end of text token before new document
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. 22 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 invisible end-of-document marker that separates one web page from the next. It is mostly silent otherwise. Turning it up made the model write long, rambling text with no clear topic.
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.
-
environmental variables, or logging properties⏎- Monitor Spark using the web UIs, metrics, and external instrumentation⏎- Understand performance tuning considerations⏎- This course is<|endoftext|>Class 1M have been learning the story of Snow White and the Seven D
-
real-time scanner hooks deep into your operating system's core functions and looks for any signs of malicious code. Any time the operating system accesses a file, the on<|endoftext|>The Environmental Inequality Lab⏎How Can We Reduce Disparities in Environmental Quality
-
, in its decision dated 23.12.2021, specifically mentioned three main models as services of cloud computing contracts: Software as a<|endoftext|>The pelvic floor is a group of muscles that act as a sling to the bottom
-
credits sourced from sustainable practices in climate-affected households in Mongolia. The project's digital MRV (measurement<|endoftext|>In this blog post we will use a Python script to work out how many leg
-
which can represent either a 0 or a 1. Quantum computers, on the other hand, use quantum bits or qubits, which can exist in multiple<|endoftext|>|Classification and external resources|⏎It is typically due to aspiration by alcoholics,
-
holdinger F., Tucker M.R., Himmelbach A., Stein N., Mascher M., Nagel K.A., De Gara L., Simmonds<|endoftext|>Last Census - 2005⏎Cameroon is a culturally diverse cost
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 if we get outside of the three different 'redum' (here, 'redu' and 'redu' in both) and also, get outside the three 'redu' (here, three 'redu'
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
a matrix⏎Find nth maximum⏎Matlab Basics II - Create a vector with a repeated entry⏎Create sequnce 1 4 9 16 25........
-
stationed at the legionary fortress at York. A few 'centurial stones' have been uncovered at the fort, which record building work undertaken by<|endoftext|>« PreviousContinue »⏎MACMILLAN & CO.'S BOOKS⏎STUD
-
.<|endoftext|>Using Command Line using Visual C++ 2008⏎- Select Start >> All Programs >> Microsoft Visual C++ 20
-
few minutes with the drag-and-drop MakeCode programming site, learn computer science using the CS Discoveries class on code.org, jump into CircuitPython to learn Python and hardware together, TinyGO, or even use the Arduino IDE. Circuit Playground Express
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
- L7:1400 weakly suppresses
- L11:1974 weakly suppresses
- L19:251 weakly excites
- L1:2964 weakly excites
- L6:2865 weakly suppresses
- L3:303 weakly suppresses
- L12:437 weakly excites
- L6:2912 weakly excites
- L6:556 weakly suppresses
- L14:72 weakly excites
- L9:263 weakly suppresses
- L19:2762 weakly suppresses
- L11:804 weakly excites
- L19:2781 weakly suppresses
- L5:1910 weakly excites
- L8:15 weakly excites unclear: mid-phrase words across many languages
- L13:2954 weakly excites
- L21:2910 weakly excites unclear: mid-sentence words before a choice
- L17:1554 weakly suppresses
- L13:3 weakly suppresses
- L15:29 weakly suppresses
- L7:1638 weakly suppresses
- L8:366 weakly suppresses
- L10:2018 weakly excites
Feeds later layers
- L24:1091 excites
- L27:1540 weakly excites
- L24:1736 weakly suppresses
- L23:2297 weakly suppresses
- L23:2602 weakly excites
- L24:1205 weakly suppresses
- L24:1192 weakly excites
- L26:345 weakly suppresses
- L23:2683 weakly suppresses opening parenthesis before an acronym
- L27:963 weakly suppresses
- L24:2900 weakly excites
- L23:337 weakly excites
- L24:1999 weakly excites
- L26:2736 weakly excites
- L24:913 weakly excites
- L25:438 weakly excites
- L26:869 weakly suppresses word before him or he
- L24:1476 weakly suppresses
- L24:772 weakly suppresses
- L23:577 weakly suppresses
- L24:2875 weakly excites
- L25:2973 weakly suppresses
- L23:1591 weakly excites
- L24:909 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
- ainen
- <State
- Python
- laces
- <Project
- <Address
- �
- argout
- /<
- cke
Suppresses
- ��
- pmat
- 也只能
- onian
- TCHA
- 各异
- avings
- echa
- 生素
- >');⏎⏎