Neuron Explorer

Qwen3-0.6B

abbreviations in parentheses

A neuron in layer 21 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 fires on the short letters inside brackets that stand for a term, like 'gbh' or 'PPCPs' or 'IEQ', in the spot where an acronym is being introduced. When turned up, the output became scrambled text full of capital letters.

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

    also analyzes the impact of various tree functional traits on liana colonization success in the PCWS. All trees ≥ 10 cm girth at breast height (gbh) was measured along with lianas (≥ 1 cm diameter at

  2. #2 peak

    alienates feminism, and the discrimination it fights, from the very current issues of gendered violence, rape culture, sexual objectification, men’s rights activists (MRAs) and the very real but less-explicit-than-they-once-w

  3. #3 peak

    health.”⏎In September, Governor Schwarzenegger signed into law legislation banning sales of children’s jewelry that contain more than 300 parts per million (ppm) total cadmium. In advocating for the law, which had bipartisan and

  4. #4 peak

    stress and to resist disease.”⏎The most popular IF methods include:⏎- Alternate day fasting⏎- 5/2 fasting, which

  5. #5 peak

    GreatSchools.org School Profiles.⏎What are the Sunshine State Standards (SSS)?⏎The Sunshine State Standards are Florida’s version of statewide learning standards which define

  6. #6 peak

    treating nausea, motion sickness and a variety of skin issues. Dramamine or Bonine are also useful for dogs suffering from motion sickness. Before giving your dog an OTC medication, check with your vet, since they could worsen certain conditions or

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 lot of people are **constrated** to a **short** **length** to **elegant** **length**. It's a **constrated** **length** to **elegant** **length** to **

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

    analyses of experiments to improve the understanding of the phenomenology.⏎The Large Atomic/Molecular Massively Parallel Simulator (LAMMPS) is a Discrete Element Method (DEM) code developed at

  2. #2 peak

    results are based on the widely accepted Greenhouse Gas Protocol [GHGP] established by the World Resources Institute [WRI] and World Business Council for Sustainable Development [WBCSD and were announced during an AMD sponsored media panel on energy efficient information

  3. #3 peak

    s. This case study investigates the resolution of the conflict. Questions remain as to precisely how this was achieved. Once these questions are resolved<|endoftext|>Wi-Fi Protected Access (WPA), Wi-Fi Protected Access II (WPA2), and Wi-Fi

  4. #4 peak

    of three concepts undergoing preliminary studies for the three-year, €7 million ($9 million) European hypersonic research project, Long Term Advanced Propulsion Concepts and Technologies (LAPCAT).⏎With €4 million from the European Commission and the rest from

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. L19:1829 suppresses opening bracket before an acronym
  2. L16:812 excites
  3. L16:722 suppresses
  4. L19:1011 excites
  5. L20:912 suppresses
  6. L17:148 suppresses
  7. L20:2862 excites
  8. L19:1778 excites
  9. L16:369 excites
  10. L8:1976 suppresses
  11. L17:2755 suppresses
  12. L19:2368 excites
  13. L19:409 excites
  14. L6:1478 weakly suppresses
  15. L14:2419 weakly excites
  16. L20:366 weakly excites
  17. L20:18 weakly excites
  18. L20:556 weakly suppresses
  19. L18:1733 weakly excites
  20. L10:86 weakly suppresses
  21. L20:225 weakly suppresses
  22. L20:987 weakly suppresses
  23. L10:310 weakly excites
  24. L17:342 weakly excites

Feeds later layers

  1. L23:404 excites
  2. L24:699 excites
  3. L24:1278 excites
  4. L25:246 excites
  5. L25:200 excites
  6. L26:761 excites
  7. L23:349 excites comma after author surname in citations
  8. L25:446 excites
  9. L25:735 excites
  10. L24:355 excites
  11. L24:288 excites
  12. L27:379 excites
  13. L26:1597 excites
  14. L23:180 excites
  15. L26:816 excites
  16. L27:782 excites
  17. L25:617 weakly excites
  18. L23:630 weakly excites
  19. L24:2002 weakly excites
  20. L22:1393 weakly excites
  21. L22:1784 weakly excites
  22. L24:1970 weakly excites
  23. L24:481 weakly excites
  24. L26:23 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 -9.20 and -8.291 values between -6.45 and -5.533 values between -3.70 and -2.784 values between -2.78 and -1.8694 values between -1.86 and -0.953108 values between -0.95 and -0.03612 values between -0.03 and 0.8923 values between 0.89 and 1.818 values between 1.81 and 2.7310 values between 2.73 and 3.6414 values between 3.64 and 4.562 values between 4.56 and 5.484 values between 5.48 and 6.404 values between 6.40 and 7.312 values between 7.31 and 8.232 values between 8.23 and 9.151 values between 9.15 and 10.071 values between 10.98 and 11.901 values between 11.90 and 12.821 values between 14.65 and 15.573 values between 15.57 and 16.491 values between 16.49 and 17.412 values between 17.41 and 18.323 values between 20.16 and 21.081 values between 22.91 and 23.831 values between 23.83 and 24.751 values between 26.58 and 27.50offnegativepositiveoften

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. FP
  2. RL
  3. SI
  4. CB
  5. C
  6. oS
  7. cs
  8. DC
  9. BT
  10. TF

Suppresses

  1. iaux
  2. idade
  3. ướng
  4. ieten
  5. INCT
  6. urers
  7. productName
  8. icular
  9. 多余的
  10. ******************************************************************************⏎