unclear: periods inside numbers and 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. 9 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 one reacts to periods that sit inside things like version numbers, IP-style numbers, and some names or abbreviations, and it is most active on the dot itself. It is mixed, though, and the steering outputs show no clear theme. Turning it up made the text repetitive and off topic.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
a list of a single compound type like Entity or Structure, was already supported before 9.0.⏎Until 188.8.131.52, if you wanted to create a list of users, you would create
-
NSM2012. You'll have 30 days to complete the course.⏎For the younger driver
-
be (at time of writing) 22.214.171.124-kgpd-071097. This is the version also known as 0.5.0. If you
-
with symptoms such as numbness and tingling in the legs and feet, may cause loss of strength in the muscles it stimulates, depending on the degree of nerve compression.⏎Sciatica occurs when the bones in the spine press on the sciatic nerve
-
taking their pills as directed, the study found.⏎“It does appear that there is a modest reduction in cancer among middle-aged and older men,” said Dr. J. Michael Gaziano, chief of the division of aging at Brigham and Women’s Hospital
-
and other features.⏎Merriam-Webster Medical Dictionary⏎Sponsored by the National Library of Medicine, this site includes⏎links to encyclopedias, glossaries,
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 thinking about the The following, I have been thinking about the The following, I have I have been thinking about lately, I have been thinking about I have been thinking about I have been thinking about I have been thinking about I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
or and donor layers, or domains, in polymer-based solar cells could increase their efficiency, according to a team of researchers that included physicists from North Carolina State University. Their findings shed light on the inner workings of these solar cells, and could lead
-
Maths Chapter 6 Exercise 6.5, we have discussed the Pythagoras theorem. Other important topics related to the chapter are covered in the following exercises.<|endoftext|>What you will learn in this module⏎Understand different tasks that have to
-
Solid Shapes⏎✔ Telling Time⏎✔ Comparing Numbers⏎Why They’ll Love It⏎Your students will review this year’s math concepts in a fun way! They will get:⏎☆ summertime jokes⏎☆ knock knock jokes⏎☆ cell
-
related descendants of that dog to fix the traits they want, said Dr. Donald F. Patterson, a professor of medicine and medical genetics at the University of Pennsylvania. If a popular sire also carries a genetic defect, it too spreads rapidly through the
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
- L5:155 suppresses
- L4:7 excites
- L4:1546 suppresses
- L6:2302 suppresses 'and' joining two parallel items
- L4:848 excites
- L5:168 suppresses
- L6:305 suppresses
- L8:309 suppresses
- L6:1736 suppresses
- L2:9 suppresses sentence-final periods vs dots in code
- L6:466 excites
- L8:143 weakly suppresses
- L8:304 weakly suppresses
- L1:130 weakly suppresses
- L8:740 weakly excites
- L6:871 weakly suppresses
- L8:381 weakly excites
- L8:174 weakly suppresses
- L8:2014 weakly excites
- L6:327 weakly excites
- L5:2159 weakly suppresses
- L8:1848 weakly excites
- L6:2399 weakly suppresses
- L0:1475 weakly excites
Feeds later layers
- L11:1025 suppresses
- L17:1105 suppresses
- L16:894 suppresses full stops ending informational sentences
- L11:132 suppresses period ending an opening topic sentence
- L14:153 suppresses
- L14:155 suppresses sentence-ending periods in prose
- L21:1086 suppresses end of sentence before a claim or statistic
- L16:1545 suppresses
- L19:344 suppresses periods ending explanations
- L12:1062 suppresses
- L22:67 suppresses
- L18:791 suppresses
- L16:436 suppresses
- L17:200 suppresses
- L24:353 suppresses decimal point inside a number
- L19:433 suppresses
- L17:347 suppresses full stops ending informative sentences
- L17:1772 suppresses
- L13:9 excites
- L18:1437 suppresses
- L10:717 suppresses
- L13:345 suppresses
- L16:797 excites
- L20:505 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
- og
- id
- oph
- 引用日期
- encore
- ro
- .pipeline
- idia
- puedo
- imiters
Suppresses
- 然而
- 而在
- however
- 这其中
- whereas
- holders
- tery
- 而对于
- however
- 但这