unclear: word fragments, weak
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. 10 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 responds only weakly and mostly to pieces of longer words that were split apart, like 'Zeb' in Zebulon or 'Gunn' in Gunnuk. There is no clear everyday theme, and turning it up did not reveal one.
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.
-
Montgomery Pike⏎In mid-July 1806 Lewis and Clark were on their way back from the Pacific. At the same time young army Lieutenant Zebulon Montgomery Pike left St. Louis with twenty-three men to reconnoiter
-
the waterway. A village steeped in Native tradition and pride, the people live in harmony with the bears as they frequent the area during salmon runs. Gunnuk Creek watershed is a vital source of subsistence for the people of Kake,
-
God can use imperfect people for His purposes. Christ<|endoftext|>Published on November 23rd, 2008 | by What2Learn0⏎Printable coloring pages – the alphabet⏎Free coloring pages to print for kids to help them
-
different. Although the latest legs have exceeded all expectations in stress-testing–the 3-D-printed plastic is about as strong as a traditional prosthetic-–coons and associate professor Matt Ratto have yet to find the perfect thickness, the
-
cost less than €1 each to produce. 10 confirmed patients have been connected to a breathing machine via a 3D printed valve.⏎Although Fracassi is<|endoftext|>Key: "S:" = Show Synset (semantic) relations,
-
in reality they are plagued with lies and evil. There appearance will make it very difficult for Hamlet to uncover the truth.⏎Polonius is constantly cncerned with his appearance. He always wants to keep up the appearance of loving and
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: What is the relationship between the two things: the first one is the subject of the second one? (The second one is the object of the first one) Okay, so that is the first one. The second one is the object of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
lower voltage without necessarily losing power.⏎Figure 4: MPPT Charge Controller⏎Figure 5: Power Curve for a PV Panel with Charging Ranges for PWM Controllers⏎MPPT controllers are able to pull this off because
-
172.16.128.0/17 SUBNET3 : 10101100.00010001.0yyyyyyy.yyyyyyyy => 172
-
Human Development for Helping Professionals (2006, 2nd ed. ) by Patriia C.⏎John Winthrops City on a hill EssayJohn winthrops city on a hill John Winthrops 1630 speech
-
of chrysanthemum tea, which is made with crushed dried flowers and boiling water.⏎What forms of chrysanthemum are available?⏎Dried chrysanthemum flowers can be found<|endoftext|>Copyright © University of Cambridge. All
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
- L6:24 strongly excites
- L8:258 strongly excites
- L5:4 strongly suppresses
- L9:870 excites
- L5:25 excites
- L6:115 excites
- L9:15 excites
- L3:2870 excites
- L9:305 suppresses
- L9:99 suppresses
- L7:1201 excites
- L8:1426 excites
- L6:169 excites
- L8:154 excites
- L9:536 suppresses
- L3:22 suppresses
- L8:677 weakly excites
- L5:187 weakly excites
- L6:198 weakly suppresses
- L6:148 weakly excites
- L9:254 weakly excites
- L5:498 weakly suppresses
- L6:374 weakly suppresses
- L5:36 weakly suppresses
Feeds later layers
- L13:2270 strongly suppresses
- L12:623 suppresses
- L13:105 suppresses
- L11:1191 excites
- L13:260 excites
- L12:134 excites
- L14:185 excites
- L15:118 excites
- L14:697 excites
- L12:243 excites
- L12:1307 suppresses
- L12:1217 excites
- L14:224 excites
- L12:1042 excites
- L11:2437 excites
- L14:98 excites
- L13:606 excites
- L11:779 excites
- L11:1701 excites
- L14:508 suppresses
- L14:74 excites
- L11:70 suppresses
- L11:908 suppresses
- L13:381 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
- PickerController
- цев
- aceous
- acious
- füh
- ulla
- �
- Cock
- ятия
- klä
Suppresses
- fi
- Topic
- ined
- g
- istent
- .VisualStudio
- ersh
- !';⏎
- verity
- 你自己