Neuron Explorer

Qwen3-0.6B

unclear: words inside program or organisation names

A neuron in layer 19 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 is hard to pin down. It fires on words in the middle of official program, agency or title names, like 'and' in 'Planning and Partnerships' or 'of' in 'Management of Health'. Turning it up made the text repetitive and unfocused.

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

    . Each performance task test is scored by two trained readers.⏎You can see some of the 2006 tests at The Florida Department of Education’s Assessment and School Performance page.⏎When is the FCAT given?⏎The FCAT is given

  2. #2 peak

    Board Learning Session⏎Leaders from Stillwater Area High School and Lake Elmo Elementary School provided an update on the successes of the BARR - Building Assets, Reducing Risks - program.⏎BARR is a research-based approach to meeting students’

  3. #3 peak

    . He paved the way for post-modern artists to develop their own vocabulary, drawing upon traditional sources and creating a new language that liquefied the barriers between old and new. His work titled ‘Shah Jahan’, a triptych

  4. #4 peak

    The Oakland Food Policy Council takes a comprehensive approach toward improving food access and urban agriculture in disadvantaged communities.⏎- Urban Forests: Urban Releaf and Friends of the Urban Forest are two grassroots organizations that beautify our neighborhoods by planting trees.⏎- Aqu

  5. #5 peak

    for smaller organizations or communities with limited resources.⏎What is MAPP?⏎The development of the Mobilizing for Action through Planning and Partnerships (MAPP) process was a joint project of the non-profit National

  6. #6 peak

    .⏎Whilst the specific process of preparing a formal risk assessment is not a requirement of the CDM regulations, regulation 3(6)(a) of the Management of Health and Safety at Work Regulations does require that risks are controlled in the workplace,

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: A part of a certain old, very old, and, in the past, always, still, and, in the past and, in the still, and, in the past, and, in the, in this, and, in

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

    Black, Brown, and Beige. Originally debuted at Carnegie Hall on January 23rd, 1943, the composition received mixed, if not critical feedback.⏎Much of that criticism weighed on the "pretentiousness" such

  2. #2 peak

    eye, and often affects your vision. At the first sign of any visual disturbance, including flashes, floaters, a curtain over your eye, or the sudden or gradual onset of blurred vision, your eye doctor should be consulted immediately.⏎The retina

  3. #3 peak

    Christ’s redemptive sacrifice for our sins). Jesus did not come only<|endoftext|>WEDNESDAY, July 4 (HealthDay News) -- For safety's sake, Americans should avoid using fireworks of any kind this Independence Day, experts say

  4. #4 peak

    ?⏎- How do land use practices and management efforts interact with natural disturbances and climatological cycles to affect estuarine water quality and suitability for ecologically (and economically) important species?<|endoftext|>Not Just for Kids⏎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. L18:569 suppresses
  2. L0:1224 excites
  3. L0:1301 excites
  4. L18:269 excites
  5. L3:591 excites
  6. L18:529 excites
  7. L6:17 weakly excites
  8. L18:77 weakly excites
  9. L0:178 weakly excites
  10. L2:1321 weakly excites
  11. L2:75 weakly excites
  12. L14:874 weakly suppresses
  13. L14:312 weakly excites
  14. L4:33 weakly excites
  15. L14:343 weakly suppresses
  16. L14:801 weakly excites 'and' or 'or' joining list items
  17. L6:2355 weakly excites
  18. L18:396 weakly excites
  19. L14:323 weakly suppresses
  20. L1:3019 weakly excites
  21. L0:313 weakly excites "and" joining paired nouns
  22. L5:133 weakly suppresses
  23. L18:164 weakly excites
  24. L14:2651 weakly excites

Feeds later layers

  1. L20:1865 weakly suppresses
  2. L21:749 weakly suppresses
  3. L25:1801 weakly suppresses
  4. L21:498 weakly suppresses
  5. L23:610 weakly excites
  6. L20:1957 weakly suppresses
  7. L20:1292 weakly suppresses
  8. L21:2692 weakly suppresses
  9. L24:2607 weakly suppresses
  10. L22:205 weakly excites
  11. L23:2699 weakly suppresses
  12. L20:22 weakly excites
  13. L21:909 weakly suppresses
  14. L20:2119 weakly excites
  15. L20:1174 weakly suppresses
  16. L20:379 weakly excites digits inside dates
  17. L21:729 weakly excites
  18. L20:2016 weakly suppresses
  19. L20:953 weakly suppresses
  20. L20:1727 weakly suppresses
  21. L20:2418 weakly suppresses
  22. L20:1922 weakly suppresses
  23. L20:873 weakly suppresses
  24. L21:1591 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 -8.69 and -8.116 values between -8.11 and -7.5410 values between -7.54 and -6.9610 values between -6.96 and -6.3812 values between -6.38 and -5.8121 values between -5.81 and -5.2336 values between -5.23 and -4.6651 values between -4.66 and -4.08103 values between -4.08 and -3.51140 values between -3.51 and -2.93194 values between -2.93 and -2.35286 values between -2.35 and -1.78375 values between -1.78 and -1.20441 values between -1.20 and -0.63433 values between -0.63 and -0.05413 values between -0.05 and 0.53302 values between 0.53 and 1.10265 values between 1.10 and 1.68190 values between 1.68 and 2.25144 values between 2.25 and 2.83131 values between 2.83 and 3.40100 values between 3.40 and 3.9881 values between 3.98 and 4.5646 values between 4.56 and 5.1342 values between 5.13 and 5.7127 values between 5.71 and 6.2813 values between 6.28 and 6.868 values between 6.86 and 7.4310 values between 7.43 and 8.015 values between 8.01 and 8.594 values between 8.59 and 9.164 values between 9.16 and 9.741 values between 9.74 and 10.311 values between 10.31 and 10.891 values between 13.77 and 14.34offnegativepositiveoften

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. 预
  2. quir
  3. comed
  4. amps
  5. chy
  6. коль
  7. đệ
  8. 掠
  9. ughty
  10. кро

Suppresses

  1. nonetheless
  2. nevertheless
  3. istically
  4. 技术研发
  5. ará
  6. ** ⏎
  7. enie
  8. --> ⏎
  9. Himself
  10. 还有一些