Neuron Explorer

Qwen3-0.6B

verb right after a subject in clauses

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 responds to the word right after a noun in the middle of a sentence, especially in 'when' or 'whenever' clauses, like 'when a fire [is] sensed' or 'whenever someone [wishes]'. Turning it up did not produce anything clearly related, so the link is only from the examples.

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

    easier for you to know when something is wrong.⏎As busy as your routine might be<|endoftext|>The Most Common Precancer⏎

  2. #2 peak

    mental function. Some people have one or more disabilities.⏎A handicap is a physical or attitudinal constraint imposed upon a person, regardless of whether that person has a disability. Webster's Ninth New Collegiate Dictionary defines the handicap as "to put

  3. #3 peak

    . The epidemic of “Swollen shoot” requires that these artisans know the state of health of their plant. Moving an agronomist every time a plant has anomalies would be a waste of time and cost the planter enormous costs. It

  4. #4 peak

    alarm type would be most efficient for your residence.⏎Having a professionally installed, monitored fire alarm system in your home offers you the best overall protection. When a fire is sensed, your alarm will activate all evacuation horns throughout your home. Even if you

  5. #5 peak

    percentage of the total cost of a healthcare service, while the insurance company covers the remaining portion.⏎e. Out-of-Pocket Maximum: The maximum amount individuals have to pay during a policy period, after which the insurance company covers all

  6. #6 peak

    on the internet. This is a good thing since paperwork is not transformed into computer work. Almost all the data is stored on the computer and thus, whenever someone wishes to go through it, all he has to do is switch on the computer and

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 number of the same person who has been in the same position in the same position of the same value. This is the beginning of a new problem in a new year. What is the problem of the same value? This is the beginning

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

    , likewise called vanity publishing or self-publication, takes place when an author consents to allow a vanity press to replicate, copywrite, modify and also offer his/her book. The regular agreement establishes the settlement terms along with an outline of all

  2. #2 peak

    but are moved to a less desirable territory because it is a minority neighborhood, while a white employee with much lower sales is given your territory and client base, enabling him to make much more in commissions than you will make for several years.⏎- Job

  3. #3 peak

    the Indian courts are overburdened with the backlog of cases and the regular courts are to decide the Cases involve a lengthy, expensive and tedious procedure. The court takes years together to settle even petty cases. LOK ADALATS, therefore

  4. #4 peak

    Insufficient vocabulary, comprehension and literacy skills negatively impact children for the rest of their lives. Once a child falls behind in language development, it is very difficult for him to catch up after 3rd grade. That’s why it’s so alarming

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. L2:72 suppresses
  2. L2:286 suppresses
  3. L2:397 suppresses
  4. L2:40 excites
  5. L2:165 excites
  6. L2:2242 suppresses
  7. L2:816 suppresses
  8. L2:974 excites
  9. L2:940 weakly suppresses
  10. L2:844 weakly suppresses
  11. L2:898 weakly suppresses
  12. L0:1225 weakly suppresses
  13. L2:139 weakly suppresses
  14. L2:434 weakly excites
  15. L2:996 weakly suppresses
  16. L2:332 weakly suppresses
  17. L2:680 weakly excites 'a' before a noun phrase
  18. L2:540 weakly suppresses
  19. L2:260 weakly suppresses
  20. L2:1159 weakly suppresses
  21. L2:1996 weakly excites
  22. L2:452 weakly suppresses
  23. L1:2992 weakly excites
  24. L2:756 weakly excites

Feeds later layers

  1. L5:576 strongly suppresses
  2. L5:188 strongly excites
  3. L6:263 strongly suppresses
  4. L6:201 strongly suppresses
  5. L6:337 excites
  6. L4:484 excites
  7. L4:29 excites
  8. L8:46 suppresses
  9. L5:167 excites
  10. L5:93 suppresses
  11. L4:218 suppresses
  12. L4:553 suppresses
  13. L5:151 suppresses
  14. L8:53 excites
  15. L6:307 excites
  16. L5:253 suppresses
  17. L5:175 excites
  18. L19:1135 excites
  19. L7:34 suppresses
  20. L9:201 excites
  21. L5:22 excites
  22. L7:267 excites
  23. L10:170 excites
  24. L6:488 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 -3.24 and -3.101 values between -2.83 and -2.691 values between -2.42 and -2.291 values between -2.29 and -2.151 values between -2.15 and -2.021 values between -2.02 and -1.883 values between -1.88 and -1.745 values between -1.74 and -1.6112 values between -1.61 and -1.475 values between -1.47 and -1.349 values between -1.34 and -1.2018 values between -1.20 and -1.0724 values between -1.07 and -0.9338 values between -0.93 and -0.8054 values between -0.80 and -0.6679 values between -0.66 and -0.5293 values between -0.52 and -0.39176 values between -0.39 and -0.25261 values between -0.25 and -0.12898 values between -0.12 and 0.021015 values between 0.02 and 0.16753 values between 0.16 and 0.29304 values between 0.29 and 0.4390 values between 0.43 and 0.5631 values between 0.56 and 0.7013 values between 0.70 and 0.837 values between 0.83 and 0.976 values between 0.97 and 1.102 values between 1.10 and 1.241 values between 1.38 and 1.511 values between 1.51 and 1.652 values between 1.65 and 1.782 values between 2.05 and 2.19offnegativepositiveoften

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. lector
  2. diver
  3. aveled
  4. legate
  5. dernier
  6. últimos
  7. último
  8. últ
  9. LEG
  10. asiat

Suppresses

  1. .".
  2. liệu
  3. мож
  4. Anywhere
  5. ệnh
  6. ".⏎
  7. ickness
  8. Cube
  9. ็
  10. entence