Neuron Explorer

Qwen3-0.6B

acronym introductions

A neuron in layer 24 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 reacts to the spot right before a short form like an abbreviation or acronym is given, such as the opening bracket in "Happy Planet Index (HPI)" or the word "known as" before "RGGI". Turning it down or up did not produce any clear change in 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

    that we need new measures of progress. It is critical that these measures clearly reflect what we value – something the current approach fails to do.⏎The Happy Planet Index (HPI) measures what matters. It tells us how well nations are doing in

  2. #2 peak

    recent National Resources Defense Council analysis shows.⏎Maine and the five other New England states are among the nine states participating in the Regional Greenhouse Gas Initiative, known as RGGI. The nine would gain a total of $3.2 billion

  3. #3 peak

    to see that turning the remains of a sailing schooner more than a century old into a 65 ft tall interactive sculpture for the Museum of History and Industry (MOHAI) in Seattle, required a great deal of artistic vision. But

  4. #4 peak

    , which primarily covers individuals aged 65 and older, and Medicaid, designed to assist low-income individuals and families. Additionally, the Children’s Health Insurance Program (CHIP) offers coverage for eligible children from low-income families.⏎3. Key Termin

  5. #5 peak

    <|endoftext|>Appreciative inquiry is about looking for the best in people – in the way they work, they live, and they behave.⏎Initially, appreciative inquiry (AI) was a “fundamental shift in the overall perspective of organizational development that

  6. #6 peak

    Heating, Refrigeration and Air-Conditioning Engineers (ASHRAE)⏎Structural Engineers Association of Kansas & Missouri (SEAKM)⏎Structural Engineering Institute (SEI)⏎Illuminating Engineering Society (IES)⏎American Society of Plumbing Engineers

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: it is not a general probability distribution to it is not a general probability distribution it is not an indirium it it it it it it it it it it it it it it it it it it it it it it it it it it

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

    Queen Mary and superintendent of the Hampton Court garden. He published Phytographia in which he described and illustrated more than 2 700 rare

  2. #2 peak

    was conveyed to Cuellar, and interred in the church of Santa Marina, where his tomb is still to be seen.⏎His most famous work is the "Historia General de los Hechos de los Castellanos en las Islas y

  3. #3 peak

    disease of man. Nevertheless, the dead body of the victim may have exhibited some purplish discoloration. The term 'Black Death' was introduced into historiography some five hundred years after the tragic events of the mid fourteenth century

  4. #4 peak

    , his incisive and wide-ranging views on jazz and related musical forms, and his central place in the development and dissemination of jazz.⏎This article explores the historiography of early black music and its legacy. In examining texts from the

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. L23:255 suppresses
  2. L23:363 suppresses
  3. L23:1808 suppresses
  4. L23:383 weakly suppresses
  5. L23:404 weakly suppresses
  6. L22:2314 weakly suppresses
  7. L22:2855 weakly suppresses digits inside citation and DOI strings
  8. L20:2479 weakly suppresses
  9. L19:738 weakly excites
  10. L6:1706 weakly excites
  11. L4:1006 weakly suppresses
  12. L6:679 weakly excites
  13. L13:1097 weakly suppresses
  14. L22:3046 weakly excites
  15. L10:1136 weakly suppresses
  16. L17:246 weakly excites
  17. L15:135 weakly excites
  18. L4:1284 weakly suppresses
  19. L1:782 weakly excites
  20. L20:86 weakly excites
  21. L4:386 weakly suppresses
  22. L19:251 weakly excites
  23. L3:832 weakly excites
  24. L21:451 weakly suppresses

Feeds later layers

  1. L27:923 weakly suppresses
  2. L25:2421 weakly suppresses
  3. L26:1706 weakly suppresses
  4. L25:1625 weakly excites
  5. L26:2797 weakly suppresses
  6. L26:893 weakly suppresses
  7. L26:1645 weakly suppresses
  8. L27:419 weakly suppresses
  9. L26:1010 weakly suppresses
  10. L26:1854 weakly suppresses
  11. L25:1363 weakly excites
  12. L26:1101 weakly suppresses
  13. L27:1526 weakly excites
  14. L27:2903 weakly suppresses
  15. L26:2577 weakly excites
  16. L27:2176 weakly suppresses
  17. L25:2007 weakly suppresses
  18. L25:724 weakly suppresses
  19. L27:1973 weakly suppresses
  20. L25:2757 weakly suppresses
  21. L25:1919 weakly suppresses
  22. L27:878 weakly suppresses
  23. L27:3026 weakly suppresses
  24. L26:2705 weakly suppresses

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.

2 values between -18.73 and -17.761 values between -13.85 and -12.882 values between -11.90 and -10.931 values between -9.95 and -8.971 values between -7.02 and -6.041 values between -6.04 and -5.072 values between -5.07 and -4.097 values between -4.09 and -3.123 values between -3.12 and -2.1416 values between -2.14 and -1.16706 values between -1.16 and -0.193092 values between -0.19 and 0.7948 values between 0.79 and 1.776 values between 1.77 and 2.747 values between 2.74 and 3.721 values between 3.72 and 4.693 values between 4.69 and 5.672 values between 5.67 and 6.651 values between 9.58 and 10.551 values between 11.53 and 12.501 values between 12.50 and 13.481 values between 15.43 and 16.412 values between 18.36 and 19.341 values between 19.34 and 20.31offnegativepositiveoften

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. ingly
  2. uki
  3. akter
  4. DI
  5. OI
  6. чь
  7. Di
  8. agi
  9. step
  10. чи

Suppresses

  1. 不下
  2. iad
  3. iram
  4. 辗
  5. inium
  6. ients
  7. ières
  8. fram
  9. idend
  10. aments