"D" in vitamin D and version digits
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. 8 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
Fires mostly on the letter D in "vitamin D", and also on digits in version numbers such as "Ubuntu 22.04" or "OS X 10.1". The steering results do not show a theme linked to vitamins, so the effect on writing is unclear.
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.
-
computer, this engaging CD makes learning the parts of speech fun and challenging for the entire class.⏎PC: Windows XP and up⏎Mac: OS X (10.1) and up⏎Resolution: 800 x 60
-
unzipping the .tgz files using the terminal. Let’s get into these.⏎Note: For this particular tutorial, we will use the Ubuntu 22.04. However, the commands used here are same for all the distributions
-
even 10% may be too much of a weekly mileage increase for some runners.⏎The causes of low bone density include genetics, poor dietary calcium or vitamin D intake, and a history of menstrual disorders, meaning a gap of six months or
-
working with resistanec bands or weights.⏎Diet: Calcium and Vitamin D are important for strong bones. High-calcium foods include:⏎- Dairy products
-
10.3390/nu8020068⏎- Calcium/Vitamin D Requirements, Recommended Foods & Supplements.” National Osteoporosis Foundation, https://
-
ake?⏎The amount of research about vitamin D and its relation to mood swings and depression is growing.⏎In addition to its well-known role in calcium absorption, vitamin D activates genes that regulate the immune system and release neurotransmitters (e.g., dopamine
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 am not a mathematician. (A) the number of my math... something / something / the number of my math... something / something / the number of my math... something / something / the number of my math... something /
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
), and the Piemonte-Toscana region of northern/central Italy (images only, February-March, 2009).⏎In Fort Yukon, Alaska, salmon is a mainstay of the diet. Yet, among the
-
nowhere to lay His head."⏎King James BibleAnd Jesus said unto him, Foxes have holes, and birds of the air [have] nests; but the Son of man hath not where to lay [his] head.⏎Parallel VersesInternational
-
is a remote and relatively undeveloped region in the Yukon that spans an area of 67,431 square kilometres. The Government of Yukon (“Yukon”) manages 97.3% of the Watershed
-
Jesus, but we need to consider why the Father sent His Son. God performed a miracle (the Incarnation) not as an end to itself
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
- L5:2552 weakly excites
- L7:57 weakly excites
- L7:1683 weakly suppresses
- L3:533 weakly excites
- L3:2335 weakly excites
- L7:954 weakly excites
- L7:1912 weakly excites
- L6:2583 weakly excites
- L3:707 weakly suppresses
- L1:924 weakly suppresses
- L2:1895 weakly suppresses
- L7:985 weakly excites
- L5:832 weakly excites
- L1:1103 weakly suppresses
- L1:1119 weakly suppresses
- L5:2587 weakly suppresses
- L5:2800 weakly suppresses
- L5:1426 weakly excites
- L6:797 weakly suppresses
- L2:1710 weakly suppresses
- L7:1873 weakly suppresses
- L5:1597 weakly excites
- L0:1413 weakly suppresses
- L2:2095 weakly suppresses
Feeds later layers
- L9:716 excites
- L10:952 suppresses
- L9:1026 excites
- L15:646 excites
- L13:1001 suppresses
- L10:1316 excites
- L10:2105 excites
- L11:122 suppresses
- L13:1513 excites
- L12:576 excites 'of' in fixed phrases like table of contents
- L9:1298 suppresses
- L13:450 suppresses
- L9:116 excites
- L9:845 excites
- L14:2288 excites
- L10:1705 excites
- L10:129 suppresses
- L11:393 excites
- L15:1299 suppresses
- L12:887 excites
- L9:401 excites
- L11:966 suppresses
- L13:2328 suppresses
- L17:870 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
- 且
- rib
- 檔案
- ,),⏎
- bery
- uhan
- tienes
- 一封信
- Ệ
- 点
Suppresses
- bidden
- ertainment
- imony
- 骝
- article
- ourt
- ewise
- -fashioned
- imageSize
- anness