Neuron Explorer

Qwen3-0.6B

unclear: word pieces in long names

A neuron in layer 12 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 a fragment in the middle of a long or technical word, such as the middle of a drug or chemical name, or the start of a legal citation. It does not clearly have one theme. When it was turned up, the model's writing got repetitive and drifted without a clear pattern.

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

    organisms and the environment. Sulfoxaflor will control a number of difficult insect pests and is proven to work against challenging pests that carbamate, neonicotinoid, organophosphate, and pyrethroid insecticides fail to

  2. #2 peak

    itation: Formal references to statutes that describe where they are published. For instance, the citation 23 Vt. Stat. Ann. § 1185 tells us that this cited statute is

  3. #3 peak

    result was the growth of two human embryonic stem cell lines that lack PIG-A, and therefore do not contain any proteins known as glycosylphosphatidy

  4. #4 peak

    dental problems before they worsen. In addition, your dentist can advise you on enhancing your dental health and propose fluoride treatments and sealants. Read more Composite Fillings.⏎4. Limit sugary and acidic foods or

  5. #5 peak

    city council of Trier, a German town where the revolutionary socialist thinker was born. The statute, designed by world-renowned Chinese sculptor Wu Weishan, is a gift to celebrate the 200th anniversary of Marx’s birth

  6. #6 peak

    Gratis is used as both an adjective (The drinks were gratis) and an adverb (Drinks were served gratis).What is the definition of free?⏎The definition of

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 have been thinking about this, and I know this is going to happen." But I think the idea is going through here, and I can't be sure. It seems to be a good opportunity. I've been thinking about this

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

    .⏎On the other hand, the program drops the decryption key on your desktop. It is in the key.txt file and you can use to restore<|endoftext|>What is insulin resistance?⏎This educational content is brought to you by the ADA x BT1 Coll

  2. #2 peak

    Draw a boxplot of these data. Use the boxplot and quartiles to ...The following data represent the ages of chief justices of the U.S. Supreme Court when they were appointed. Justice ....... Age John Jay

  3. #3 peak

    panmixia of an African fruit bat facilitates transmission of potentially zoonotic viruses.<|endoftext|>The AR-10 is a very important firearm in the history of weapons. It served as the predecessor to the AR-15 and the AR platform

  4. #4 peak

    .S. Greenhouse Gas Emissions and Sinks: 1990-2019<|endoftext|>Following directions with color words is a fun practice activity. View other primary units and freebies at www.creativeclassroomconnections.com.⏎

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. L11:1692 strongly excites
  2. L11:957 strongly suppresses
  3. L11:1328 excites
  4. L11:444 suppresses
  5. L11:1064 suppresses
  6. L11:211 excites
  7. L9:203 suppresses
  8. L11:928 excites
  9. L9:794 excites
  10. L11:539 suppresses
  11. L6:34 suppresses
  12. L11:401 suppresses
  13. L8:139 excites
  14. L11:1041 suppresses
  15. L10:78 suppresses
  16. L5:299 suppresses
  17. L11:70 suppresses
  18. L6:109 excites
  19. L9:769 excites
  20. L11:59 suppresses
  21. L11:659 excites
  22. L6:126 excites
  23. L11:1107 excites
  24. L7:461 suppresses

Feeds later layers

  1. L14:1194 excites
  2. L13:680 suppresses
  3. L14:308 excites
  4. L14:763 suppresses
  5. L18:397 excites
  6. L15:1752 suppresses
  7. L14:1197 suppresses
  8. L14:1603 suppresses
  9. L14:1365 suppresses
  10. L16:1132 excites
  11. L13:230 suppresses
  12. L15:465 excites
  13. L13:715 excites
  14. L18:942 excites
  15. L16:638 suppresses
  16. L15:2228 excites
  17. L13:1067 excites
  18. L13:1168 excites
  19. L15:2641 excites
  20. L16:2169 suppresses
  21. L13:1018 weakly excites
  22. L15:2475 weakly suppresses
  23. L13:1045 weakly excites
  24. L14:863 weakly excites unclear: document ends and timestamps

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 -3.24 and -3.111 values between -2.46 and -2.331 values between -2.07 and -1.943 values between -1.94 and -1.815 values between -1.81 and -1.684 values between -1.68 and -1.554 values between -1.55 and -1.4110 values between -1.41 and -1.2821 values between -1.28 and -1.1526 values between -1.15 and -1.0257 values between -1.02 and -0.8977 values between -0.89 and -0.76131 values between -0.76 and -0.63174 values between -0.63 and -0.50255 values between -0.50 and -0.37362 values between -0.37 and -0.24446 values between -0.24 and -0.11646 values between -0.11 and 0.02589 values between 0.02 and 0.15423 values between 0.15 and 0.28276 values between 0.28 and 0.41158 values between 0.41 and 0.5496 values between 0.54 and 0.6881 values between 0.68 and 0.8127 values between 0.81 and 0.9414 values between 0.94 and 1.0713 values between 1.07 and 1.203 values between 1.20 and 1.331 values between 1.33 and 1.461 values between 1.46 and 1.591 values between 1.72 and 1.851 values between 1.85 and 1.98offnegativepositiveoften

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. 分泌
  2. dec
  3. uste
  4. uled
  5. ulet
  6. ML
  7. 酮
  8. theless
  9. andest
  10. MLElement

Suppresses

  1. ?
  2. ?⏎⏎
  3. ?> ⏎ ⏎
  4. ?」
  5. ? ⏎
  6. ?> ⏎
  7. _xlabel
  8. ournaments
  9. ?⏎
  10. орг