closing parenthesis ending an aside
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. 13 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 closing bracket or dash that ends a short aside inside a sentence, such as '(usually)' or '(maximum)'. The steered text did not show a clear link to this pattern.
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.
-
LOOKUP (click here for more information).⏎What is a 2D lookup?⏎In many cases, you only look for a value in – let’s say – column A and return the
-
traffic with neither party incurring Internet Transit fees. We describe the process of peering, introduce (a lot of) terminology, and enumerate the top five motivations ISPs give for peering. The chapter
-
U.S. territory, but the Livermore atmospheric center is looking at how wind changes or rain could affect the spread of nuclear materials in Japan.⏎"(The center) has been activated to provide the U.S. government with a predictive modeling capability to
-
both of these Conventions are obliged to bring their national traffic laws in conformity with (one of) these Conventions. Both Conventions entail rules on, among others, traffic
-
the synergy of every team member.⏎n.大綱, 提要 [ summary or outline of a book, play, etc ]⏎note: 字根組合的概念是 "general view",可以用來
-
), data 0 (the library globals), and rloc 0 (the data relocation table). By contrast, an application if a resource database with (usually) five resources: code 0, code 1, data 0, r
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 user's query is in the blank, which is in the end (or else, which one, actually?...[...], as the...[...], which is in the end, and which one is actually the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
the key challenges facing education around the globe and to promote visionary thinking about new education models.⏎The summit was opened by Her Highness Sheikha Mozah Bint Nasser al-Missned, chair of the foundation, who said, “
-
hours before commanding Admiral Chuichi Nagumo began launching strike aircraft, the minesweeper USS Condor spotted a midget submarine outside the harbor entrance and alerted destroyer USS Ward. Ward was initially unsuccessful in locating the target. Hours later, Ward fired
-
marker. Battleship U.S.S. Ohio (approx<|endoftext|>“Snow White,” the beloved, first, full-length animated feature film
-
all the steps, you will see one shortcut on the desktop. Cut it and paste it at the following path⏎Now open Run (ctrl+R) and type your command name(Here my command name is “DOSBox“) and
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
- L12:2203 excites
- L10:1883 weakly excites closing parenthesis after a number or unit
- L8:1902 weakly excites
- L12:968 weakly suppresses 'of' after geographic nouns
- L12:2516 weakly suppresses
- L12:1398 weakly excites
- L8:2208 weakly excites
- L5:3016 weakly suppresses
- L12:72 weakly suppresses
- L5:539 weakly excites
- L10:3054 weakly suppresses
- L9:830 weakly suppresses
- L9:2352 weakly excites
- L4:1220 weakly suppresses
- L8:507 weakly suppresses
- L4:133 weakly excites
- L12:3036 weakly suppresses
- L0:2766 weakly suppresses
- L8:1847 weakly excites
- L10:1517 weakly excites
- L4:2132 weakly suppresses
- L10:1706 weakly suppresses
- L4:518 weakly suppresses
- L1:2530 weakly excites
Feeds later layers
- L16:2306 suppresses
- L18:1891 suppresses
- L15:1657 suppresses
- L14:1943 excites
- L19:891 weakly suppresses
- L24:473 weakly excites
- L15:1640 weakly suppresses
- L18:2889 weakly excites
- L18:1097 weakly suppresses
- L16:1856 weakly suppresses
- L16:2436 weakly suppresses
- L16:592 weakly suppresses
- L16:2024 weakly excites
- L20:564 weakly excites
- L14:44 weakly excites
- L24:168 weakly excites
- L21:2800 weakly excites
- L14:1410 weakly suppresses
- L15:485 weakly suppresses
- L27:1932 weakly excites
- L27:451 weakly excites
- L20:1178 weakly excites
- L24:1072 weakly suppresses
- L17:1186 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
- _:
- BASIS
- 第一天
- AML
- kea
- 的日
- 意義
- дать
- expectations
- 明天
Suppresses
- icut
- ifer
- crow
- ies
- ione
- CDATA
- utions
- berry
- istes
- ucha