document end token, mixed
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. 12 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 invisible end-of-page marker, but also on some ordinary words in the middle of sentences. The pattern is only partly clear. Turning it up gave repetitive, plain sentences.
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.
-
velocity (km/s)...?⏎- Inversely proportional⏎- Directly proportional⏎- Linearly related⏎- Non-linearly related⏎Similar Physics resources:<|endoftext|>RNA functions as the central conduit for information transfer in biology - in simple replicating
-
viewed by: Khanh-Van Le-Bucklin, Stephanie D'Augustine⏎Last reviewed: May 07, 2008<|endoftext|>As we creep into the peak of hurricane season, many media outlets have dispatched reporters
-
, u, t)$.⏎<|endoftext|>While many students view summer as a time to forget about school, it remains an
-
Wenjun Chen, Yan Huang, Jiufeng Sun, Jingtao Men, Hailiang Liu, Fang Luo, Lei Guo, Xiaoli Lv<|endoftext|>Laws and Cases: How to Do Legal Research⏎Sometimes you need to look
-
round," he said.⏎The results are published in the American Journal of Hypertension.<|endoftext|>Nicolaus Copernicus’ 540th birthday: A man
-
y (Zurich).⏎250 pages, 20 x 17 cm, softcover, Edition Patrick Frey (Zurich).<|endoftext|>Pronunciation: (lus'trum), [key]⏎—pl. -
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: the number of the phone number. The number of the phone number is the number of the phone number, which is the number of the phone number. The number of the phone number, which is the number of the phone number, which is
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Nobel Prize a few years before.⏎"I want to build a steam car which will be practical enough for general use," Howard announced. "I need engineering helpmen who understand the dynamics of steam and can supply them to a workable automobile."⏎Dr
-
and some mesotrophic, lakes. Laboratory determination of P release rates from sediment cores is one approach for determining the role of internal P loading and guiding management decisions. Two principal alternatives to experimental determination of sediment P release exist for estimating internal load:
-
, show promise. Magnetic devices could be used to move such cells to diseased areas of the body. But current ways of attaching SPIONs to therapeutic cells are difficult to use and may damage the therapeutic cells. So the researchers set out to
-
photocathode, often based on NiO, is needed. An important and unknown question is the role of water molecules adsorbed on the NiO surface. Kaijian Zhu performs research on the effects of this adsorption.⏎"In this work
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
- L9:1045 suppresses
- L8:3040 excites
- L7:2652 suppresses
- L8:1731 suppresses
- L11:1285 excites
- L8:2703 excites
- L10:981 suppresses
- L10:907 suppresses
- L8:1937 excites
- L4:296 excites
- L9:373 suppresses
- L9:1815 weakly suppresses
- L10:1448 weakly excites
- L8:2524 weakly excites
- L4:1122 weakly excites
- L10:710 weakly suppresses
- L7:1112 weakly excites
- L11:1482 weakly excites
- L10:1577 weakly excites
- L10:1579 weakly suppresses
- L8:824 weakly suppresses
- L10:607 weakly suppresses
- L11:3069 weakly excites
- L7:563 weakly excites
Feeds later layers
- L14:3029 suppresses
- L15:1748 suppresses
- L15:2327 suppresses
- L14:2184 suppresses
- L15:506 excites
- L13:1338 suppresses
- L13:1326 suppresses
- L14:2373 suppresses
- L13:1396 suppresses
- L13:681 suppresses
- L14:354 suppresses
- L14:410 excites
- L14:2284 suppresses
- L15:2476 suppresses
- L17:2299 suppresses
- L15:2047 suppresses
- L16:3000 suppresses
- L13:2422 weakly suppresses
- L14:2186 weakly suppresses
- L13:2361 weakly suppresses
- L15:1769 weakly suppresses
- L14:576 weakly suppresses
- L14:654 weakly suppresses
- L15:1782 weakly 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
- ירות
- EMBER
- adget
- sgiving
- ConfigurationException
- ced
- 宥
- INCIDENTAL
- .Factory
- افة
Suppresses
- oster
- ashes
- держ
- pto
- ftar
- /res
- pps
- tems
- ��
- abs