decimal point inside a number
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. 24 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 dot in decimal numbers like '29.5 days' or '0.25V'. Turning it up gave random, scrambled text, so the steering does not show a clear theme beyond the snippets.
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.
-
28 days and a lunar cycle lasting 29.5 days.⏎In addition, folklore often suggests that women are more likely to be
-
please remember to carry a converter!⏎Precaution 2: Choosing a power bank⏎So far, the standard voltage of USB is 5V±0.25V. However, the output voltage of some power banks on the market
-
best performance in delay time, area, and power product results. A 64-bit carry-look-ahead adder with RALA is fabricated by 0.25-mum CMOS technology to verify its feasibility and functionality. The area
-
regularity improves at each step).⏎The big-picture takeaway from TSMC’s announcement is that while double patterning is already in<|endoftext|>- Размер: 2.3 Mегабайта⏎- Количество слайдов: 1
-
well as Ataluren pontent inhibitor an summary of the Ataluren pontent inhibitor pathways/proteins that are activated in response to each type of damage. A. A double strand break DNA break (DSB) most often results in
-
<|endoftext|>Originally intended for use as a video connector in radar applications, UHF coaxial connectors are general purpose units developed for use in low frequency systems from 0.6 – 300 MHz Invented for use in the radio industry in
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: we can get a lot of information from the we of the we we we we we we we we we we we we we we we we we we we we we we we we we we we we we we we we we we we we
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
reach out to Neuro-Behavioral Clinical Research for help and information on anxiety.⏎RECEIVE A $75 BONUS⏎WHEN YOU REFER A FRIEND!<|endoftext|>- About CSAIL⏎- News + Events⏎- Alumni & Friends
-
toured South India extensively and composed songs on the various deities of the religious pantheon. He had met both Tyāgarāja and Śyām<|endoftext|>Different cultures have been using psychedelic substances in rituals for thousands of years. In many
-
can safely remove.⏎But remember that you should never cut off more than two-thirds of a root at any time.⏎Also,<|endoftext|>Title VII makes sexual discrimination, i.e., using gender as a reason for making
-
acclaimed literary biographer of Kurt Vonnegut and Harper Lee comes the life story of a song, one of the most iconic ever written: John Lennon's "Imagine".⏎"Twenty-two lines of graceful, plain-spoken faith in the power
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
- L23:509 strongly suppresses
- L23:179 strongly excites
- L23:561 strongly excites
- L22:67 strongly suppresses
- L21:524 strongly suppresses
- L22:597 strongly suppresses end of document token
- L22:3045 strongly suppresses
- L22:820 suppresses
- L19:1197 suppresses
- L20:473 suppresses
- L18:382 suppresses
- L22:376 excites
- L18:227 suppresses
- L18:1808 suppresses
- L16:432 suppresses period after month or initial
- L15:403 suppresses period inside an abbreviation
- L19:68 excites full stops closing a claim
- L19:344 excites periods ending explanations
- L19:3015 excites
- L17:347 excites full stops ending informative sentences
- L22:430 suppresses
- L16:762 excites
- L9:16 suppresses unclear: periods inside numbers and names
- L15:184 suppresses
Feeds later layers
- L27:446 strongly excites
- L26:1200 strongly excites
- L27:1637 strongly excites
- L27:137 strongly excites
- L27:1588 excites
- L27:248 excites
- L26:114 excites
- L26:517 excites
- L27:479 excites
- L27:647 excites
- L27:508 excites
- L27:59 excites full stops in running text
- L26:450 excites
- L26:773 excites
- L26:157 suppresses
- L27:94 excites
- L27:854 excites
- L26:778 excites capital letters inside acronyms
- L27:452 excites
- L26:630 suppresses
- L27:123 excites
- L26:372 excites
- L26:27 suppresses
- L27:228 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
- The
- That
- However
- This
- Then
- The
- →If
- →The
- There
- These
Suppresses
- 的方向
- 已久
- 月以来
- alytics
- eness
- 的身影
- matter
- utta
- �能
- 好