Neuron Explorer

Qwen3-0.6B

unclear: weak mixed activations

A neuron in layer 13 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron fires only weakly and on a mixed bag of things, including grammar exercises, digits in lists and a few single words. Nothing clear ties these together. Turning it up produced repetitive or wandering text with no obvious theme.

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 activation was highest. Amber marks positive activation; the outlined token is the peak. Hover a token for its strength.

  1. #1 peak

    Sustainability 2013, 5(7), 3006-3008; doi:10.3390/su5073006⏎I am honored to introduce this

  2. #2 peak

    day, all three dogs were very hungry.⏎They started searching for food, but they couldn’t find anything. At last, they found a bone lying in the water.⏎They tried hard to get it but failed.⏎One of them thought that if

  3. #3 peak

    the house The house is cleaned by Helen Some people collect pictures Pictures are collected by some people The dog killed a ratA rat was killed by the dog The teacher will explain the story The story will be explained by the teacher⏎Examples Simple present:

  4. #4 peak

    , snatched off the order of St. George which he bore on his neck, and the Admiral, to show himself as great an enemy in adversity as he had been thought

  5. #5 peak

    Protecting Your Family and Loves Ones, No Matter What Happens⏎The program uses direct cash transfers for girls who remain unmarried and in school, but somehow the factories have had more of an impact.⏎How can this be?⏎“When

  6. #6 peak

    of various survey questions. Here we will go over what cross tabs are used for, how to set them up, and what the output will look like.⏎As mentioned, cross tabs are used for showing a side-by-side comparison of two or more

Steering

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.

turned down a lot
Prompt

Here is something I have been thinking about lately:................................................................................................................................................

Most negative on

Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.

  1. #1 peak

    , Sea Level, Renewable Energy, Natural Resources, Climate, Weather, Non-Renewable Energy, Land,

  2. #2 peak

    The answer to soaring gas prices lies in renewable heating, a technology that is ready to replace gas

  3. #3 peak

    00,000), with mortality rates twice as high for African Americans as Whites (4.4 versus 2.2 per 100,000).⏎These researchers

  4. #4 peak

    suggest that the symptoms of anxiety could exacerbate slow processing speed. For example, children with Obsessive Compulsive Disorder might appear to complete work slowly because they have developed internal rules such as needing to re-read an entire passage if they make a

Connections

Show on map

Which neurons in earlier layers feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.

Fed by earlier layers

  1. L12:673 excites
  2. L12:56 suppresses
  3. L12:583 excites
  4. L12:827 excites
  5. L11:990 excites
  6. L11:481 excites
  7. L10:1509 excites
  8. L12:730 excites
  9. L12:994 suppresses
  10. L9:536 excites
  11. L10:1873 suppresses
  12. L12:1567 excites
  13. L7:380 suppresses
  14. L11:569 weakly suppresses
  15. L12:2488 weakly excites
  16. L10:48 weakly suppresses
  17. L11:196 weakly suppresses
  18. L12:1346 weakly suppresses
  19. L12:744 weakly excites
  20. L11:928 weakly excites
  21. L12:184 weakly excites
  22. L1:1007 weakly excites name pieces inside a multi-part word
  23. L11:718 weakly suppresses
  24. L9:1524 weakly suppresses

Feeds later layers

  1. L14:991 strongly suppresses
  2. L15:1750 strongly excites
  3. L18:997 excites
  4. L18:397 excites
  5. L14:2280 excites
  6. L14:561 excites
  7. L15:1471 excites
  8. L14:1451 suppresses
  9. L15:2263 excites
  10. L15:567 suppresses
  11. L15:1027 excites
  12. L15:946 suppresses
  13. L15:1947 suppresses
  14. L14:978 excites
  15. L17:1414 suppresses
  16. L15:159 suppresses
  17. L16:866 suppresses
  18. L14:2827 suppresses
  19. L16:2101 excites
  20. L16:1904 excites
  21. L15:597 suppresses
  22. L15:2673 suppresses
  23. L15:2475 excites
  24. L15:341 suppresses

Activation histogram

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.

1 values between -1.57 and -1.521 values between -1.29 and -1.241 values between -1.24 and -1.181 values between -1.18 and -1.121 values between -1.07 and -1.014 values between -0.84 and -0.782 values between -0.78 and -0.735 values between -0.73 and -0.679 values between -0.67 and -0.619 values between -0.61 and -0.5618 values between -0.56 and -0.5023 values between -0.50 and -0.4428 values between -0.44 and -0.3952 values between -0.39 and -0.3351 values between -0.33 and -0.2768 values between -0.27 and -0.22122 values between -0.22 and -0.16212 values between -0.16 and -0.10357 values between -0.10 and -0.05863 values between -0.05 and 0.01762 values between 0.01 and 0.07485 values between 0.07 and 0.12322 values between 0.12 and 0.18206 values between 0.18 and 0.24137 values between 0.24 and 0.2985 values between 0.29 and 0.3535 values between 0.35 and 0.4030 values between 0.40 and 0.4612 values between 0.46 and 0.522 values between 0.52 and 0.573 values between 0.57 and 0.631 values between 0.63 and 0.69offnegativepositiveoften

Logit lens

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 layers instead.

Promotes

  1. 民主党
  2. igli
  3. тер
  4. -addons
  5. barang
  6. 对学生
  7. iente
  8. HING
  9. %)⏎⏎
  10. (png

Suppresses

  1. dff
  2. ells
  3. ators
  4. PC
  5. ight
  6. GBT
  7. too
  8. CP
  9. gates
  10. ifier