empty table cell separator bars
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. 9 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 vertical bar symbols that start or separate empty cells in wiki-style tables, the spot where a new table cell is about to begin. Turning it up mostly pushed the text into lists of numbers, which fits the table style, but turning it down did not give a clear theme.
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.
-
to remove this template message)|⏎The Central Business District as seen from the top of the Carlton Centre⏎|Municipality||City of Johannesburg|⏎|• Total||1.09 km2 (0.42 sq mi)|⏎|
-
||Announced||May 4, 1990|⏎|-||Restored||August 21, 1991|⏎|-||Total||64,589 km2 (124th)⏎2
-
, the following chart applies.⏎Fall 2001 Jewish Holiday Calendar⏎|Biblical Name||Modern Holiday||Hebrew Date||English Date|⏎|Feast of Trumpets (Lev. 23:23–
-
200 m||200000 mm|⏎|Numbers||Mm in m||M in mm|⏎Some another meter to mm
-
2:15|⏎|Last Modified:||11 May 2011 04:
-
ivut||678 - 687|⏎|Julkaisu||JOURNAL OF LANGUAGE TEACHING AND RESEARCH|⏎
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 know that the concept of a "Math is Beautiful" at the moment in the website's location, and that's what I was talking about. Let me know if you need more information. The question is related to the mathematical and
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, and Interactions⏎|Herb||Why People May Use This Herb||Examples of Side Effects||<|endoftext|>In the statements, It rains; it is cold
-
Succeeded by||Daniel François Malan|⏎3 September 1919 – 30 June 1924⏎|Preceded by||Louis Botha|⏎|Succeeded by||James Barry Munnik Hertz
-
reports, he continued "wielding his machete until the foe had<|endoftext|>|Theme||Visible||Selectable||Appearance||Zoom Range (now: 0)|⏎Well-to-well tracer tests contribute
-
Plant Operator⏎Tasks | Tools & Technology | Knowledge | Skills | Abilities | Work Activities | Detailed Work Activities | Work Context | Job Zone | Education | Interests | Work Styles | Work Values | Wages & Employment | Job Openings⏎-
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
- L7:335 excites
- L6:1691 suppresses
- L8:1384 suppresses
- L6:1056 weakly suppresses
- L5:809 weakly suppresses
- L6:1841 weakly suppresses
- L8:1394 weakly excites
- L6:2156 weakly excites
- L1:871 weakly excites
- L7:62 weakly excites
- L6:234 weakly suppresses
- L6:2650 weakly suppresses
- L5:1311 weakly excites
- L8:18 weakly suppresses
- L5:222 weakly suppresses
- L6:1081 weakly excites
- L7:2073 weakly suppresses
- L5:258 weakly excites
- L8:1023 weakly excites
- L5:1906 weakly excites
- L2:2952 weakly excites
- L7:77 weakly suppresses
- L4:947 weakly excites
- L2:493 weakly suppresses
Feeds later layers
- L11:1496 excites
- L12:1387 excites
- L18:1621 excites
- L11:2508 excites
- L12:2336 excites
- L13:2294 weakly excites
- L11:280 weakly excites
- L10:2274 weakly excites
- L15:1864 weakly excites
- L11:2434 weakly excites
- L24:401 weakly suppresses 'it' referring back to a noun
- L17:1547 weakly suppresses
- L17:2611 weakly excites
- L16:3014 weakly excites
- L16:1550 weakly excites
- L11:2383 weakly excites
- L22:729 weakly suppresses
- L15:2545 weakly excites
- L13:2370 weakly suppresses
- L13:2024 weakly suppresses
- L19:476 weakly excites
- L19:1367 weakly excites
- L24:411 weakly excites
- L19:2595 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
- ystate
- 的动作
- 家中
- engers
- 标志
- же
- usterity
- illage
- หาย
- rei
Suppresses
- /cop
- LOC
- locale
- _memcpy
- .Win
- fits
- 是一款
- López
- LPARAM
- PHY