capital e starting a word
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. 27 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 a capital E that begins a word or acronym, such as Eierbaum, Elysia, EPOCH, Emissivity or EBUS. The strongest negative values are on the letter e in words like Eating and Zero-Emission. Turning it up made the model repeat words like 'modern modern', which is unrelated to the letter.
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.
-
are called Ostereierbaum (literally: Easter egg tree) and a notable one, decorated since 1965, was the Saalfeld Eierbaum. 2015 was its last year open to the public
-
In] Sea Slug Forum. Australian Museum, Sydney. Available from http://www.seaslugforum.net/factsheet/elyscfcang⏎Re: Elysia from Gulf of Mexico⏎From: Kathe R. Jensen, October
-
if you have 1000 training examples, you can choose to pick 100 random samples (batch size=100) for each EPOCH (Training cycle)<|endoftext|>|Origin:||co
-
the groundwater thus there are significantly reduced mass transfer limitations compared to commonly used gas sparging (bubbling) techniques⏎Practical implications for use of the Waterloo Emitter
-
by manufacturer, window type, and age, but some basic information will get you a long way.⏎Please understand that the “e” stands for emissivity. Emissivity is the ability of a material to radiate energy. When heat or
-
global vaccine meeting in Nairobi, Kenya. AAP Global Immunization Project Advisory Committee Chair, Meg Fisher, MD, FAAP, served as faculty along with representatives from EAPA, KPA, Gavi, and others organizations. The meeting resulted in
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: why would previous least whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole whole
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
now Blaze and Logan are at the Smithsonian<|endoftext|>A recent study on high protein has been making the media rounds lately, with some news sites using headlines such as “Eating large amounts of meat, cheese may be as deadly as smoking.”⏎Note:
-
government incentive programs available for those who purchase these vehicles. For example, in May of 2019, the Government of Canada implemented a Canada-wide Zero-Emission Vehicle (ZEV) incentive, where customers can receive a rebate when they
-
, SpaceX launched a Dragon spacecraft from the Cape Canaveral Air Force Station. The successful flight demonstrated the components necessary to transport humans to low-Earth orbit. SpaceX expects to be ready to fly its first manned mission in
-
in the periodic table), the greater the shielding effect is. Essentially, the core electrons shield the valence electrons from the positive charge of the nucleus.⏎Electron-Electron Repulsions ⏎When two electrons are in the same shell, they will
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
- L22:2428 excites
- L18:221 excites
- L17:542 suppresses
- L21:435 excites
- L17:594 suppresses
- L19:645 suppresses
- L16:56 suppresses
- L2:1749 excites
- L17:1717 excites
- L6:401 weakly suppresses
- L16:829 weakly excites
- L10:2401 weakly excites
- L15:1272 weakly suppresses
- L16:248 weakly excites
- L18:139 weakly excites
- L16:2851 weakly excites
- L3:332 weakly suppresses
- L0:1291 weakly suppresses
- L15:2839 weakly excites
- L11:393 weakly excites
- L5:5 weakly excites digit after a label like paper or type
- L0:54 weakly excites
- L5:1231 weakly excites
- L15:1831 weakly excites
Feeds later layers
None: this is the last layer.
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
- pared
- gnore
- 每当
- mue
- 当之
- onNext
- 生
- elyn
- ئ
- 孰
Suppresses
- thic
- ighth
- VAL
- 終わり
- previous
- cob
- ╚
- held
- つか
- icles