Neuron Explorer

Qwen3-0.6B

periods and colons in DOIs and citations

A neuron in layer 11 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 punctuation inside academic citations, especially the dot in 'doi:10.1096' and the colon after 'Authors'. When it was turned up, the model's text drifted into unrelated, repetitive material.

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

    Prev Med Rep⏎Authors: Altman EA, Lee KL, Hecht CA, Hampton KE, Moreno G

  2. #2 peak

    elevation of erythropoietin in the brain improves exercise performance in mice without inducing erythropoiesis. FASEB J. doi:10.1096<|endoftext|>Valentine’s Day Task Cards (2nd –

  3. #3 peak

    - freely available⏎Sustainability 2013, 5(7), 3006-3008; doi:10.3390/su5073006⏎I am

  4. #4 peak

    iodiversity and Conservation (2004) 13: 2511. doi:10.1023/B:BIOC.00000484

  5. #5 peak

    that some parts of the moon spend about a quarter of the moon's orbit protected by Earth's magnetosphere (Geophysical Research Letters, DOI: 10.1029/2007GL030507

  6. #6 peak

    No. AAAJMR-22-82571(R); Published: 29-Nov-2022, DOI:10.35841/ aaajmr-6.11.1

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: "The President, President (c) and the Constitution, (c) is the basis of the U.S. Constitution and the U.S. government, the U.S. Constitution, (c) and the U.S. government, the

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

    secretpassword';⏎mysql> GRANT ALL PRIVILEGES ON database_name.* TO user_name@localhost;⏎mysql> FLUSH PRIVILEGES;⏎Step 5

  2. #2 peak

    of this page: http://www.cal-ipc.org/ip/management/plant_profiles/Rumex_acetosella.php⏎Rumex acetosella (red sorrel, sheep sorrel)⏎Rumex acetosella (red

  3. #3 peak

    or can be readily restored to shoot, automatically more than one shot, without manual reloading, by a single function of the trigger.” 26 U.S.C. § 5845(b). The machine gun was first widely used during

  4. #4 peak

    was completed in 1988. The pharmaceutical research<|endoftext|>© 2016 by the authors. The mixed Cunninghamia lanceolata (Lamb.) Hook., Pinus massoniana Lamb., and hardwood forest in southeastern

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. L10:789 weakly suppresses
  2. L0:2487 weakly suppresses
  3. L9:195 weakly excites
  4. L8:2109 weakly suppresses
  5. L10:2512 weakly suppresses
  6. L3:2563 weakly excites
  7. L7:1280 weakly suppresses
  8. L6:728 weakly suppresses
  9. L9:1233 weakly suppresses
  10. L0:1944 weakly suppresses
  11. L4:1853 weakly suppresses
  12. L0:2735 weakly suppresses
  13. L4:1487 weakly suppresses
  14. L9:2070 weakly suppresses
  15. L6:1955 weakly suppresses
  16. L8:1717 weakly excites
  17. L1:2993 weakly suppresses
  18. L7:1423 weakly excites
  19. L4:3050 weakly excites
  20. L10:446 weakly excites
  21. L10:785 weakly suppresses
  22. L2:2918 weakly suppresses
  23. L5:1603 weakly suppresses
  24. L10:2463 weakly excites

Feeds later layers

  1. L12:1591 suppresses
  2. L17:584 suppresses
  3. L15:1640 suppresses
  4. L13:1706 suppresses
  5. L22:324 weakly excites
  6. L12:2232 weakly suppresses
  7. L13:2189 weakly suppresses
  8. L16:2942 weakly suppresses
  9. L13:2995 weakly excites
  10. L22:414 weakly excites
  11. L16:944 weakly suppresses
  12. L18:1455 weakly suppresses
  13. L18:932 weakly suppresses
  14. L18:1650 weakly suppresses
  15. L12:2555 weakly excites
  16. L14:2288 weakly suppresses
  17. L17:1588 weakly suppresses
  18. L16:2986 weakly suppresses
  19. L13:2269 weakly suppresses
  20. L14:1248 weakly excites
  21. L13:2030 weakly suppresses
  22. L23:943 weakly excites
  23. L21:1332 weakly excites
  24. L17:849 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.

1 values between -2.05 and -1.921 values between -1.66 and -1.531 values between -1.27 and -1.141 values between -0.88 and -0.753 values between -0.75 and -0.626 values between -0.62 and -0.4951 values between -0.49 and -0.36233 values between -0.36 and -0.23759 values between -0.23 and -0.101440 values between -0.10 and 0.03956 values between 0.03 and 0.16358 values between 0.16 and 0.2980 values between 0.29 and 0.4215 values between 0.42 and 0.551 values between 0.55 and 0.681 values between 0.68 and 0.811 values between 3.02 and 3.15offnegativepositiveoften

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. aram
  2. anoi
  3. _third
  4. →hs
  5. :http
  6. 'https
  7. :boolean
  8. :max
  9. blade
  10. [email

Suppresses

  1. prive
  2. mute
  3. igin
  4. ง
  5. surfaced
  6. statt
  7. isa
  8. asion
  9. ứng
  10. 垣