"etc" closing a list
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. 19 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 word "etc" at the end of a list of examples, such as "bread, rice, grains... etc." When it was turned up, the model's writing mostly fell into repetitive loops, so the effect on the text is not clear.
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.
-
approach to crude oil causation is, the operating expenses tend to increase due to more spending on transportation and indirect expenses like employee travel allowance, third party fleet charges etc. This reduces corporate earnings. The reverse should be true
-
apples are good for heart patients. It and keeps the digestive track balanced. The nutritional composition of apples is very rich in vitamins, minerals, carbohydrates and amino acids etc. Apples are also very good source of vitamin B-Complex.⏎Apple contains
-
reflects our crisis with growing racism and intolerance. Clearly, discrimination against different groups in Egypt is not uncommon whether it was based on gender, religious beliefs, class, etc. Activist Fatma Emam recently wrote an article of her experience as a
-
race of Kurooshdusha – some of famous persons of this race are Sacepenthu, Seyaamagan, Vetharppan etc., Vetharppan’s third son belonged to the Saethe race,
-
as the name of a mountain ridge. The name is derived from a postulated basic meaning “throat, neck,” which Christian Bartholomae translated as neck etc. of daevic beings (see DĒW), and mentioned manao
-
cotton is a popular choice and its demand is increasing.⏎Organic cotton is used in several products such as cotton swabs, baby diapers, towels, sheets, etc. Hemp has high durability and requires less water. Organic hemp is used in paper
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 most effective. That is to say, the most effective. The most effective. That is to say. The most effective. I have not had much. I have not made much. I have not made much. I have not made
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
and chemical requirements for specific purposes."— Presentation transcript:⏎Types of Portland Cement Different types of portland cement are manufactured to meet various physical and chemical requirements for specific purposes. Portlant cements are manufactured to meet the specifications of ASTM C1
-
is another paragraph, with a sentence that is split over several lines.Will get formatted as:⏎This is another paragraph, with a sentence that is split over several lines.<|endoftext|>Insufficient vocabulary, comprehension and literacy skills negatively impact children for the rest
-
CD prevalence, their risk factors, and treatments.⏎Copyright © 2020 The Royal Society for Public Health. Published by Elsevier Ltd. All rights reserved.<|endoftext|>We all grew up hearing the phrase, “Good night, sleep tight
-
⏎- 1st Photos of Mars by Curiosity Rover (Gallery)⏎Copyright 2012 SPACE.com, a TechMediaNetwork company. All rights reserved. This material may not
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
- L18:1437 strongly suppresses
- L18:1471 suppresses
- L18:2155 suppresses
- L18:791 excites
- L16:527 excites
- L17:642 excites
- L17:1218 excites
- L16:432 excites period after month or initial
- L11:1047 weakly excites pieces of software and library names
- L16:826 weakly excites
- L18:113 weakly suppresses
- L15:2237 weakly excites
- L17:347 weakly suppresses full stops ending informative sentences
- L17:1772 weakly excites
- L17:11 weakly suppresses
- L15:2561 weakly suppresses
- L13:961 weakly excites
- L18:48 weakly excites
- L17:2932 weakly excites
- L9:1164 weakly excites
- L8:539 weakly suppresses
- L14:422 weakly suppresses periods after initials and abbreviations
- L10:2963 weakly excites
- L15:2226 weakly suppresses
Feeds later layers
- L24:228 strongly suppresses
- L25:294 strongly suppresses
- L20:536 strongly suppresses
- L26:11 strongly excites
- L27:319 suppresses
- L26:255 suppresses
- L25:656 suppresses
- L26:593 suppresses
- L26:1169 suppresses
- L25:387 suppresses
- L21:148 suppresses
- L26:174 suppresses
- L27:1418 suppresses
- L26:943 suppresses
- L27:25 excites
- L27:1328 suppresses
- L20:622 excites
- L26:62 excites
- L27:580 suppresses
- L27:385 suppresses
- L27:362 suppresses
- L20:1806 excites slash inside web addresses
- L22:1058 suppresses
- L26:1391 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
- 亞
- いたら
- ⏎
- %m
- HasForeignKey
- 是为了
- %
- 第二
- 同一
- 虽然
Suppresses
- .\
- .`
- .]
- .)
- .;
- .�
- .:
- .'
- .”
- .,