Neuron Explorer

Qwen3-0.6B

unclear: mid-word and mid-phrase fragments

A neuron in layer 16 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 is hard to pin down. Its strongest moments fall on pieces of words or phrases that are still unfinished, such as the start of a name split across tokens. Turning it up or down did not produce a recognisable 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

    to reaffix it, who, in turn, donated it to the public library. The scalp remained on display there for over 75 years, grossing out generations of Omahans. Now, the scalp is preserved in a dark room

  2. #2 peak

    ’t recognize the logic of that switch, review how fact families work by using smaller numbers: 5 = 9 – 4 means also that 4 = 9 – 5.⏎Made = 650⏎Sold = ?⏎Left

  3. #3 peak

    , he was offered a lucrative contract with the Yankees. Playing for the New York Yankees can bring a good deal of scrutiny and pressure. Under this pressure, Knoblauch began having problems with his throw to first base. Balls would sail over

  4. #4 peak

    . In fact, forces such as water, wind, and heat work to transform rocks into the soil as we know it over thousands of years.⏎Mixtures And Compounds⏎Before we understand whether soil is a heterogeneous mixture or not, it is

  5. #5 peak

    directress does her best to win over Wiliam by her show of affection and coquetry but he soon finds put that she is engaged to M. Pelet and returns her efforts with coolness and disdain. Wili

  6. #6 peak

    in molecule, it triggers a chemical reaction called a double proton transfer. This discovery raises the possibility that hypericin and similar molecules that are also activated by light could be used in therapies to treat AIDS, hepatitis, brain tumors, and

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: for, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and,

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

    remember yin and yang. If you use a light background, use dark or black colour for text. If you have a dark background, use white or light colour for text. The text needs to stand out from the background.⏎Also, when

  2. #2 peak

    well positioned in the photovoltaic⏎world. Instead of having one large inverter that handles all of the⏎current conversion from DC to AC, these new solar panels have a⏎solar micro inverter that takes the DC current generated by

  3. #3 peak

    , by the late 1970s, planetary researchers were receiving thousands of colour images from Mars revealing evidence of dry riverbeds hinting at the planet’s wetter, warmer history. These insights into the Martian past demonstrated the science

  4. #4 peak

    vote for Slavery in Oregon? Yes, or No.⏎And in the poll books shall be columns headed respectively.⏎“Constitution, Yes.” “Constitution, No"⏎“Slavery, Yes." “Slavery, No."⏎And the

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. L0:1301 excites
  4. L2:1321 excites
  5. L2:75 excites
  6. L2:49 excites
  7. L8:3 suppresses
  8. L1:1464 suppresses
  9. L0:1224 excites
  10. L1:2801 suppresses
  11. L6:17 excites
  12. L8:14 suppresses
  13. L8:16 excites
  14. L5:50 suppresses
  15. L8:4 suppresses
  16. L0:178 excites
  17. L6:7 excites
  18. L3:115 excites
  19. L5:46 suppresses
  20. L5:110 excites
  21. L6:18 suppresses
  22. L5:158 excites
  23. L6:8 excites
  24. L2:1710 excites

Feeds later layers

  1. L17:267 weakly excites
  2. L18:1268 weakly suppresses
  3. L17:873 weakly excites
  4. L17:1652 weakly excites
  5. L17:281 weakly excites
  6. L25:2575 weakly excites
  7. L17:174 weakly excites
  8. L18:1383 weakly suppresses
  9. L18:895 weakly excites
  10. L17:1237 weakly suppresses
  11. L17:1453 weakly excites
  12. L17:2058 weakly suppresses
  13. L19:1765 weakly excites
  14. L18:1050 weakly excites
  15. L17:1732 weakly excites
  16. L17:2029 weakly excites
  17. L18:146 weakly excites
  18. L26:490 weakly excites the digit 1 starting a number
  19. L21:415 weakly excites
  20. L24:821 weakly excites
  21. L17:1371 weakly excites
  22. L24:2191 weakly suppresses
  23. L18:425 weakly suppresses
  24. L19:909 weakly 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.

2 values between -4.64 and -4.361 values between -4.36 and -4.073 values between -4.07 and -3.799 values between -3.79 and -3.5011 values between -3.50 and -3.2210 values between -3.22 and -2.9412 values between -2.94 and -2.6520 values between -2.65 and -2.3736 values between -2.37 and -2.0851 values between -2.08 and -1.8070 values between -1.80 and -1.5183 values between -1.51 and -1.2399 values between -1.23 and -0.95140 values between -0.95 and -0.66175 values between -0.66 and -0.38212 values between -0.38 and -0.09277 values between -0.09 and 0.19243 values between 0.19 and 0.48304 values between 0.48 and 0.76289 values between 0.76 and 1.05270 values between 1.05 and 1.33309 values between 1.33 and 1.61268 values between 1.61 and 1.90214 values between 1.90 and 2.18203 values between 2.18 and 2.47172 values between 2.47 and 2.75103 values between 2.75 and 3.0499 values between 3.04 and 3.3280 values between 3.32 and 3.6048 values between 3.60 and 3.8937 values between 3.89 and 4.1728 values between 4.17 and 4.4611 values between 4.46 and 4.748 values between 4.74 and 5.035 values between 5.03 and 5.313 values between 5.31 and 5.591 values between 5.59 and 5.881 values between 5.88 and 6.161 values between 6.45 and 6.73offnegativepositiveoften

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. Câm
  2. idelity
  3. ählen
  4. 사이트
  5. ué
  6. IEL
  7. üß
  8. arel
  9. 𝛼
  10. -league

Suppresses

  1. 鼓
  2. izational
  3. �
  4. igr
  5. دو
  6. isation
  7. isations
  8. nr
  9. ising
  10. izations