Neuron Explorer

Qwen3-0.6B

the word 'panel' and its relatives

A neuron in layer 3 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 lights up on the word 'panel' (and pieces of similar words like 'panelists'). Turning it up didn't produce anything clearly related to panels, so the effect on the written text is hard to read.

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

    Published: 09/28/2006 - Updated: 03/19/2019⏎Author: Casa Taller Saluz⏎Establishing good habits during childhood decreases the possibility

  2. #2 peak

    published in 1966.⏎Yet, though many of the panellists called it a “perfect” novel, “Wide Sargasso Sea”

  3. #3 peak

    Choset, robotics professor at CMU, speaking on the panel. The university recently launched an artificial intelligence major rather than opting to revamp the

  4. #4 peak

    Of Relays⏎Transtutors is the best place to get answers to all your doubts regarding classification of relays with examples. Transtutors has a vast panel of experienced in classification of relays electrical engineering tutors who can explain the

  5. #5 peak

    hour long debate. The panelists argued amongst themselves about what exactly a service or a microservice means. By

  6. #6 peak

    ’s 200th birthday, and the 10th anniversary of the premier Caribbean literary festival, the NGC Bocas Lit Fest, a panel that included American writer Lauren Elkin, British novelist Linda Grant, and Trinidad

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 am the 18th of May, 2023. So, the user is the first day of January, 2023, and the rest of the month is the last day of January, 2

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

    et au Parlement sur l’évolution des charges et produits de l’Assurance Maladie au titre de 2014 (loi du 13 août 20

  2. #2 peak

    iations are not used in any Polish sources, The standard one was single-turret variant, and could be called just: "7TP".<|endoftext|>Art. XXIX.—The Older Gravels of North Canterbury.⏎[Read before the New Zealand

  3. #3 peak

    ’re concerned about authorized principle. It’s adopted by the Industrial Revolution (XVIII century) and the golden age of British education throughout the Victorian era (XIX century).⏎With the information that training, and particularly science schooling, is an endeavor

  4. #4 peak

    chervil blooms from mid spring to early summer. It is an herbaceous biennial or short-lived perennial plant in the family Apiaceae (Umbelliferae), plants that bear their flowers in umbrella-like clusters.⏎The related

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. L0:1224 weakly suppresses
  2. L1:625 weakly suppresses
  3. L1:956 weakly suppresses
  4. L2:111 weakly suppresses
  5. L1:1665 weakly suppresses
  6. L0:2008 weakly excites
  7. L1:651 weakly suppresses
  8. L0:2233 weakly suppresses
  9. L2:75 weakly suppresses
  10. L2:2240 weakly suppresses
  11. L2:2912 weakly excites
  12. L2:1899 weakly excites
  13. L1:2284 weakly suppresses
  14. L2:2242 weakly suppresses
  15. L0:1301 weakly suppresses
  16. L1:2124 weakly excites
  17. L2:614 weakly excites
  18. L2:2519 weakly excites
  19. L0:1080 weakly excites
  20. L2:2970 weakly excites
  21. L0:815 weakly excites
  22. L1:2420 weakly suppresses
  23. L1:1499 weakly suppresses
  24. L0:2180 weakly suppresses

Feeds later layers

  1. L4:2984 suppresses
  2. L22:298 suppresses
  3. L6:890 weakly suppresses
  4. L5:906 weakly excites
  5. L5:2883 weakly excites
  6. L13:1894 weakly excites
  7. L11:2907 weakly excites
  8. L4:2364 weakly suppresses
  9. L12:1893 weakly excites
  10. L6:486 weakly suppresses
  11. L4:1209 weakly suppresses
  12. L4:2349 weakly excites
  13. L12:991 weakly excites
  14. L6:2373 weakly excites
  15. L24:2526 weakly excites
  16. L4:3050 weakly excites
  17. L25:1796 weakly excites
  18. L6:448 weakly excites
  19. L7:1563 weakly suppresses
  20. L23:1345 weakly excites
  21. L4:2497 weakly suppresses
  22. L10:1755 weakly excites
  23. L4:2743 weakly suppresses
  24. L5:2416 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 -0.66 and -0.571 values between -0.57 and -0.485 values between -0.48 and -0.3925 values between -0.39 and -0.3195 values between -0.31 and -0.22348 values between -0.22 and -0.131042 values between -0.13 and -0.041572 values between -0.04 and 0.05514 values between 0.05 and 0.14209 values between 0.14 and 0.2362 values between 0.23 and 0.3222 values between 0.32 and 0.415 values between 0.41 and 0.503 values between 0.50 and 0.591 values between 0.59 and 0.671 values between 0.94 and 1.031 values between 1.48 and 1.561 values between 2.81 and 2.90offnegativepositiveoften

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. 主页
  2. vette
  3. 首页
  4. issy
  5. .@
  6. erten
  7. @mail
  8. .lastname
  9. ıcı
  10. iron

Suppresses

  1. …”⏎⏎
  2. ouncing
  3. profund
  4. deja
  5. яз
  6. áo
  7. …⏎⏎
  8. ISCO
  9. 偿
  10. dred