engineering and measurement wording
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. 22 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 tends to fire in technical writing about machines, threads, sizes and materials, such as 'main circuit', 'major diameter' and 'plate thickness'. It is not clear what turning it up or down does, since the generated text shifted to unrelated topics.
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.
-
the AC asynchronous motor, improve the operation precision, change the power factor, and Realize overcurrent, overvoltage, overload protection and other functions. The main circuit of the frequency converter can be roughly divided into two categories: one is the
-
, generous and clean.⏎Cleaner uniforms must be equipped with appropriate personal protective equipment, such as gloves, water shoes, dust masks, etc., which can not only improve work efficiency, but also reduce or avoid occupational hazards for cleaners.⏎It can
-
forced) and a NO-GO gauge is not answered, the threaded part is deemed to be accepted in totality.⏎But in some specific cases where the thread major diameter (for screws) or minor diameter (for nuts) is critical, separate
-
For example, you may find a threaded rod measuring M8-1.25 x 1 Meter. The M8 (8mm) is the rod's diameter,
-
gauging of threaded parts.⏎Before that, a word about thread gauging practice:⏎In a threaded part, several features are involved – the thread’s major diameter, minor diameter, pitch diameter, pitch (in case of metric threads)
-
.22–95 filler wire of size 1.2 mm diameter and the base metal used was IS:2062 structural steel plate of 20 mm thickness. An
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 want to the same point of a point of a point of a point... like a human perspective, so that for every input, output, same point, same point, same point, same point, same point... i.e., the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
technology for COVID-19 diagnosis, vaccination, or therapy.⏎Xuexin Duan, School of Precision Instrument and Opto-electronics Engineering, Tianjin University, China⏎Weihua Guan, School of<|endoftext|>The flu is
-
1,* , Jingbiao Xu 2⏎Nanjing Xiaozhuang University Affiliated Primary School, Nanjing, Jiangsu 211171, China⏎2 Nanjing Xiaozhuang University, Nan
-
many parts of Tibet, with shops as well, where they were not seen previously.⏎Tibet's Overall Population⏎U'Tsang (Tibet Autonomous Region)⏎In the capital Lhasa, Tibetans offficially outnumber
-
the Department of Construction, Chaoyang University of Technology, Taichung 41349, Taiwan (
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
- L18:233 suppresses
- L21:525 excites
- L19:615 excites
- L18:1050 excites
- L20:395 suppresses
- L9:207 suppresses
- L18:607 excites
- L10:873 excites
- L3:130 excites
- L13:2796 suppresses
- L9:232 suppresses
- L21:158 excites
- L20:859 excites
- L7:92 weakly excites
- L4:42 weakly suppresses
- L4:177 weakly excites
- L21:1405 weakly suppresses
- L9:281 weakly excites
- L9:46 weakly suppresses
- L18:768 weakly excites
- L10:155 weakly excites
- L4:267 weakly excites
- L18:1721 weakly excites
- L5:46 weakly excites
Feeds later layers
- L27:2103 weakly excites
- L25:2829 weakly excites
- L25:1491 weakly excites
- L24:1932 weakly excites
- L25:2597 weakly excites
- L23:1869 weakly excites
- L23:1397 weakly excites
- L25:1537 weakly excites
- L27:1029 weakly excites
- L27:1439 weakly excites
- L24:2610 weakly suppresses
- L24:889 weakly excites
- L27:2749 weakly excites
- L27:2656 weakly excites
- L24:1567 weakly excites
- L23:1879 weakly suppresses
- L24:2461 weakly excites
- L26:365 weakly suppresses
- L24:3010 weakly suppresses
- L24:2565 weakly suppresses
- L24:214 weakly suppresses
- L23:241 weakly suppresses
- L27:2544 weakly suppresses
- L23:2792 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
- etter
- |`⏎
- 国度
- official
- bras
- OMX
- picture
- Official
- pictured
- racially
Suppresses
- utan
- rends
- �建
- entar
- alike
- ervation
- uto
- eterminate
- rightness
- 遞