end of document boundary
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. 21 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 responds to the invisible marker the training data uses to show that one web page has ended and another has begun. It fires on the spot right before a new, unrelated text starts. Turning it up did not produce a clear theme in the model's writing.
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.
-
Reducer coupling nuts offer the ability to couple different rod sizes. Threaded rods are often
-
locate.<|endoftext|>Science Fair Project Encyclopedia⏎The World Factbook⏎The World Factbook is an
-
the<|endoftext|>Nature has given us all the possible amenities to keep ourselves healthy. Interestingly, our
-
the<|endoftext|>This Article explores the legacy of the U.S. environmental justice movement for the U
-
it was celebrated for a<|endoftext|>Learn something new every day⏎More Info... by email⏎A quesadilla
-
would pass away before<|endoftext|>The Badrouzi Festival or “Kajin” was held in ancient Iran
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: when you think of the natural forerites, which is the general public's natural or naturalistic, but when you think of the social forerites, which is the general public's social or socialish, but when you think of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
are several shell recycling drop-off sites across the coast including at the North Carolina Coastal Federation located at 3609 Hwy 24, Ocean, NC 28570 and at Hammocks Beach State Park located at 1
-
015 The University of North Carolina Press⏎116 South Boundary Street, Chapel Hill, NC 27514-3808⏎How to Order | Make
-
Planetarium (who's internet⏎presence is limited to⏎StarHustler at the moment) usually has set up an event⏎hotline (Phone: 305-854-2222) and features shows
-
ooteka (John Jolly) of the Cherokee, negotiated the January 14, 1836, treaty between Chief Bowl of the Cherokee and the Republic of Texas.
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
- L20:431 excites
- L18:363 excites
- L16:1023 excites
- L3:46 suppresses
- L3:500 suppresses
- L8:407 excites
- L20:508 excites
- L16:1295 excites
- L20:1402 weakly suppresses
- L18:1203 weakly excites end of document marker
- L15:1165 weakly suppresses
- L2:1152 weakly excites
- L18:529 weakly excites
- L6:2570 weakly excites
- L4:2084 weakly excites
- L20:1829 weakly suppresses
- L4:1700 weakly suppresses conservation laws and invariance
- L14:1495 weakly suppresses
- L9:2169 weakly excites
- L20:85 weakly suppresses
- L20:1053 weakly suppresses
- L19:989 weakly excites
- L20:2663 weakly suppresses
- L9:114 weakly suppresses
Feeds later layers
- L25:465 weakly suppresses
- L26:2004 weakly suppresses
- L24:2919 weakly suppresses
- L22:1905 weakly excites
- L23:1539 weakly excites
- L27:1436 weakly excites
- L22:2564 weakly suppresses perpendicular and geometry terms
- L23:1648 weakly suppresses
- L23:2704 weakly suppresses
- L23:232 weakly excites
- L24:1450 weakly suppresses
- L24:1979 weakly suppresses
- L22:1790 weakly excites
- L25:2883 weakly suppresses
- L24:1511 weakly suppresses
- L24:1060 weakly suppresses
- L23:618 weakly suppresses
- L22:2670 weakly suppresses
- L23:2728 weakly suppresses second digit of years like 2020
- L22:499 weakly excites
- L23:907 weakly suppresses
- L25:1544 weakly suppresses
- L24:2146 weakly excites
- L24:1847 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
- const
- const
- →const
- ({...
- 스크
- cry
- plemented
- ,const
- interface
- <const
Suppresses
- .Connect
- ?⏎
- ije
- ?⏎
- ||⏎
- hormone
- ?
- 吗
- ?
- otional