Neuron Explorer

Qwen3-0.6B

unclear: mid-phrase words, mixed

A neuron in layer 12 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 on a grab bag of words that sit in the middle of a phrase, such as 'Placement', 'behind', 'succeed' and 'after'. No single theme stands out. When it was turned up, the model's writing drifted into unrelated rambling about math and reasoning.

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

    websites and started applications. It is also measuring time in front of the monitor.<|endoftext|>DRDO Placement Paper (EMT & Microprocessor)⏎Paper: DRDO Placement Paper (EMT & Microprocessor)⏎EMT and microprocessor:⏎Make sure

  2. #2 peak

    in “Internet Explorer” into the bar at the bottom left of the screen and click on it. It should open up after clicking.⏎· Type HydroShare.org into the browser and click on it – it should open up.⏎·

  3. #3 peak

    central and southern Andes for its edible root. In Peru and Bolivia, for example, this tuberous root vegetable is the second most widely grown root crop behind the potato. In these countries, the

  4. #4 peak

    agreed-on method for creating citations. It provides formats for citing differnet information resources within the body of the paper, for listing different information types on the Works Cited page, and even ways to set up headings and margins for the paper. There

  5. #5 peak

    confederation by distributing the major cities among his sons; the eldest was to be grand prince of Kiev, and the brothers were to succeed each other, moving up the hierarchy of cities toward Kiev, filling vacancies left by

  6. #6 peak

    a way to land the probe safely on the celestial body.⏎When Philae reaches the surface of the comet, the lander will harpoon itself into place

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: the following is a summary of the first and last paragraph, but it's not a math problem. (The first and last paragraph. The first and last. I need not understand the meaning. But that is. The first and last.

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

    helps to boost fertility.⏎- Avocado plays an important role in women’s health. It helps in estrogen production and reproductive hormonal balance.⏎- Nutrients like folic acid, potassium, fiber and other minerals make it super food for good health.⏎

  2. #2 peak

    on family size and future increase in the overall population. In addition, a mother’s age, whether younger or older has an effect on outcomes of births including birth weight, multiple births, and birth

  3. #3 peak

    , level surface. From one or more of these vantage points, you should draw an imaginary plumb line from the top of the leg to the ground. If you cannot draw a

  4. #4 peak

    begin, both mosquito and microbial small and large subunit rRNA fragments are amplified from a metagenomic community DNA sample. Then, the community specific biotinylated antisense ribosomal RNA probes are synthesized in vitro using T7 RNA

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. L8:7 suppresses
  2. L6:13 suppresses
  3. L8:16 excites
  4. L8:10 weakly suppresses
  5. L11:163 weakly excites
  6. L8:14 weakly suppresses
  7. L8:1319 weakly suppresses
  8. L6:8 weakly excites
  9. L6:12 weakly suppresses
  10. L8:12 weakly suppresses
  11. L10:535 weakly excites
  12. L7:303 weakly excites
  13. L9:1157 weakly suppresses
  14. L6:16 weakly suppresses
  15. L5:158 weakly excites
  16. L9:1766 weakly suppresses
  17. L3:1662 weakly excites
  18. L5:133 weakly suppresses
  19. L7:49 weakly excites
  20. L9:665 weakly suppresses
  21. L8:59 weakly excites
  22. L6:17 weakly excites
  23. L8:4 weakly suppresses
  24. L10:1373 weakly suppresses

Feeds later layers

  1. L14:585 excites comma opening an inserted clause
  2. L13:2153 excites
  3. L14:1951 weakly excites
  4. L13:803 weakly excites
  5. L17:2082 weakly suppresses
  6. L14:2192 weakly excites
  7. L15:3027 weakly excites
  8. L15:1381 weakly suppresses
  9. L21:248 weakly suppresses
  10. L16:2742 weakly excites end-of-document token after a finished text
  11. L16:1251 weakly suppresses
  12. L17:2673 weakly excites
  13. L15:270 weakly excites
  14. L14:2119 weakly excites
  15. L14:1618 weakly excites
  16. L15:1787 weakly excites
  17. L14:2203 weakly excites
  18. L16:1436 weakly suppresses
  19. L17:2567 weakly suppresses
  20. L23:189 weakly suppresses
  21. L20:1201 weakly excites
  22. L16:1449 weakly suppresses
  23. L18:1501 weakly suppresses
  24. L16:3027 weakly excites

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 -3.28 and -3.141 values between -2.73 and -2.592 values between -2.32 and -2.181 values between -2.18 and -2.042 values between -2.04 and -1.901 values between -1.90 and -1.774 values between -1.77 and -1.636 values between -1.63 and -1.499 values between -1.49 and -1.3614 values between -1.36 and -1.2227 values between -1.22 and -1.0843 values between -1.08 and -0.9470 values between -0.94 and -0.81102 values between -0.81 and -0.67182 values between -0.67 and -0.53213 values between -0.53 and -0.39315 values between -0.39 and -0.26412 values between -0.26 and -0.12529 values between -0.12 and 0.02517 values between 0.02 and 0.16403 values between 0.16 and 0.29313 values between 0.29 and 0.43237 values between 0.43 and 0.57163 values between 0.57 and 0.7098 values between 0.70 and 0.84101 values between 0.84 and 0.9839 values between 0.98 and 1.1236 values between 1.12 and 1.2524 values between 1.25 and 1.3916 values between 1.39 and 1.539 values between 1.53 and 1.677 values between 1.67 and 1.804 values between 1.80 and 1.944 values between 1.94 and 2.083 values between 2.08 and 2.22offnegativepositiveoften

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. erus
  2. shedding
  3. isure
  4. ett
  5. ỵ
  6. "profile
  7. itches
  8. :block
  9. chod
  10. udes

Suppresses

  1. antic
  2. 麦克
  3. assic
  4. нем
  5. енд
  6. 分
  7. Indo
  8. ней
  9. 绿城
  10. ает