pieces of software and library names
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. 11 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 later pieces of technical names, such as 'lib' in 'GLib', 'Bool' in a Haskell module, '.curve' in a scipy function or '2' in 'urllib2'. It is mostly quiet or slightly negative on names in ordinary text, like 'Journal' or 'Base Ten'. Turning it up or down did not give a clear story.
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.
-
since it is a function instead of syntax, it can be partially applied and passed around to higher order functions, like so:⏎ghci> :m + Data.Bool.Extras⏎ghci> let yesOrNo = bool
-
as resource databases, with a database type of GLib, and a library-specific creator ID. The GLib library will usually contain three
-
:|| Web page: http://www.tandfonline.com/VECE20|⏎|Order Information:||Web: http://www.tandfonline.com/pricing/journal/VECE20|⏎When requesting a correction,
-
ography. arXiv - Computer Science > Cryptography and Security (21 October 2016) | arXiv: 1610.06918 | PDF⏎We ask whether neural networks can learn
-
up your code into a main portion (of at most 32K), and a number of libraries (each of at most 32K).⏎GLib (pronounced ``gee-lib'') shared libraries are a way to implement shared
-
The widget uses scipy.curve_fit to find the optimal values of the parameters.⏎The widget works only on regression
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: I will be a scientist? I will be a doctor? I will be a teacher? I will be a nurse? I will be a writer? I will be a pilot? I will be a doctor? I will be a researcher?
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
write to show their mathematical thinking.⏎Contents 4-8 Counting and Cardinality Task Cards (CC) 9-13 Numbers and Operations in Base Ten Task Cards (NBT) 14-18 Operations and Algebraic
-
Gall, head of the French Space Agency CNES.⏎Courtesy ESA/Rosetta/MPS for OSIRIS Team MPS/UPD/LAM/IAA/SSO/INTA/UPM/DASP/IDAThe spacecraft will spend
-
the brown dwarfs, identified as CFBDSIR 1458+10B, in a paper in the current issue of The Astrophysical Journal.⏎"At such temperatures we expect the brown dwarf to have properties that are different from
-
iztli Calmecac performs Aztec dance from Pennsylvania, another in the 2009 Homegrown Concert Series sponsored by the American Folklife Center.⏎Calmecac, Ollin Yoliztli -
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
- L8:509 strongly excites
- L9:1164 strongly excites
- L7:1681 suppresses
- L8:1031 excites
- L10:2154 excites
- L9:2303 suppresses
- L6:673 suppresses "to" after adjectives like obvious or meaning
- L10:2787 suppresses unclear: document boundaries and digits
- L6:2192 suppresses
- L10:1642 suppresses
- L4:1183 suppresses
- L4:1797 suppresses
- L9:1366 weakly excites
- L3:2128 weakly suppresses
- L7:1755 weakly excites
- L4:2121 weakly excites
- L7:900 weakly excites
- L9:814 weakly suppresses
- L3:2733 weakly suppresses the word 'form' in 'formative'
- L4:1765 weakly excites
- L3:1561 weakly excites
- L6:728 weakly excites
- L9:461 weakly excites
- L4:905 weakly suppresses
Feeds later layers
- L14:1248 excites
- L27:362 excites
- L26:174 excites
- L16:1446 excites
- L24:228 excites
- L13:1354 excites
- L19:1596 excites
- L20:536 excites
- L25:387 excites
- L17:1674 excites
- L19:913 excites
- L26:593 excites
- L13:1576 excites
- L13:1845 excites
- L27:875 excites
- L25:294 excites
- L27:370 excites
- L16:2651 excites
- L21:389 excites
- L23:372 excites
- L25:1135 excites
- L12:498 suppresses
- L21:1842 excites
- L19:954 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
- iswa
- wipe
- ู้
- ',(
- ",(
- ascus
- olves
- Incorporated
- ::_
- życ
Suppresses
- ?
- ??
- ?
- ???
- -
- "
- 5
- "?
- 義
- 义