'in' after a size measurement
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. 11 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 'in' when it comes right after a size or range, as in 'range in color', '4 inches in diameter' or '3 metres in height'. Those snippets are about how big or wide something is. Turning it up gave text that did not clearly follow this pattern, so the link to the snippets is only partly confirmed.
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.
-
from boulders extracted from riverbeds.⏎The cylindrical shaped jars have a lip rim to support a lid, and range from one to more than three metres in height, weighing up to 14 tons. Very few examples of stone lids
-
the sweetest taste. Thick slices work well for sandwiches or caprese. Some varieties can produce fruit weighing up to several pounds.⏎Oxheart tomatoes can range in from small to as big as beefsteaks, and are shaped like large heart
-
28% protein. They are also an excellent supply of vitamins A, B, C, and E.⏎Sea vegetables come in many varieties and can range in color from red, blue-green, brown, and green.⏎- Kombu
-
demospongiae, which live in fresh water.⏎A wide variety of sponges exist, ranging in size from 1 inch to 4 feet in diameter. Inside a sponge there is a system of openings and passages called a canal
-
panel is a formed sheet of molded concrete, usually 2 feet to 12 feet (0.61 m to 3.66 m) in width, and 8 feet to 12 feet (2.44
-
drooping, and on this part, the 9-12 twelve flowers are produced. Flowers are splendid and of large size, being nearly 4 inches in diameter when fully expanded. In color the flowers 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: I need to get this problem in the book 10000000000012323000101117112411271127
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
case of a number, you can use the following formulas: 1350pc, x, or %. Then, multiply the result by 100 to get the percent. This method is also useful for finding the exact value of
-
blue and red on this chart denote republican and democratic rule respectively and form the background for some salient information on the US economy. The grey line displays the annual average unemployment rate which has been seasonally adjusted. The black line shows historical GDP percentage
-
| Volume 25, Issue 1⏎That was the end of the campaigning in 177
-
in 2 patients (5%). The mean age of patients was 29 years. 2 patients (5%) reported
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
- L9:2164 suppresses
- L6:1318 suppresses
- L8:2201 excites
- L5:1247 weakly excites
- L9:1317 weakly suppresses
- L1:2070 weakly suppresses
- L6:398 weakly excites
- L3:2106 weakly excites
- L0:2496 weakly excites
- L4:456 weakly suppresses
- L1:2153 weakly suppresses
- L10:1839 weakly excites
- L4:1954 weakly excites
- L10:299 weakly excites
- L6:2252 weakly excites
- L0:2367 weakly excites
- L7:2705 weakly suppresses
- L5:2446 weakly excites
- L5:2163 weakly suppresses
- L9:707 weakly suppresses
- L8:586 weakly excites
- L7:121 weakly excites
- L3:1458 weakly excites
- L10:1712 weakly suppresses
Feeds later layers
- L13:901 excites
- L17:1979 excites
- L17:305 excites
- L18:557 excites
- L12:1925 suppresses
- L13:1335 excites 'of' inside fixed noun phrases
- L18:1358 excites
- L12:1518 suppresses
- L19:1361 excites
- L19:1176 suppresses
- L16:940 excites
- L20:803 excites
- L20:712 excites
- L17:546 excites
- L20:1284 excites
- L12:1496 excites
- L14:1393 excites
- L18:56 suppresses
- L15:1310 excites
- L13:938 excites
- L14:1737 excites
- L13:843 excites
- L22:507 excites
- L20:904 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
- volume
- grace
- height
- functionality
- width
- appearances
- size
- 容量
- direction
- 宽度
Suppresses
- ionale
- autiful
- miền
- swana
- 私立
- Paleo
- ivist
- chemes
- chooser
- enment