colon after a metadata label
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. 2 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 lights up on the colon after labels in catalogue-style records, such as "Published:", "Date:" or "Publication Date:". Turning it up did not make the writing look like those records.
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.
-
Science Grants by Hughes Institute⏎By WILLIAM CELIS 3d⏎Published: October 4, 1994⏎An $86 million grant
-
801-1869⏎Articles of Agreement and Specifications, Topeka School House Construction⏎Authors: New England Emigrant Aid Company⏎Date: June 9, 1857⏎Charles H. Branscomb
-
0.3390/ijerph9114170⏎Published: 15 November 2012⏎Abstract: Understanding the effects of
-
Cook, James (Capt.)⏎Subject: Kerguelen Islands, Christmas Island⏎Period: 1785-87 (circa)⏎Publication: Troisieme Voyage De Cook, ou Voyage a l'Ocean Pacifique…⏎
-
results⏎Authors: Walker, Robert J. (John), 1801-1869⏎Date: May 27, 1857⏎In this long and formal
-
an attachment loop.⏎Date: Circa 1st-3rd century AD Condition: Extremely fine. The
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: what is the name of the object and the value of the expression or the result statement and the combination of the result statement and the m and the n and the c and the grouping statement and the m and the n and the c and the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
⏎But here thing is, save() method can return that primary key id value which is generated by hibernate and we can see it by⏎long s = session.save(k);⏎In this same case, persist() will never give any value
-
10 0 20. I hope i have been clear enough. ------------------ Come on in !! Drinks are on the house in the Big Moose Saloon !!⏎Udayan Naik<BR>Sun Certified Programmer for the Java
-
around the world once more to witness another Venus Transit. It's the last that any of us will see
-
system, and multithreading.⏎Also included is coverage of some of Java's most powerful features,⏎such as generics, lambda expressions, and the very important⏎modules feature. A number of recent additions to the
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
- L0:1559 excites
- L0:1981 weakly suppresses
- L0:185 weakly excites
- L0:9 weakly suppresses
- L1:2944 weakly suppresses
- L0:2632 weakly suppresses
- L1:1340 weakly suppresses
- L1:2638 weakly excites
- L1:465 weakly suppresses
- L1:2714 weakly excites
- L1:1777 weakly excites
- L0:334 weakly suppresses
- L1:1393 weakly excites
- L0:2685 weakly excites
- L0:1197 weakly suppresses
- L1:1887 weakly excites
- L1:2674 weakly suppresses
- L1:1620 weakly suppresses
- L1:1877 weakly suppresses
- L0:1591 weakly suppresses
- L1:1918 weakly suppresses
- L1:190 weakly excites
- L1:1543 weakly excites
- L1:535 weakly suppresses
Feeds later layers
- L17:343 suppresses
- L6:378 suppresses
- L20:431 suppresses
- L20:936 suppresses
- L26:1014 suppresses
- L4:368 suppresses
- L21:448 suppresses
- L9:195 suppresses
- L25:1298 suppresses
- L4:2962 suppresses
- L24:1550 suppresses
- L23:917 suppresses the digit '1' starting a year
- L25:1461 suppresses
- L4:1388 excites
- L24:1850 suppresses
- L3:88 suppresses
- L24:479 weakly suppresses
- L22:567 weakly suppresses
- L19:989 weakly suppresses
- L17:388 weakly excites
- L22:2975 weakly suppresses
- L21:2052 weakly suppresses
- L4:310 weakly suppresses
- L7:433 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
- 的一员
- 当晚
- 受益
- 对她
- .Debugger
- 렴
- 的问题
- cmath
- CID
- 請求
Suppresses
- uncate
- osi
- sealing
- orges
- otlin
- @Override
- riding
- andscape
- urga
- нести