Neuron Explorer

Qwen3-0.6B

't' after a hyphen in animal names

A neuron in layer 25 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 letter t right after a hyphen, most often in bird and animal names like red-tailed hawk or long-tailed manakin. The activations are weak compared with the other neurons here. Turning it up or down gave unrelated text.

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

    , and red tail. They have big, rounded wings and big bodies.⏎Red-tailed hawks often have a brown upper body and a white underside.⏎They are

  2. #2 peak

    of birds?⏎Of the two hundred and eighty migratory bird species in Palo Verde there is the scarlet macaw, the great curassow, the long-tailed manakin, the jabiru, hawks, ducks, the pic

  3. #3 peak

    species, as well as discuss where you may be able to see them in the state.⏎|Red Shoulder Hawk⏎Types of Hawks in Michigan⏎1. Red-tailed Hawk⏎In both the summer and winter months, the state of Michigan is

  4. #4 peak

    ats, since the males share the same red gorget as their East Coast cousins. The predominant species of hummingbirds in our valley, though, is the broad-tail. These birds look similar to the ruby-throats but are slightly larger

  5. #5 peak

    bed bug is their habitat. While the common<|endoftext|>448 pp., 6.125 x 9.25, 25 halftones, notes, bibl., index⏎Catholic Sisters

  6. #6 peak

    grid.⏎- Passive systems: The building design includes many passive features such as maximizing natural daylight and ventilation from the outdoors and, low-<|endoftext|>Five Questions about Long-Term Care⏎1. What is long-term care?⏎Long-term care refers to

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: I want to out out out out out out out out out out out out out out out out out out out out out out out out out out out out out out out out out out out out out out out out out out out out out

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

    of the availability of the more secure and complex WPA2, which became available in 2004 and is a common shorthand for the full IEEE 802.11i (or IEEE 802.11

  2. #2 peak

    is a common anode type⏎This code is for a common anode 7 segment display, the code was written using the mikroC PRO for 8051 compiler⏎P0 = 0xFF;⏎//activate segments

  3. #3 peak

    .⏎3. K.P.Soman, ShyamDiwakar, V.Ajay, “Insight into Data Mining Theory & Practice, Prentice Hall India,2012,⏎4. G.H.Gupta,

  4. #4 peak

    a Working Group was established.⏎At this time, Australia had the most gender-segregated workforce in the Organisation for Economic Co-operation and Development (OECD).⏎Affirmative Action (Equal Employment Opportunity for Women)

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. L5:2 suppresses
  2. L9:18 suppresses text boundary after a document break
  3. L0:9 weakly suppresses
  4. L11:18 weakly suppresses
  5. L9:114 weakly suppresses
  6. L21:295 weakly suppresses
  7. L6:1817 weakly excites
  8. L20:1360 weakly excites
  9. L13:2317 weakly excites
  10. L4:1059 weakly suppresses
  11. L20:1013 weakly suppresses
  12. L3:2890 weakly suppresses
  13. L10:1772 weakly excites
  14. L16:1693 weakly excites
  15. L21:1347 weakly suppresses
  16. L20:1711 weakly excites
  17. L15:1989 weakly excites
  18. L0:423 weakly excites
  19. L2:1809 weakly excites
  20. L19:738 weakly suppresses
  21. L17:280 weakly suppresses
  22. L3:2376 weakly suppresses
  23. L16:270 weakly suppresses
  24. L12:31 weakly suppresses

Feeds later layers

  1. L27:2512 weakly excites
  2. L27:125 weakly suppresses
  3. L26:2338 weakly excites
  4. L26:1489 weakly excites
  5. L26:618 weakly excites
  6. L26:1406 weakly suppresses
  7. L27:2316 weakly excites
  8. L27:2708 weakly suppresses
  9. L26:609 weakly suppresses
  10. L26:1755 weakly excites
  11. L26:1942 weakly suppresses
  12. L27:3016 weakly excites
  13. L26:1321 weakly suppresses
  14. L26:1313 weakly excites
  15. L27:285 weakly excites
  16. L27:592 weakly suppresses
  17. L27:2653 weakly suppresses
  18. L27:1735 weakly excites
  19. L27:850 weakly suppresses
  20. L26:2113 weakly suppresses
  21. L26:1468 weakly suppresses
  22. L26:3018 weakly excites
  23. L26:2190 weakly excites
  24. L26:911 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 -19.08 and -18.352 values between -16.89 and -16.161 values between -13.97 and -13.252 values between -13.25 and -12.521 values between -12.52 and -11.792 values between -11.79 and -11.063 values between -10.33 and -9.602 values between -9.60 and -8.874 values between -8.87 and -8.146 values between -8.14 and -7.419 values between -7.41 and -6.688 values between -6.68 and -5.957 values between -5.95 and -5.2313 values between -5.23 and -4.5019 values between -4.50 and -3.7735 values between -3.77 and -3.0469 values between -3.04 and -2.3184 values between -2.31 and -1.58182 values between -1.58 and -0.85886 values between -0.85 and -0.122252 values between -0.12 and 0.61230 values between 0.61 and 1.3442 values between 1.34 and 2.0720 values between 2.07 and 2.8011 values between 2.80 and 3.5210 values between 3.52 and 4.253 values between 4.98 and 5.711 values between 6.44 and 7.172 values between 7.17 and 7.901 values between 9.36 and 10.09offnegativepositiveoften

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. enaire
  2. irable
  3. lot
  4. ickle
  5. 一緒
  6. contracting
  7. (body
  8. ocol
  9. enha
  10. etal

Suppresses

  1. ph
  2. 法师
  3. 抜
  4. usk
  5. EMON
  6. ук
  7. keer
  8. ude
  9. دق
  10. _Native