Neuron Explorer

Qwen3-0.6B

mid-word pieces in anatomy and place names

A neuron in layer 3 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 middle piece of a longer word that has been split up, such as 'patel' in 'patellar', 'axon' in 'axoneme', and 'Namib' in 'Namibian'. It is pushed negative by words like 'father' and 'Duke'. Steering gave text without a clear link to these words.

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

    ialis, vastus intermedius, and the rectus femoris. These muscles are present just above the patella or the kneecap and form the patellar tendon. These muscles perform the function of extending the leg at the knee and facilitate

  2. #2 peak

    the patellar tendon mostly.⏎Achilles Tendon: This tendon is present behind the foot

  3. #3 peak

    axoneme's length. The outer arm-less doublet of each flagellum faces

  4. #4 peak

    ambique and Angola: Advances and Absences Chapter 11 Peacekeeping, Alphabet Soup, and Violence Against Women in the Balkans Chapter 12 The Namibian Peace Operation in a Gender Context Part 13 Peacekeeping Operations, International

  5. #5 peak

    and dredged material. Furthermore, wasting food is equivalent to wasting water because we are wasting the resources used to produce food.⏎Inspiration from Nature⏎The Namib desert fogstand beetle (Stenocara gracilipes) is inspirational.

  6. #6 peak

    ries always occur in the same doublets from section to section, indicating that certain doublets have consistent morphological specializations. These unique doublets give the axoneme an inherent structural polarity. All three specializations are present in the proximal

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 great thing in the government. The people of the state of Han Yu. In the Han Dynasty, the great virtue is not the same as the great thing in the state of Sichuan. The people of the state of Anh

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

    idiom two different meanings, though somewhat related to each other.⏎The first story of origin came from the early Spring and Autumn period. The story goes when The Duke Huan of Qi, the first super power emerged since the demise of Zhou King

  2. #2 peak

    the tongue lashing from Claudius and Gertrude. Denied his studies, the crown, the alliance of his mother and the guidance of his father, Hamlet feels like he has no purpose. He says it is only the

  3. #3 peak

    ǐ), describes the “proper” social relationships and the set of rituals which regulate them. The fundamental social relationships are those between the emperor and his ministers, between father and son, between husband and wife

  4. #4 peak

    neighbor. He nevertheless won a major victory over the Edomites. Afterward, however, his veneration of the gods of Edom earned him a prophetic condemnation.⏎Seeking to punish Israel for raids against Judean towns and other insults

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. L0:1706 weakly excites
  2. L0:1174 weakly excites
  3. L2:800 weakly excites
  4. L1:277 weakly excites
  5. L2:920 weakly excites 'of' after quantity words
  6. L0:1626 weakly suppresses
  7. L2:2595 weakly excites
  8. L0:299 weakly excites
  9. L1:951 weakly excites
  10. L1:908 weakly excites
  11. L2:36 weakly excites
  12. L2:175 weakly excites
  13. L1:3041 weakly excites
  14. L0:1978 weakly suppresses
  15. L2:483 weakly suppresses
  16. L1:3069 weakly excites
  17. L2:1137 weakly excites
  18. L0:1737 weakly excites
  19. L0:1559 weakly suppresses
  20. L1:1801 weakly excites
  21. L1:1652 weakly suppresses
  22. L1:1873 weakly suppresses
  23. L1:2825 weakly excites
  24. L1:2082 weakly excites

Feeds later layers

  1. L6:3026 suppresses
  2. L9:2452 suppresses
  3. L10:446 suppresses
  4. L18:2878 suppresses
  5. L4:410 suppresses
  6. L24:970 suppresses
  7. L6:331 suppresses
  8. L5:1592 suppresses
  9. L20:673 excites
  10. L7:247 weakly excites
  11. L11:308 weakly suppresses
  12. L14:1815 weakly suppresses
  13. L21:2058 weakly suppresses
  14. L22:2581 weakly excites
  15. L9:1480 weakly excites
  16. L19:2218 weakly suppresses
  17. L4:2202 weakly suppresses
  18. L14:1204 weakly excites
  19. L13:2017 weakly excites
  20. L9:1673 weakly excites
  21. L22:2657 weakly suppresses
  22. L5:255 weakly suppresses
  23. L4:840 weakly excites
  24. L6:973 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 -1.35 and -1.301 values between -1.30 and -1.261 values between -1.12 and -1.071 values between -0.98 and -0.931 values between -0.93 and -0.881 values between -0.74 and -0.701 values between -0.70 and -0.651 values between -0.65 and -0.601 values between -0.56 and -0.513 values between -0.51 and -0.465 values between -0.46 and -0.423 values between -0.42 and -0.3714 values between -0.37 and -0.3230 values between -0.32 and -0.2865 values between -0.28 and -0.23123 values between -0.23 and -0.18177 values between -0.18 and -0.14296 values between -0.14 and -0.09431 values between -0.09 and -0.04638 values between -0.04 and 0.01614 values between 0.01 and 0.05473 values between 0.05 and 0.10363 values between 0.10 and 0.15252 values between 0.15 and 0.19179 values between 0.19 and 0.24116 values between 0.24 and 0.2958 values between 0.29 and 0.3331 values between 0.33 and 0.3813 values between 0.38 and 0.4312 values between 0.43 and 0.473 values between 0.47 and 0.52offnegativepositiveoften

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. ilst
  2. 栲
  3. Reg
  4. _accessible
  5. 腔
  6. 操作
  7. cls
  8. addin
  9. groupId
  10. 荷

Suppresses

  1. 告诉他
  2. wort
  3. führen
  4. wróci
  5. TERS
  6. bè
  7. 帮他
  8. VERTISEMENT
  9. ynthesis
  10. 厩