leading digit of metric conversions
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 lights up on the first digit of a number given in metric units inside brackets, like the 2 in '87 hectares (217 acres)' or '83 square miles (210 km2)'. When it was turned up, the written text did not show any clear theme, so the steering tells us little about it.
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.
-
state of Tabasco which was almost unknown until a group of experts began excavating it three months ago.⏎The archaeological site, which is 87 hectares (217 acres) in size, is located adjacent to the ranching community of
-
Norman cathedral. It is the seat of the Bishop of Coutances and Avranches, previously the Bishops of Coutances. Standing 80 metres (295 ft) tall, it dominates the town and can be seen from as
-
710 km2), of which 576 square miles (1,490 km2) is land and 83 square miles (210 km2) (13%) is water. It is the second
-
town, and there were no organized churches.⏎According to the United States Census Bureau, the town has a total area of 97.3 square miles (252 km2), of which, 96
-
is 31.0" (787.4 mm). The month with the most precipitation on average is January with 8.0" (203.2 mm) of precipitation. The month with the least precipitation on
-
the distance down the lane from the<|endoftext|>Clare, New York⏎|Clare, New York|⏎|• Total||97.3 sq mi (251.9 km2)|⏎
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:... this action, my action, my action, my action, my action, my action, my,. this action. my. this. a. a. a. f. a. f. a . a. f
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
more complex sentences, main idea, narrative development, and inferencing.⏎- Martin Luther King Jr Holiday 5 Resources for AAC Learners⏎- Visual Supports, Core Words, and Martin Luther King⏎- Core Values: Suggestions for Language Building with
-
front grille. …⏎- Bumper End. Bumper ends are an important addition to the bumper itself. …⏎- Valance Panel. …⏎- Inner Fender. …⏎- Fender. …⏎- Cowl Panel.
-
12 nations: Fiji, Papua New Guinea, Vanuatu, Tonga, Samoa, Solomon Islands, Kiribati, Tuvalu, Nauru, Palau, Federated States
-
.S. portion of the Pacific Flyway alone.⏎Pacific Flyway States⏎Alaska | Arizona | California | Idaho | Nevada | Oregon | Utah | Washington | Colorado | Montana | New Mexico | Wyoming<|endoftext|>Please boost your Plan to download papers⏎
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
- L11:1633 strongly excites
- L11:689 excites
- L11:1659 excites
- L10:1271 suppresses
- L10:951 suppresses
- L10:715 excites
- L10:514 suppresses
- L10:1417 excites
- L11:1220 excites
- L11:5 excites
- L5:603 weakly suppresses
- L10:110 weakly suppresses
- L5:463 weakly excites
- L11:902 weakly excites
- L4:2176 weakly suppresses
- L5:374 weakly suppresses
- L11:4 weakly excites
- L1:2095 weakly excites
- L0:2093 weakly excites
- L1:1802 weakly excites
- L7:3024 weakly suppresses
- L7:282 weakly excites
- L10:761 weakly excites
- L8:462 weakly suppresses
Feeds later layers
- L13:1771 strongly suppresses
- L13:15 strongly excites
- L13:2215 strongly excites
- L14:2141 strongly suppresses
- L13:1113 strongly excites
- L13:1139 suppresses
- L16:1868 excites
- L13:2412 suppresses
- L13:2606 excites
- L13:2948 suppresses
- L13:2036 excites
- L17:670 excites
- L14:1752 excites
- L13:1757 excites
- L16:900 suppresses
- L14:1247 suppresses
- L13:2769 excites
- L13:674 weakly suppresses
- L22:2341 weakly suppresses
- L13:1104 weakly excites percentages and shares in statistics
- L14:2118 weakly suppresses
- L15:1682 weakly suppresses
- L14:2654 weakly excites
- L14:324 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
- 窝
- head
- ibel
- onto
- oids
- head
- arcy
- (head
- amento
- frage
Suppresses
- ioned
- .LatLng
- Fuse
- ),$
- prite
- nominee
- 使用者或
- '],['
- ripping
- renom