'of' or 'the' after 'part of' or 'subset of'
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. 14 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 words right after phrases like 'part of', 'subset of' or 'during', where the next word names a larger group or category. Turning it up gave text about other topics, so the output does not clearly reflect this.
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.
-
against a predetermined set of requirements and validating that the product meets those requirements.⏎Examples of QC include technical reviews, software testing and code inspections.⏎Testing is a subset of QC. It is the process of executing a system in order to detect bugs in
-
can have as many base search paths as you like. Here's one that includes⏎/etc as a search path too.<|endoftext|>Salivary gland (Endocrine system)⏎: #1 is , #2 is , #3 is⏎The sal
-
floxacin is an antibiotic that is part of the fluoroquinolone family. It effectively kills bacteria, while dexameth
-
, Thamserku, Kantega, Mera Peak and Island Peak.⏎Mt. Everest, which is part of the Himalaya range, is located on the border between Nepal and Tibet. Rising to
-
However, the Supreme Court justices are chosen by the president, the head of the executive branch. But while the president makes that selection, the Senate, part of the legislative branch, approves the decision.⏎Each branch must depend on the other in order
-
<|endoftext|>The Battle of the Kasserine Pass was a battle that took place during the Tunisia Campaign of World War II in February of 1943. It
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 have been, and has been, and is most well known by me. I have been, and is most well known by me, and is most well known by me. 100% of the first person. 1
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
the 1999 State of the Union Address, and then determine how newspaper articles and television reports describe and analyze the event. Use this...⏎6th - 12th English Language Arts CCSS: Adaptable⏎The State
-
in-depth reports from Israel⏎Connect with Israel, right from your home.⏎Lift up the voice of truth and hope⏎Support Jerusalem-based Zionist journalism.⏎Already a member? Login here.<|endoftext|>The fractured rock surfaces were observed under different loading stages
-
characters shorter!⏎The current version of bool-extras is 0.3.0 and is available on Hackage.⏎If you have cabal-install (which you really should have, by the way), the
-
elements of a successful speech. Not only do they get to become better-informed...⏎7th - 11th Social Studies & History⏎President Obama's State of the Union
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
- L12:1356 excites
- L13:2420 excites
- L8:1885 suppresses
- L10:2016 weakly suppresses
- L13:3062 weakly suppresses
- L9:1737 weakly suppresses
- L13:1489 weakly suppresses
- L3:647 weakly suppresses
- L7:2108 weakly suppresses
- L13:2564 weakly suppresses
- L9:242 weakly excites
- L5:1096 weakly excites
- L12:1590 weakly excites
- L12:995 weakly suppresses
- L1:199 weakly excites
- L12:2040 weakly suppresses
- L0:338 weakly excites
- L6:784 weakly excites
- L8:1804 weakly suppresses
- L13:2999 weakly suppresses
- L2:2516 weakly excites
- L5:1477 weakly excites
- L12:3006 weakly excites
- L12:3017 weakly suppresses
Feeds later layers
- L15:2333 weakly excites
- L15:978 weakly suppresses
- L20:2988 weakly excites
- L18:1691 weakly excites
- L15:1698 weakly suppresses
- L16:1759 weakly suppresses
- L18:1858 weakly excites
- L19:2443 weakly excites
- L15:1095 weakly suppresses
- L15:2058 weakly suppresses
- L17:2175 weakly excites
- L22:2127 weakly suppresses
- L16:1121 weakly excites
- L16:1584 weakly suppresses
- L18:1699 weakly suppresses
- L18:1356 weakly excites
- L17:909 weakly suppresses
- L17:209 weakly excites
- L18:1879 weakly suppresses
- L15:2221 weakly excites
- L18:327 weakly excites
- L15:2195 weakly excites
- L20:1817 weakly suppresses
- L16:2788 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
- ialect
- @update
- extras
- ubyte
- ungs
- NodeType
- CIT
- toc
- egov
- avia
Suppresses
- ancel
- 泣
- 干
- 精子
- åde
- edian
- `,`
- 从
- rance
- rence