 weakly excites
- L0:1186 weakly excites
- L0:2791 weakly suppresses
Feeds later layers
- L17:3027 weakly excites
- L26:380 weakly excites
- L2:694 weakly excites
- L22:742 weakly excites
- L21:524 weakly suppresses
- L19:328 weakly excites
- L3:259 weakly suppresses
- L10:213 weakly suppresses
- L3:2930 weakly excites
- L17:949 weakly suppresses
- L18:1085 weakly suppresses full stops ending sentences
- L15:1998 weakly excites
- L8:740 weakly excites
- L11:841 weakly excites
- L6:62 weakly suppresses
- L8:2128 weakly suppresses
- L19:2680 weakly excites
- L22:481 weakly excites
- L6:59 weakly suppresses
- L19:630 weakly excites
- L22:1596 weakly excites
- L4:1470 weakly excites
- L15:1334 weakly suppresses
- L3:2785 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
- 瓢
- openh
- 叨
- )':
- 氡
- 醋
- 饮
- 一杯
- מין
- Uran
Suppresses
- ctors
- stances
- itives
- itive
- cư
- leine
- /lgpl
- cba
- sumer
- successes