end of document token
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. 13 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 when one web page or document ends and the text is about to switch to something unrelated. It is mostly quiet elsewhere. Turning it up made the model's writing wander and repeat itself, but that does not clearly connect to the trigger.
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.
-
, but pointers can be uninitialized, which is a nasty security hole. An eagerly evaluating functional language without hacks, would only allow for acyclic graphs of pointers.<|endoftext|>The Death And Life Of The Great Lakes Pdf⏎- and pdf⏎- Tuesday
-
the other hand, are designed to handle complex logic functions that require high performance and low latency.⏎Developing software for microcontrollers is relatively easy compared to FPGA development<|endoftext|>The deepest human need is to engage with the world: confidently, productively,
-
work, including popular decision trees⏎- Get a comprehensive view of the machine learning ecosystem in R⏎- Explore the powerhouse of tools available in R’s caret package<|endoftext|>I just discovered a blog post by Ann M. Little, a historian at Colorado
-
manage operations in your warehouse.⏎- Lay out a proper strategy for handling damaged products and returns from customers.⏎- Implement pest control measures to protect your products.⏎-<|endoftext|>Nucleation properties of respirable coal dust (terminal progress report).⏎Physics Department
-
00 training examples, you can choose to pick 100 random samples (batch size=100) for each EPOCH (Training cycle)<|endoftext|>|Origin:||co
-
will discuss a basic MySQL installation on a Linux virtual server, and some of the basic MySQL commands.⏎Full documentation can be found on the MySQL.com website here.<|endoftext|>The law applicable to military strategic use of outer space⏎New technologies and the law
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: how to make a community that is open, and how to make the community members feel safe and connected. This is what I am working on in my community. I have been thinking about the importance of these values. This is what I have
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
hook them to books.⏎- Read in front of your child. Be a role model for them. Kids are the best observers, and they tend to imitate there parents.⏎- Show the various options to your child, discuss books with them,
-
the entire crisis on p-values.⏎I think to develop a better understanding of why there is a “crisis”, we need to step back and look across differend fields of study. There is one key question you should be asking yourself: Is
-
your life. And you are to make them known to your children and to your children’s children – the day you stood before the L-ord, your G-d, at Sinai” (Deuteronomy 4:9-10
-
uscule script and its splendid and complex derivative, known as Gothic, were the principal calligraphic styles from the 9th to the 14th cent.⏎The humanistic handwriting style of the Renaissance, a deliberate imitation of Caroling
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
- L6:1524 strongly excites
- L4:1203 suppresses
- L10:214 suppresses
- L6:689 excites
- L10:2199 suppresses
- L7:114 excites
- L8:1728 suppresses
- L8:492 suppresses
- L9:1593 excites
- L12:992 excites
- L11:2145 excites
- L7:220 suppresses
- L2:281 suppresses
- L11:345 excites
- L11:710 suppresses
- L11:258 excites
- L4:1963 suppresses
- L5:24 excites digit inside a date or number
- L12:661 excites
- L12:1334 suppresses
- L11:1348 excites
- L12:701 suppresses
- L11:639 excites
- L10:2250 suppresses
Feeds later layers
- L18:467 strongly suppresses
- L14:1200 strongly suppresses
- L18:897 suppresses
- L14:523 suppresses
- L16:1004 suppresses
- L17:279 suppresses
- L14:1332 suppresses
- L19:402 suppresses
- L18:1186 suppresses
- L19:47 excites
- L14:1590 suppresses
- L17:848 suppresses
- L17:523 suppresses
- L18:214 excites end of document marker
- L15:1616 suppresses word fragments mid-word
- L16:1251 excites
- L16:387 suppresses
- L16:822 suppresses
- L18:592 suppresses
- L15:2087 suppresses
- L15:1065 excites
- L16:444 suppresses
- L15:1404 suppresses
- L18:283 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
- am
- ishly
- ran
- sympath
- mlin
- amic
- amins
- amel
- ucker
- translation
Suppresses
- "]->
- _Msk
- ")->
- gard
- eña
- \")
- 糠
- ㏄
- >")⏎
- ()")⏎