unclear: mixed, Chinese and name fragments
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. 25 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 scattered set of pieces, including a Chinese character, a syllable inside a name, and a word fragment in a plant or holiday name. Turning it down made the model repeat the Chinese character for 'have' over and over.
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.
-
已发表论文 ⏎“助动词”在汉语中具有较大的作用,使用率高,但是泰国学生经常在学习和使用汉语助动词中产生各种的偏误。本文通过调查法、统计
-
the former closely resembles that of the latter told in Târanâtha. Târanâtha states that when Açvaghosha became a sthavira and advocate of the⏎[paragraph continues] Tripitaka, he
-
id ul-Azha Day 2||Feast of Sacrifice|⏎|Saturday||September 30||Ashura||10th day of Muharram|⏎|Friday||November 03||Urs of Shah Abdul
-
a USDA report. "The pollinators will thank you."⏎Looking for some ideas? Check out these flowering plants that can help give bees a boost.⏎Crocus are a good choice to attract bees in the early spring. They're also
-
s and quizzes, 5269 literature essays, 1584 sample college application essays, 204 lesson plans, and ad-free surfing in this premium content, “Members Only” section of the site! Membership includes
-
properties, points of interest and hindrances.⏎Glass is certainly the most widely recognized type of packaging waste. Be that as it may, glass holders are oftentimes supplanted with plastic ones, in light of the fact that dissimilar to glass,
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:有有有有有有有有有有有有有有有有有有有有有有有有有有有有有有有有有有有有有有有有有有有有有有有有
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, until 1865. Other well known names were John Pringle, John Hogg, Charles Murray, John Scott, Charles Sheddon and David Greig the elder. But already there were two factors which between them were to kill
-
other tragic events the general, underlying message of the decade was one of peace, love and hope.⏎In 1957, Jerry Landis and Tom Grath had a moderate hit with their song “Hey Schoolgirl”. The duo’s
-
Indus Valley Civilisation of Harappa and Mohenjo-daro from the Bronze Age. Nowadays, the Thar Desert is
-
no epidemic diseases.⏎“No one got the flu, not even the common cold, until about 10,000 years ago,” said Dr. Michael Greger, author of Bird Flu: A Virus of Our Own Hatching.⏎What
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
- L5:621 excites
- L19:89 excites
- L16:3 suppresses
- L16:36 excites
- L2:761 suppresses
- L18:853 suppresses
- L1:259 suppresses
- L16:83 excites
- L23:1408 excites
- L5:334 excites
- L16:1 excites
- L0:2215 suppresses
- L20:61 suppresses
- L16:110 excites
- L20:2984 excites
- L2:1091 suppresses
- L16:143 excites
- L21:1809 excites
- L16:493 excites
- L20:1852 weakly suppresses
- L4:345 weakly excites
- L16:0 weakly suppresses
- L20:78 weakly suppresses
- L0:2090 weakly suppresses
Feeds later layers
- L27:756 strongly suppresses
- L27:1350 weakly suppresses
- L27:1307 weakly excites
- L27:809 weakly suppresses
- L27:225 weakly suppresses
- L27:464 weakly suppresses
- L26:550 weakly excites
- L26:1776 weakly suppresses
- L27:2479 weakly suppresses
- L27:2809 weakly suppresses
- L27:160 weakly suppresses
- L26:1914 weakly suppresses
- L27:2419 weakly suppresses
- L27:561 weakly suppresses
- L26:891 weakly suppresses
- L26:751 weakly suppresses
- L27:721 weakly suppresses
- L27:1596 weakly suppresses
- L26:2620 weakly suppresses
- L27:710 weakly suppresses
- L26:2529 weakly suppresses
- L27:185 weakly suppresses
- L27:2286 weakly excites
- L27:1982 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
- tó
- s
- าว
- t
- in
- uo
- dict
- inant
- lec
- grips
Suppresses
- 有
- 有这样的
- 有大量的
- 有不同的
- 有足够的
- have
- (has
- 有能力
- 有更多的
- Have