table column header boundaries
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. 8 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 double bar that separates a column heading from the next heading in pasted tables, such as the gap between 'Format' and 'File Header'. Turning it up made the model's output drift into unrelated text, with no clear theme.
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.
-
any parameters. Other file formats can be read by creating an NMRView parameter file that contains a description of the layout of the file.⏎|Program||Format||File Header||Block Header||Parameters|⏎|Varian||Serial||32
-
Gulf, down 2.5 percent⏎|Region||December 2013||December 2012||Percent Change|⏎|Region||December 2013||November 2013||Percent Change
-
reports, he continued "wielding his machete until the foe had<|endoftext|>|Theme||Visible||Selectable||Appearance||Zoom Range (now: 0)|⏎Well
-
of Sukkot. But everywhere else, the following chart applies.⏎Fall 2001 Jewish Holiday Calendar⏎|Biblical Name||Modern Holiday||Hebrew Date||English Date|⏎|Feast of Trumpets (Lev
-
GNU Oflox⏎This is a GNU package:octave⏎released on 31 May 2012⏎|License||Verified by||Verified on||Notes|⏎|GPLv2||Kelly Hopkins||6 December
-
ARAB 300- Advanced Arabic⏎Arabic Long Term Schedule (updated Spring 2012)⏎|FALL 2012||ARAB 100|⏎|SPRING 2013||
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 a 4th year in college, and I want to know how to calculate the probability of getting a higher number of 3 in my 4th year. Let's see if I can calculate the probability of getting a higher
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, we could:⏎|Vitamin B12 Need||94%||Confirm|⏎|Lupus (SLE)||2%||Ruled out|⏎|A
-
Place of Publication||Perth, Western Australia|⏎|Number of pages||173|⏎|ISBN (Print)||9 7
-
FF’s Daniel Raimi gives a balanced and accessible view of oil and gas development, drawing on both scholarly literature and his t...<|endoftext|>|Full name||Noam Chomsky
-
17.0.0/17⏎- We need a network with <|endoftext|>|Classification||Electronic musical instrument, synthesizer|⏎|Inventor(s)||Suzuki Musical Instrument Corporation|⏎|Tronichord, autoh
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
- L6:1056 suppresses
- L7:335 excites
- L5:1512 suppresses
- L5:894 suppresses
- L2:974 excites
- L7:63 weakly excites
- L6:1691 weakly suppresses
- L6:16 weakly excites
- L4:556 weakly excites
- L7:1717 weakly suppresses
- L7:69 weakly excites
- L6:234 weakly suppresses
- L5:50 weakly excites
- L7:1242 weakly suppresses
- L1:2801 weakly excites
- L2:916 weakly excites
- L6:2156 weakly excites
- L5:75 weakly excites
- L3:115 weakly suppresses
- L7:271 weakly suppresses
- L3:1962 weakly excites
- L0:2821 weakly excites
- L6:7 weakly suppresses
- L5:64 weakly excites
Feeds later layers
- L9:1066 excites
- L13:1850 excites
- L10:2609 excites
- L14:2303 weakly excites
- L9:1734 weakly excites
- L18:1900 weakly suppresses
- L11:2508 weakly excites
- L13:201 weakly excites
- L23:2766 weakly suppresses
- L14:3039 weakly excites
- L12:660 weakly excites
- L11:2507 weakly excites
- L10:1849 weakly excites
- L15:234 weakly excites
- L10:2219 weakly suppresses
- L20:2520 weakly suppresses
- L9:1638 weakly excites
- L16:1647 weakly suppresses
- L22:853 weakly suppresses
- L15:457 weakly excites
- L10:2862 weakly excites
- L23:1323 weakly excites
- L19:2291 weakly excites
- L18:3015 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
- amba
- 门前
- ��态
- ,key
- abbo
- atable
- acted
- ='.
- rama
- =key
Suppresses
- illery
- erville
- ọi
- eros
- глав
- qrt
- 伍
- 辜负
- UTURE
- éré