Neuron Explorer

Qwen3-0.6B

opening parenthesis before an acronym

A neuron in layer 26 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

It fires on the open bracket that comes right after a full name, just before its short form, as in 'Civilian Conservation Corps (CCC)' or 'Individualized Program Plan (IPP)'. Turning it up did not produce anything obviously related, and the output mostly became repetitive or drifted off topic.

Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens an email to the maker with the neuron already filled in).

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

    with your Pediatrician⏎“Parents should express any concerns about symptoms seen in their children to their pediatrician who will<|endoftext|>A novel logic concept, Race Logic Architecture (RALA), is proposed. RALA is a new logic operation architecture in that

  2. #2 peak

    , and living or life skills training.⏎- Eligible California residents may connect with the San Andreas Regional Center (SARC) to create an Individualized Program Plan (IPP). This is a contract between SARC and the individual that sets up an

  3. #3 peak

    Belgium was a founding member of NATO (it is the site of NATO Headquarters) in 1949 and the European Economic Community (EEC) (now the European Union (EU)) in 195

  4. #4 peak

    network of 150 miles of trails, including the Wild Oak National Recreation Trail and a portion of the Great Eastern Trail. The historic Civilian Conservation Corps (CCC) built many of the trails.⏎With foresight, good judgment, determination

  5. #5 peak

    courtesy by USGS)⏎Contact Information: Nicole Wright, Program Manager Toxicity Testing Unit⏎As part of the RIBS program, the Toxicity Testing Unit (TTU) uses bioassays to look for toxicity in surface waters and sed

  6. #6 peak

    providers in properly storing and handling their vaccine supply. In this toolkit, the CDC recommends the use of a specific type of TMD known as a Digital Data Logger (DDL) for continuous temperature monitoring and recording.⏎The Benefits of Using<|endoftext|>LYB

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 do the work I need to do to help the community. I want to help my I-AC (I-AC-2) I-AC-1 I-AC-2-1, I-AC-1

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

    or otherness, since until before 1950 only manuscripts with conquerors version `were known, such as Hernan Cortes or Diaz del Castillo chronicles. So this study shows disagreements and contradictions of winners.⏎The historian thinks Jose

  2. #2 peak

    at top sides.⏎The triptych is one of the most important early works by Yoshitoshi. It depicts a difficult battle from the Taiheiki (Chronicle of Great Peace), a Japanese historical epic written in the late 14

  3. #3 peak

    word "lexicography" derives from the Greek λεξικογράφος lexikographos, "lexicographer", from λεξικό

  4. #4 peak

    Catechism of the Catholic Church|⏎IntraText - Text⏎THE SACRAMENTAL CELEBRATION OF THE PASCHAL MYSTERY⏎The catechesis of the liturgy entails first of all an understanding

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. L21:272 excites
  2. L21:350 excites
  3. L21:268 excites
  4. L21:273 excites
  5. L25:138 suppresses
  6. L21:907 excites
  7. L23:363 suppresses
  8. L21:610 excites
  9. L21:459 suppresses
  10. L21:430 suppresses
  11. L21:415 suppresses
  12. L21:494 suppresses
  13. L23:404 suppresses
  14. L21:319 suppresses
  15. L21:337 suppresses
  16. L21:261 excites
  17. L23:383 suppresses
  18. L20:2808 suppresses
  19. L22:78 weakly suppresses
  20. L25:237 weakly suppresses
  21. L24:1147 weakly suppresses
  22. L21:259 weakly excites
  23. L9:620 weakly excites
  24. L19:1218 weakly suppresses

Feeds later layers

  1. L27:544 strongly excites
  2. L27:89 weakly suppresses
  3. L27:1326 weakly suppresses
  4. L27:1261 weakly excites
  5. L27:1515 weakly excites
  6. L27:245 weakly suppresses
  7. L27:859 weakly excites unclear: token just before a k sound
  8. L27:1254 weakly excites
  9. L27:321 weakly excites
  10. L27:2487 weakly excites
  11. L27:2707 weakly excites
  12. L27:1857 weakly excites
  13. L27:2057 weakly excites
  14. L27:1455 weakly excites
  15. L27:807 weakly suppresses
  16. L27:2346 weakly excites
  17. L27:1140 weakly excites
  18. L27:308 weakly excites
  19. L27:1604 weakly excites
  20. L27:2770 weakly suppresses
  21. L27:1796 weakly suppresses
  22. L27:425 weakly suppresses
  23. L27:2691 weakly suppresses
  24. L27:1546 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 -29.64 and -27.101 values between -22.00 and -19.461 values between -19.46 and -16.912 values between -16.91 and -14.371 values between -14.37 and -11.824 values between -11.82 and -9.2810 values between -9.28 and -6.738 values between -6.73 and -4.1832 values between -4.18 and -1.643711 values between -1.64 and 0.91125 values between 0.91 and 3.453 values between 3.45 and 6.003 values between 6.00 and 8.551 values between 8.55 and 11.091 values between 13.64 and 16.181 values between 28.91 and 31.462 values between 34.00 and 36.551 values between 69.64 and 72.19offnegativepositiveoften

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. YY
  2. bbb
  3. YY
  4. ZZ
  5. YYY
  6. DDD
  7. dd
  8. BBB
  9. HH
  10. hhh

Suppresses

  1. Ac
  2. acr
  3. Ac
  4. AC
  5. 镇江
  6. AC
  7. _ac
  8. (ac
  9. ac
  10. .ac