letter k starting a unit or fragment
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. 26 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 responds to the letter k when it begins a piece of text, most clearly as the 'k' in units like kPa, kph or kV after a number. The negative side picks up similar k and Cyrillic к starts, so the pattern is somewhat mixed. Steering turned repetitive, mostly around the word 'price' and similar words.
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.
-
in feet, meters, centimeters, psi, kPa, or bar.⏎The most accurate method of obtaining changes in water level is
-
scientists. First, the heat shield helped dissipate some of the spacecraft's 13,000 miles per hour (21,000 kph) initial speed. Then, the vehicle deployed a parachute to slow down further
-
U.<|endoftext|>Power supply at medium voltage⏎From Electrical Installation Guide⏎The term "medium voltage" is commonly used for distribution systems with voltages above 1 kV and generally applied up to and including 52 kV . For technical
-
sun does not rise for 105 days during winter. Temperatures drop to -56˚C and winds blow in excess of 160 kph. Access by ship and plane is limited to a 3-month summer window
-
. Any object that follows one of these prepositions must be in accusative case: durch-throughfuer-for gegen-againstohne-without um-around Er kauft das Geschenk fuer mich.<|endoftext|>In case you are searching for
-
kits. Materials, including the investigational medical product (IMP), comparator drugs, and ancillary supplies, are collected, laid out, and analysed, and a kitting plan is developed to determine which components are included in each kit
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: one day one day one day one day one day one day one day one day one day one day one day one day one day one day one day one day one day one day one day one day one day one day one day one day
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
group is the Russian Cossacks (казаки) of the Don, Kuban, Terek and Ural rivers and the Ukrainian Cossacks (козаки), respectively. Somewhat less well known are the Polish Coss
-
thinks: "What he should be really teaching was the mysterious thing which he alone–out of ten thousand, a hundred thousand, perhaps even a million men–knew how to teach: for example–multilevel thinking," which he then
-
is in almost every part of the body including the teeth and blood vessels. It is like the glue that holds everything together and is taken from the Greek word “kola” meaning glue. Thus it is easy to see how important it is in
-
tomatoes, tomatillos, camotes (sweet potatoes), jicama (a tuber also known as sengkwang, yam bean, singkamas, Mexican turnip), amaranth (also called pigweeds)
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
- L20:613 strongly suppresses
- L22:2170 strongly excites
- L18:816 suppresses
- L19:381 excites word-start 'acc' fragments
- L17:1216 suppresses
- L21:590 excites
- L7:674 suppresses
- L19:2402 excites
- L17:437 excites
- L18:131 excites
- L19:1198 excites
- L20:97 excites
- L2:1661 excites
- L5:340 suppresses zeros inside long numbers
- L18:103 suppresses
- L16:81 weakly suppresses
- L0:1878 weakly excites
- L25:413 weakly suppresses
- L18:1442 weakly excites unclear: word fragments and digits inside numbers
- L19:1500 weakly excites
- L20:217 weakly suppresses
- L10:90 weakly excites
- L1:50 weakly excites
- L16:946 weakly excites the word 'late' in time phrases
Feeds later layers
- L27:1851 weakly suppresses capital K starting a word
- L27:855 weakly excites
- L27:879 weakly excites
- L27:1528 weakly excites
- L27:876 weakly excites
- L27:384 weakly suppresses
- L27:2099 weakly excites
- L27:2567 weakly excites
- L27:1493 weakly excites
- L27:541 weakly excites
- L27:1453 weakly excites
- L27:2433 weakly suppresses
- L27:1537 weakly excites
- L27:1395 weakly excites
- L27:2939 weakly excites
- L27:2742 weakly excites 'w' inside split words
- L27:914 weakly suppresses
- L27:62 weakly excites
- L27:754 weakly excites
- L27:927 weakly suppresses
- L27:1504 weakly excites
- L27:971 weakly excites
- L27:2511 weakly suppresses
- L27:1667 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
- it
- uncture
- ey
- imize
- lan
- itos
- uddenly
- kk
- undert
- imized
Suppresses
- finger
- onz
- stown
- -regexp
- ter
- .defaultProps
- 北京时间
- rement
- onn
- 日消息