'whole' in 'a whole lot'
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. 3 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 word 'whole' when it comes after 'a', as in 'a whole lot' or 'a whole host of'. It is the strongest on that word and quiet elsewhere. Turning it up did not visibly shift the replies.
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.
-
is a lot of work. She commented that sometimes memorization work makes a poem "fall apart" and lose meaning or beauty.⏎If you love the poem a whole lot, spending more time and attention on it through memorization just creates additional lo
-
how many petals then had fallen?<|endoftext|>At EdPlace we are passionate about planning and preparation to avoid undue stress to students, whatever their age.⏎There are a whole host of reasons why a 6-week learning plan may be useful to your child
-
to time during the semester.<|endoftext|>June 1st 1584⏎Perth's HammermenThe Hammermen’s Incorporation of Perth embraced a whole collection of crafts including silversmiths, goldsmiths, clocksmiths
-
isn't going to prematurely age your brain, but regularly consuming added sugar in large quantities may contribute to complications with your brain and cognitive health (not to mention a whole host of other health issues).⏎Read on to learn more about some of the drinks
-
generations, DNA changes at a presumably steady rate. For birds, that's about 2% change (for mitochondrial DNA) in a million years -- not a whole lot! But it's enough to figure out their evolutionary history. What have we
-
cancer.⏎Water is the essence of life, but not just human life. Almost every living thing on the planet must have water to survive, and that includes a whole host of bacteria and microorganisms that can wreak havoc on human health.
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 more I have been able to get to 20% from my savings. So, this is a common indicator of how much I have been saving, and the more I have been able to get to 20% from my
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
health care system.⏎Other species, too, suffer as a result of the way we eat. Industrial food production pollutes rivers and streams, creating massive offshore dead zones. It kills off bee colonies, consigns millions of animals to inhum
-
in the 1980s. The scientists used computer modeling for their prediction and will take shipboard measurements in July throughout the gulf. The dead zone is caused by an abundance of nutrients from sources
-
felsic composition, and is typically black in color but may also be rust-red or display whiteish ‘snowflake’ inclusions. It often displays conchoidal fracturing with very sharp edges. Obsidian is made from the rapid cooling
-
-red, dark green, or iridescent. Inclusions of white ‘snowflakes’ are common. Obsidian is very smooth, with scoop-shaped conchoidal fractures that often result in sharp edges when broken.⏎Pro Tip: I have
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
- L1:951 weakly excites
- L1:2483 weakly excites
- L0:2129 weakly suppresses
- L0:4 weakly suppresses
- L2:2909 weakly excites
- L0:571 weakly excites
- L0:976 weakly excites
- L0:2583 weakly suppresses
- L2:2001 weakly excites
- L1:2967 weakly suppresses
- L1:2724 weakly excites
- L0:2230 weakly excites
- L2:1321 weakly suppresses
- L1:718 weakly suppresses
- L1:1799 weakly suppresses
- L0:1316 weakly suppresses
- L1:1293 weakly excites
- L0:1 weakly suppresses
- L0:178 weakly suppresses
- L2:235 weakly suppresses
- L0:814 weakly excites
- L1:2559 weakly excites
- L2:2532 weakly suppresses
- L0:1826 weakly suppresses
Feeds later layers
- L4:1369 weakly excites
- L24:2959 weakly excites
- L9:2811 weakly excites
- L23:566 weakly suppresses
- L20:909 weakly suppresses
- L7:2155 weakly excites
- L10:1323 weakly suppresses
- L6:1096 weakly suppresses
- L9:2408 weakly excites
- L5:1852 weakly suppresses
- L23:1507 weakly excites
- L18:2803 weakly excites
- L23:1685 weakly suppresses
- L8:2646 weakly suppresses
- L21:497 weakly suppresses
- L11:2884 weakly excites
- L11:1007 weakly excites
- L26:3018 weakly excites
- L15:2255 weakly suppresses
- L5:2903 weakly excites
- L17:362 weakly suppresses
- L14:2117 weakly suppresses
- L14:933 weakly excites
- L8:785 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
- 废话
- 不由
- crap
- Fortune
- 命周期
- ensive
- orable
- ertificate
- lime
- 盆地
Suppresses
- func
- $log
- oster
- odor
- ador
- SB
- diá
- (symbol
- ued
- stellen