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. 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 lights up on the marker that signals the end of one document and the start of an unrelated one. When it was turned up, the model produced long runs of repeated digits, which doesn't tell us much about its job.
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.
-
300K Trees in Nicosia”) NGO, Yesil Baris (Cyprus Green Action Group) NGO.⏎Started: February<|endoftext|>Public sector workers earned 1% less per hour, on average, than private
-
.⏎They found that only 12.5% were able to correctly answer the questions while some students took six minutes to do so.⏎The average time it<|endoftext|>Since that Egypt is known as the land of civilization, its ancient history along together
-
ccoli and mushroom omelette: Saute broccoli florets and sliced mushrooms and fold them into a fluffy egg omelette for a filling and nutritious breakfast.<|endoftext|>Hanukkah begins this year on December 1st, at sundown.
-
880 genera and 134 families. Moreover, the assemblage represents approximately 10% of the vascular flora of the Mediterranean.<|endoftext|>Christian Henriot, IAO (first draft,
-
yourself a few pretty pennies. For instance, a kilogram of basic carrots will cost you 69p at Sainsburys, 75<|endoftext|>by Lori Valigra⏎Mainebiz news story⏎Maine could stand
-
spiritual refuge or seek blessings from Lord Shiva. The calm and serene surroundings of the temple attract a lot of people. Its ancient architecture is even more attractive.<|endoftext|>Japanese art has a rich and varied history, spanning thousands of years and encompassing
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: In I Am I Am But The I I I In I The So This So So So So So I So So So So So I So So So So I So So So So I So So So So So I So So So So So
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
its DataSet. Dim thisDataSet As DataSet = CType(DataGrid1.DataSource, DataSet) ' Get the DataTableCollection. Dim tablesCollection As DataTableCollection = thisDataSet.Tables ' Get a specific table by name. Dim table As DataTable = tablesCollection("
-
under CC BY-NC-SA⏎Given an integer , the exponential sum version of the Gauss sum modulo is defined by⏎and the quadratic residue version is defined by⏎where is the Jacobi symbol, which has values
-
-south linear vent system over the past 5 million years; the most recent recorded eruption took place between 600 - 700 A.D. Harrat Khaybar contains a wide range of volcanic rock types and spectacular land
-
will be placed underwater or be exposed to water consistently.In applications where greater strength and hardness are required.The Difference Between 304 and 316 Stainless Steel Metal results for this questionWhat is electrical conductivity of 304 stainless?
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
- L18:2164 strongly suppresses
- L18:1105 strongly suppresses
- L19:909 strongly excites
- L21:759 excites
- L21:717 suppresses
- L21:524 suppresses
- L18:2043 excites
- L20:505 suppresses
- L18:572 excites
- L18:1808 suppresses
- L16:1579 excites
- L17:2706 suppresses
- L21:210 excites
- L19:1516 suppresses
- L20:79 excites
- L21:466 suppresses
- L20:577 excites
- L18:1950 excites opening parenthesis after a name or title
- L19:1197 suppresses
- L16:2964 suppresses
- L17:1772 suppresses
- L20:2405 suppresses
- L19:1875 suppresses
- L19:2090 excites
Feeds later layers
- L23:179 strongly suppresses
- L24:353 strongly suppresses decimal point inside a number
- L27:446 suppresses
- L26:114 suppresses
- L25:234 suppresses
- L27:1637 suppresses
- L26:1200 suppresses
- L27:137 suppresses
- L27:647 suppresses
- L27:1588 suppresses
- L26:157 excites
- L26:517 suppresses
- L26:773 suppresses
- L23:221 suppresses
- L27:452 suppresses
- L27:228 suppresses
- L27:508 suppresses
- L23:374 suppresses
- L26:810 suppresses
- L27:248 suppresses
- L23:298 suppresses
- L24:192 excites
- L25:484 suppresses
- L27:59 suppresses full stops in running text
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
- ·
- |
- Tube
- idelberg
- →
- .jquery
- .
- onn
- �
- 的方向
Suppresses
- However
- Therefore
- Unfortunately
- The
- This
- These
- Since
- Similarly
- Meanwhile
- Thus