decimal point inside numbers
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. 22 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 responds to the dot in decimal numbers such as 0.5, 1.25 or 2.0. Its negative side fires on the word "main" in words like "mainstay". Turning it up or down gave ordinary, unrelated text, so the effect on writing is unclear.
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.
-
strands of the 9/10 grade band of the Minnesota Language Arts standards are addressed.⏎Credit: 1.0 (Semester A = 0.5 credit, Semester B = 0.5 credit.)⏎Honors Opportunity:
-
,000-lb load would use the following calculation:⏎[(4,000 lbs + 6,000 lbs) × 1.25] ÷ 4 = 3,125 lbs⏎
-
rather than just during that time period.⏎If you're walking between 3 and 4 miles per hour on a hike you should multiply your weight by 0.30 to get your net calorie burn per hour, according to Runner's World
-
connecting computers and peripherals or other devices. It is capable of supporting up to 127 daisy-chained peripheral devices simultaneously. The latest USB 2.0 specification can deliver 480Mbps data transfer bandwidth. In addition,
-
; that is, Intel had to be able to build it on the same manufacturing lines that generated the 100-MHz P5s. (0.6µm Bipolar Complementary Metal Oxide Semiconductor, or BiCMOS
-
0 (P1/P0) where P1 and P0 are two quanties that are in the same units of measure. The decibel is 0.1 bel, so the decibel value LdB is LdB = 1
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: if the entire $ \frac{ (m+1) }{ (m+2) } } $ $ \frac{ (m+2) }{ (m+3) } $ $ \int_{\frac{ (
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
the NEJM column.⏎Yet hospitals and politicians alike often tout the jobs that the health care industry creates in local communities. “The health care sector is an economic mainstay, providing stability and even growth during times of recession,” the American Hospital Association
-
Gender, Complex Political Emergencies, and International Intervention Chapter 3 Gender and the Causes and Consequences of Armed Conflict Chapter 4 Gender Mainstreaming in United Nations Peacekeeping Operations: Talking the Talk, Tripping over
-
Regulations, 1992: Main aim of this legislation is establishing tour operators' responsibility towards customers and providing remedies to
-
Gopher Prairie on the train. In the drab town are three thousand dull people, in a social strata ranging from Swede farmer to bank president. Main Street has two-story brick shops flanked by Fords and lumber wagons.
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
- L18:1758 suppresses
- L9:1854 suppresses
- L17:1552 suppresses
- L19:831 suppresses
- L18:791 weakly suppresses
- L21:763 weakly excites
- L20:1120 weakly excites
- L21:1159 weakly suppresses
- L18:1409 weakly suppresses
- L5:582 weakly excites
- L20:179 weakly excites
- L21:395 weakly excites
- L20:169 weakly suppresses
- L19:1351 weakly suppresses
- L5:1909 weakly suppresses
- L18:569 weakly excites
- L15:12 weakly excites
- L16:432 weakly suppresses period after month or initial
- L2:1369 weakly excites
- L2:75 weakly suppresses
- L1:670 weakly excites
- L20:486 weakly excites
- L18:737 weakly excites
- L20:962 weakly excites decimal points and dots in numbers
Feeds later layers
- L23:1279 excites
- L23:585 excites
- L27:435 weakly excites
- L24:256 weakly suppresses
- L26:760 weakly excites
- L24:371 weakly excites
- L23:950 weakly excites
- L26:327 weakly suppresses
- L23:2500 weakly excites
- L26:1467 weakly suppresses
- L23:2043 weakly excites
- L26:929 weakly excites
- L27:829 weakly excites
- L27:1631 weakly excites
- L25:985 weakly excites
- L24:2115 weakly excites
- L25:2159 weakly excites
- L23:2825 weakly excites
- L26:1307 weakly excites
- L23:2553 weakly suppresses
- L25:212 weakly excites
- L25:2080 weakly suppresses
- L24:1170 weakly excites
- L23:2853 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
- uong
- annon
- OMP
- aises
- ética
- 惑
- aggio
- gebung
- atters
- орт
Suppresses
- pen
- ped
- opt
- experiment
- pool
- ades
- clin
- AxisSize
- men
- esting