Neuron Explorer

Qwen3-0.6B

'Holy' in names like Holy Land

A neuron in layer 23 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 'Holy' when it is part of a name, such as Holy Land, Holy Roman Empire or Holy Family. It is weakly negative on the word 'Great' in The Great Gatsby. Turning it up or down gave text that did not relate to religion or names.

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

    biggest fear is to see our brothers and sisters starving and without any assistance. Acts and decrees of the Second Plenary Council of the Philippines: held at the Holy Apostles Seminary, Makati, Metro Manila from 20 January-

  2. #2 peak

    . This traditional Mexican liturgy is celebrated throughout the Catholic Southwest during the last nine days of Advent, a kind of public ritual theater. People accompany statues of the Holy Family from house to house around the barrio, waiting to be recognized and allowed

  3. #3 peak

    see that the success of the Church⏎has largely been dependent on national factors. Contrast, for instance,⏎the history of the Holy Roman Empire with that of the British Empire.⏎Without the temporal power of this material

  4. #4 peak

    Zucchino, previously a reporter and foreign correspondent for the The Philadelphia Inquirer, The<|endoftext|>On November 8th, we commemorate the Synaxis of the Holy Archangels and all the Bodiless Powers. This is a special time to

  5. #5 peak

    near Bethlehem with his wife, Celeste,⏎and their six children. (They're expecting number seven.)⏎A cantor who is also active in the parish's Holy Name Society,⏎Behe still sees no conflict between faith and evolution, but is

  6. #6 peak

    ,” which became the seat of the theological faculty of Paris, the most illustrious seat of learning in the medieval period.⏎He led the Seventh Crusade to the Holy Land in 1248–50 and died on

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: when I was in high school, I was also going to go to the same school where I was also going to go to the same then when I was also going to go to the same then when I was also going to go to the

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

    etc.), appears in Greek and Latin satires from the late first century. In the sexuality of the ancient Romans, a tribas was a woman or hermaphrodite who actively penetrated another woman. Until

  2. #2 peak

    . Therefore, the author supports the vision that intense pursuit of success by the individual leads to their corruption and ultimately a more corrupt society. In the novel, The Great Gatsby, Fitzgerald has a pessimistic vision of America and his depiction is that

  3. #3 peak

    part decadence, excess, and disillusionment. Being wealthy is certainly one of the main accomplishments that characterized the American society. Through his characters, the novel The Great Gatsby by F. Scott Fitzgerald reveals the consequences of this dream on the population

  4. #4 peak

    to bring them all into existence on day four of Creation with the word of His mouth.⏎“By the word of the Lord the heavens were made, and all the host<|endoftext|>Oxcarbazepine prevents fits.⏎Take it regularly, twice

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. L14:2548 weakly excites
  2. L14:2288 weakly excites
  3. L13:484 weakly excites
  4. L20:361 weakly excites
  5. L19:427 weakly suppresses
  6. L9:2717 weakly suppresses
  7. L17:985 weakly excites
  8. L6:1458 weakly suppresses
  9. L7:1249 weakly excites
  10. L4:2714 weakly suppresses
  11. L17:1971 weakly excites
  12. L20:303 weakly suppresses
  13. L1:1913 weakly excites
  14. L0:1814 weakly excites
  15. L16:2385 weakly suppresses
  16. L17:1690 weakly excites
  17. L21:48 weakly suppresses
  18. L21:2748 weakly suppresses
  19. L13:790 weakly excites
  20. L2:70 weakly excites
  21. L9:1019 weakly excites
  22. L16:792 weakly excites "the" inside a prepositional phrase
  23. L19:2875 weakly suppresses
  24. L11:130 weakly excites

Feeds later layers

  1. L27:1099 weakly suppresses
  2. L27:2392 weakly suppresses
  3. L25:1682 weakly excites
  4. L27:754 weakly suppresses
  5. L24:628 weakly suppresses
  6. L25:2540 weakly suppresses
  7. L27:1005 weakly suppresses
  8. L25:662 weakly excites
  9. L24:473 weakly suppresses
  10. L24:961 weakly suppresses
  11. L24:2188 weakly suppresses
  12. L24:2222 weakly suppresses
  13. L26:2018 weakly suppresses
  14. L27:1865 weakly suppresses
  15. L25:1801 weakly suppresses
  16. L27:2346 weakly suppresses
  17. L25:2774 weakly suppresses
  18. L25:2482 weakly excites
  19. L24:1898 weakly excites
  20. L27:1541 weakly excites
  21. L27:1897 weakly suppresses
  22. L24:1201 weakly excites
  23. L24:2742 weakly excites
  24. L25:1101 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 -11.68 and -10.733 values between -8.84 and -7.893 values between -7.89 and -6.953 values between -6.95 and -6.0010 values between -6.00 and -5.0512 values between -5.05 and -4.1117 values between -4.11 and -3.1634 values between -3.16 and -2.2156 values between -2.21 and -1.27394 values between -1.27 and -0.323003 values between -0.32 and 0.63224 values between 0.63 and 1.5757 values between 1.57 and 2.5216 values between 2.52 and 3.4715 values between 3.47 and 4.4115 values between 4.41 and 5.3610 values between 5.36 and 6.316 values between 6.31 and 7.255 values between 7.25 and 8.206 values between 8.20 and 9.152 values between 9.15 and 10.093 values between 10.09 and 11.044 values between 11.04 and 11.991 values between 13.88 and 14.831 values between 14.83 and 15.771 values between 16.72 and 17.672 values between 17.67 and 18.611 values between 18.61 and 19.562 values between 21.45 and 22.401 values between 25.24 and 26.19offnegativepositiveoften

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. imon
  2. pose
  3. _FOUND
  4. 存在
  5. man
  6. 职
  7. ifter
  8. counter
  9. _stub
  10. 很

Suppresses

  1. undos
  2. OMP
  3. adiens
  4. beth
  5. 何时
  6. 也不例外
  7. rowave
  8. ()=="
  9. ��
  10. →glUniform