'of' after all, each, or any
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. 20 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 'of' when it follows words like 'all', 'each' or 'any two', as in 'All of these cases'. When it was turned up, the model's replies did not show a clear theme, so the link to the text is only seen in the examples.
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.
-
the most important factor in determining how a person came to use one style or another in their most intimate relationships, namely with their families, is through learning.⏎All of the styles of anger management are learned styles. Consequently, if they were learned,
-
some individuals, which may be greater than another’s needs. Further, the absence of certain diseases does not indicate that a person is meeting their nutrient needs.⏎Each of us fits into one of these categories:⏎- You are deficient in one or more
-
best by hearing their instructor explain it. Some favor learning it from the textbook. Others need to play with specimens in the lab to really "get it."⏎All of these differences in learning among individuals are often called their "learning styles." There are
-
drugs, devices, and procedures. "Certain trends are increasing longevity and others are reducing it," he says. "I'm trying to figure out how much each of these trends contributes."⏎Over the period Lichtenberg studied a huge number of
-
by , which in turn join to form . These drain into ducts situated between the lobes of the gland (called or excretory ducts).⏎All of the human salivary glands terminate in the mouth, where the saliva proceeds to
-
two end systems⏎B. It specifies the quality of the data & connection⏎C. It specify the size of data⏎D. All of the above⏎6-4. The class-based addressing is also known as⏎
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: each each each, each each, each each, each each, each each, each each, each each, each each, each each, each each, each each, each each, each each, each each, each each, each each
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
and replace them when necessary. This guide provides a comprehensive look into the world of HVAC parts.⏎Understanding Different HVAC Components⏎HVAC systems consist of several key components, each with a specific function. The furnace is responsible for heating air, the air
-
fractive processor in the hands of the very type of machine learning system it's designed to accelerate.⏎The resulting processor is made up of multiple diffractive layers, made up of passive materials that allow the transmission of light. Information is encoded as
-
millions of people worldwide. It is characterized by the abnormal growth of cells in the body, which can invade and destroy surrounding tissues. There are various types of cancer, each with its own unique characteristics and treatment options. In this article, we will
-
Dry Eye” is a term used to describe a group of conditions that cause irritation and damage to the ocular surface. There are many different forms of dry eye, each with a unique cause, and each with a unique set of treatment options available
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
- L19:1837 excites
- L16:1718 suppresses
- L19:2285 suppresses
- L17:2291 excites
- L19:2445 excites
- L16:1418 excites
- L19:2682 excites
- L17:1583 excites
- L14:2704 suppresses
- L19:1422 excites
- L15:1245 excites
- L17:2808 excites
- L18:212 excites
- L16:1317 weakly excites
- L16:1955 weakly suppresses
- L15:275 weakly suppresses
- L15:1679 weakly excites
- L18:525 weakly excites
- L15:2998 weakly suppresses
- L18:2661 weakly suppresses
- L8:467 weakly excites
- L4:450 weakly excites
- L12:1088 weakly suppresses
- L11:272 weakly suppresses
Feeds later layers
- L27:1187 strongly suppresses
- L24:1149 strongly suppresses
- L26:890 strongly suppresses
- L22:477 strongly suppresses
- L23:1305 suppresses
- L23:1807 suppresses
- L22:505 suppresses
- L21:1884 suppresses
- L21:1446 suppresses
- L23:485 suppresses
- L22:940 suppresses
- L23:1240 suppresses
- L23:96 excites
- L21:744 suppresses
- L27:769 suppresses
- L25:1394 suppresses
- L23:402 excites
- L21:1983 suppresses
- L25:1720 weakly suppresses
- L24:3021 weakly suppresses
- L26:917 weakly suppresses
- L21:2646 weakly suppresses
- L22:1633 weakly suppresses
- L21:840 weakly 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
- 在内的
- 粕
- hunter
- ACCEPT
- uru
- ati
- skip
- tics
- aze
- igator
Suppresses
- each
- each
- 每个
- Each
- Each
- EACH
- 每個
- _each
- 每位
- 每一个