Neuron Explorer

Qwen3-0.6B

name fragments inside split words

A neuron in layer 26 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 weakly on a piece of a word that has been chopped up, mostly inside proper names, such as "ust" in "Eustis", "igr" in "Tigris" or "Car" in "Carpathia". The examples are varied and the signal is weak, so the pattern is not very sharp. Steering changes did not reveal a clear 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

    every titan finds its challenger eventually. This one just happened to appear in the form of a bee swarm.⏎The drama unfolded Thursday at the Joint Base Langley-Eustis in Virginia, when an F-22 aircraft from the 19

  2. #2 peak

    of different kingdoms spread out through the land between Anatolia (modern Turkey) and Babylon (modern Iraq).⏎Mitanni took brief control of the region in the Tigris and Euphrates basin between the fall of Amorite Babylon to

  3. #3 peak

    formed in 1880 from part of the Town of Peirrepont, was the last town formed in the county. Madam De Stael, a French author, at one time invested in the town lands, on

  4. #4 peak

    before you can see the other students’ posts. Submit your initial post(s) early in the module.⏎See the Discussion Board Guidelines and Rubrics⏎Download Discussion Board Guidelines and Rubrics.⏎Reh, F. J. (201

  5. #5 peak

    journalistic institution, Adafruit strives to be a fair, informative, and positive voice within the community – check it out here: adafruit.com/editorialstandards⏎Stop breadboarding and soldering – start making immediately! Adafruit

  6. #6 peak

    This is one of Atlanta’s roughest areas.⏎This mansion in the Vine neighborhood, built in 1910, was the home of Alonzo Herndon, who was born into slavery but went on to become Atlanta's first black

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: option option option option option option option option option option option option option option option option option option option option option option option option option option option option option option option option option option option option option option option option option option option option option option option option

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

    earlier version, you should certainly upgrade; this version fixes quite a few problems, including handling of certain kinds of global variables.⏎The first step in creating a GLib shared library is to create a ``static'' library (also called an ar archive

  2. #2 peak

    Acceptable is a client error that is returned by the server to indicate that the resource exists but the proactive Content Negotiation engaged in by the client using a relevant HTTP request header was not satisfiable and the server does not supply

  3. #3 peak

    treatment. Prognostic and predictive factors are often discussed together. They both play a part in deciding on a treatment plan and a prognosis.⏎The following are prognostic factors for bile duct cancer

  4. #4 peak

    show that arable reversion is an effective tool for the management of butterflycommunities. We suggest that as restoration takes time to achieve, its use as a mitigation tool against future environmental change (i.e. by decreasing isolation in fragmented landscapes)

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. L20:2117 excites
  2. L22:739 suppresses 'are' after 'here' introducing a list
  3. L21:913 suppresses
  4. L19:2379 suppresses
  5. L25:399 suppresses
  6. L20:928 excites
  7. L22:186 excites
  8. L25:320 excites
  9. L25:374 suppresses
  10. L18:872 weakly excites the digit 8 in lists and numbers
  11. L12:1507 weakly excites
  12. L17:1172 weakly suppresses
  13. L22:426 weakly excites
  14. L4:2743 weakly excites
  15. L19:2149 weakly excites
  16. L18:2147 weakly suppresses
  17. L3:1331 weakly suppresses
  18. L19:2305 weakly suppresses
  19. L10:641 weakly suppresses
  20. L25:437 weakly excites
  21. L19:2122 weakly suppresses
  22. L24:451 weakly suppresses
  23. L25:390 weakly suppresses
  24. L6:821 weakly excites

Feeds later layers

  1. L27:551 suppresses
  2. L27:814 suppresses
  3. L27:588 suppresses
  4. L27:1035 excites
  5. L27:488 weakly suppresses
  6. L27:1181 weakly excites
  7. L27:480 weakly excites
  8. L27:592 weakly suppresses
  9. L27:1218 weakly excites
  10. L27:533 weakly suppresses
  11. L27:2939 weakly suppresses
  12. L27:584 weakly suppresses
  13. L27:250 weakly suppresses
  14. L27:602 weakly suppresses
  15. L27:1313 weakly suppresses
  16. L27:403 weakly suppresses
  17. L27:2567 weakly suppresses
  18. L27:963 weakly suppresses
  19. L27:2909 weakly excites
  20. L27:2806 weakly suppresses
  21. L27:1927 weakly suppresses
  22. L27:2387 weakly excites lowercase m opening a word fragment
  23. L27:1405 weakly excites
  24. L27:1186 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 -17.91 and -17.131 values between -15.57 and -14.792 values between -10.89 and -10.122 values between -10.12 and -9.341 values between -9.34 and -8.563 values between -7.78 and -7.009 values between -7.00 and -6.226 values between -6.22 and -5.446 values between -5.44 and -4.6614 values between -4.66 and -3.8824 values between -3.88 and -3.1031 values between -3.10 and -2.3264 values between -2.32 and -1.55188 values between -1.55 and -0.771891 values between -0.77 and 0.011179 values between 0.01 and 0.79225 values between 0.79 and 1.5799 values between 1.57 and 2.3554 values between 2.35 and 3.1334 values between 3.13 and 3.9121 values between 3.91 and 4.6916 values between 4.69 and 5.4711 values between 5.47 and 6.257 values between 6.25 and 7.032 values between 7.03 and 7.804 values between 7.80 and 8.585 values between 8.58 and 9.365 values between 9.36 and 10.142 values between 10.14 and 10.921 values between 12.48 and 13.26offnegativepositiveoften

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. ections
  2. owi
  3. Methods
  4. ddb
  5. 各项工作
  6. Methods
  7. ewan
  8. methods
  9. 毯
  10. behaviors

Suppresses

  1. option
  2. (option
  3. _option
  4. requirement
  5. option
  6. property
  7. Option
  8. limitation
  9. principle
  10. capability