Neuron Explorer

Qwen3-0.6B

'to' or 'from' in 'from one X to another'

A neuron in layer 26 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the word 'to' or 'from' in phrases that describe moving between two things, such as 'from one device to another' or 'from one phase to another'. Turning it down made the model repeat the word 'another', which fits the topic.

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

  1. #1 peak

    VARIABLE v.This variable also points to the same obj. v created in method amethod().So now we have TWO REFERENCES pointing to same obj. 2)Inside method another , v.i=20; is execeuted.This means that

  2. #2 peak

    be exchanged for local products. Throughout history, cities have been founded at the intersections of transportation routes, or at points where goods must shift from one mode of transportation to another, as at river and ocean ports.⏎Religious elements have been crucial throughout

  3. #3 peak

    A lady From Your Nation and Dating a Guy by Another Countries⏎As you can see, the between online dating a girl through your country and dating a guy coming from another countries is great. It can be very confusing at first, but it’s

  4. #4 peak

    ⏎In his<|endoftext|>Refraction is the bending (change in the direction) of a wave (light or sound). When a light ray passes from one transparent medium to another, it changes speed and bends.⏎- Light traveling from air and going into

  5. #5 peak

    etc.--there's no simple way to transfer "clippings" of data from one device to another. But a new research project at MIT called Sparsh is aiming to fix

  6. #6 peak

    relationship between temperature and/or duration of extraction versus the actual amount of it you get out once you’re done.⏎The relatively easy mass transport of caffeine from one phase to another also plays a factor in its ability to be absorbed into the human bloodstream in

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: another又又又又又又又又又又又又又又又又又又又又又又又又又又又又又又又又又又又又又又又又又又又又又又又

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

    both reef fishes and macro-invertebrates (conch, lobster, Diadema); 2) To relate this information to in-situ data collected

  2. #2 peak

    Diatomaceous Earth is a naturally occurring siliceous sedimentary mineral compound, made from microscopic fossilised remains of algae-like plants called Diatoms.⏎Diatoms occur in virtually every environment that contains water. This includes

  3. #3 peak

    assays to look for toxicity in surface waters and sediments. Tests on ambient surface waters are conducted using the "water flea" (a crustacean), Ceriodaphnia dubia, to identify toxic effects on survival and reproduction. Col

  4. #4 peak

    female Salii. Another military dance was the Saltatio bellicrepa, said to have been instituted by Romulus in commemoration of the Rape of the Sabines. The Pyrrhic dance (fig. 13) was also

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. L21:724 strongly suppresses
  2. L20:2919 strongly suppresses
  3. L19:1632 strongly excites
  4. L22:1632 excites
  5. L23:371 excites
  6. L17:1900 excites
  7. L17:861 excites
  8. L16:2122 suppresses
  9. L21:2124 excites
  10. L17:2158 suppresses
  11. L21:2473 excites
  12. L18:3044 suppresses
  13. L18:2772 excites
  14. L21:2296 excites
  15. L16:1733 excites
  16. L21:2198 excites
  17. L21:784 suppresses
  18. L19:3036 excites
  19. L20:2636 excites
  20. L19:775 suppresses
  21. L19:2162 excites
  22. L19:1352 excites
  23. L19:1050 suppresses
  24. L24:92 suppresses

Feeds later layers

  1. L27:2627 suppresses
  2. L27:655 weakly suppresses
  3. L27:961 weakly suppresses
  4. L27:2970 weakly suppresses
  5. L27:674 weakly excites
  6. L27:1905 weakly excites
  7. L27:873 weakly suppresses
  8. L27:1821 weakly excites
  9. L27:2473 weakly excites
  10. L27:1381 weakly suppresses
  11. L27:2427 weakly suppresses
  12. L27:1973 weakly excites
  13. L27:2810 weakly excites
  14. L27:2996 weakly excites
  15. L27:2175 weakly excites
  16. L27:450 weakly excites
  17. L27:1078 weakly excites
  18. L27:2812 weakly excites
  19. L27:525 weakly excites
  20. L27:2137 weakly excites
  21. L27:2100 weakly excites
  22. L27:1692 weakly excites
  23. L27:2869 weakly excites
  24. L27:1181 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 -14.56 and -13.631 values between -9.93 and -9.001 values between -9.00 and -8.081 values between -7.15 and -6.221 values between -6.22 and -5.302 values between -5.30 and -4.375 values between -4.37 and -3.454 values between -3.45 and -2.526 values between -2.52 and -1.6059 values between -1.60 and -0.673203 values between -0.67 and 0.26595 values between 0.26 and 1.1825 values between 1.18 and 2.111 values between 3.03 and 3.961 values between 5.81 and 6.731 values between 8.59 and 9.511 values between 21.54 and 22.47offnegativepositiveoften

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. �
  2. CHED
  3. <article
  4. vrier
  5. 勰
  6. semble
  7. ולים
  8. ставить
  9. смотреть
  10. крут

Suppresses

  1. 又
  2. 又要
  3. 又有
  4. another
  5. another
  6. 又是
  7. 他又
  8. 又能
  9. Another
  10. 又被