first piece of a split name or term
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. 18 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 first fragment of a longer word or name that the tokenizer splits into pieces, such as "Inter" in "Interchange" or "Ass" in "Association". The examples are mostly institutional or scientific names. Turning it up gave repetitive or off-topic text, so the steering doesn't add much.
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.
-
Vashishth has found a 92 percent correlation between the cross-links and collagen stiffness. The bones for the study were obtained from the National Disease Research Interchange (NDRI).⏎The details of what happens to the collagen are tricky.
-
In early December, the East African Paediatric Association (EAPA) and Kenya Paediatric Assocation (KPA) in collaboration with the American Academy of Pediatrics and Royal College
-
waves created by the explosion allowed researchers Elizabeth Silber and Peter Brown of the Meteor Infrasound Group at the Univ. of Western Ontario to determine that the explosion
-
the scientific literature on cow-calf separation to provide perspective and, hopefully, guidance. The findings of the committee were presented at the 2019 Dairy Cattle Welfare Symposium in Tampa, Fla., followed by an active discussion among the conference
-
is the school's Early Years Policy and information regarding books and apps suitable for our Foundation Stage children as well as a recommended list of Early Years apps. Our Curriculum Overviews can be found on our class pages of this website.⏎Getting Ready for School
-
finding, the researchers have clarified the way in which this oscillator adjusts its activity throughout the day. They show that the three proteins studied, the Pseudo-Response Regulators PRR5, PRR7 and PRR9, associate with promoter
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: and the the, and the, and the, and the, and the, and the, and the, and the, and the, and the, and the, and [a, and [a, and [a, and [
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
crucial to learning, however you need to search for worksheets that show the very same skills in a selection of ways. This not only stops monotony setup in, but additionally strengthens the principles by urging recognizing too. The importance of reading
-
. v is detached from the original object it was pointing to and now LOCAL VAR. v and vh point to same object.Since vh is a new obj. instance var vh.i = 10 and HENCE v.i=10. LOCAL
-
and merchandising. Artists themselves are looking for new ways to make money, they have to think outside of the box. Since the music became a digital product record sales give the wall to the royalties. The audience is changing step by step,
-
creating sort of territorial⏎reference points was probably one of the aims for such locations, considering that sacred mountains strategical position marks the⏎northern limit of the Po valley, which was, at that time
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
- L16:1323 suppresses
- L16:1335 excites
- L16:627 excites
- L17:603 excites inside organisation and product names
- L15:1410 suppresses
- L15:1811 excites
- L14:2831 excites
- L14:553 suppresses
- L15:1701 excites
- L15:1022 excites
- L15:658 excites
- L15:2774 suppresses
- L14:817 excites
- L16:847 excites
- L15:713 suppresses
- L15:1580 excites
- L15:1653 suppresses
- L14:459 excites
- L15:2081 suppresses
- L13:1492 suppresses
- L9:392 suppresses "and" inside organisation names
- L17:1940 suppresses
- L16:1017 excites
- L8:982 suppresses
Feeds later layers
- L20:463 suppresses
- L20:353 suppresses
- L20:398 suppresses
- L19:211 excites
- L24:78 excites
- L19:155 suppresses
- L19:1009 weakly excites
- L21:243 weakly suppresses
- L19:667 weakly suppresses
- L20:225 weakly suppresses
- L20:494 weakly suppresses
- L20:328 weakly suppresses
- L19:1860 weakly suppresses
- L27:895 weakly excites
- L22:1295 weakly suppresses 'as' before 'well' or 'as much'
- L21:720 weakly suppresses
- L22:1619 weakly excites
- L20:1108 weakly suppresses
- L22:255 weakly suppresses
- L26:962 weakly suppresses
- L23:1985 weakly excites
- L19:608 weakly suppresses
- L25:334 weakly suppresses
- L25:62 weakly 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
- aken
- uncate
- FILENAME
- nameof
- الدولية
- ObjectType
- igious
- ánd
- rompt
- enerative
Suppresses
- (SS
- блок
- cek
- ces
- iards
- гля
- 的孩子
- (style
- (cert
- (car