Neuron Explorer

Qwen3-0.6B

commas and 'and' in noun lists

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 reacts to the commas and 'and' that separate items in a list of things, like 'cutting patterns, and hand writing' or 'fish, chemistry, and physics'. Turning it up made the writing fall into repetitive lists, though not clearly the same kind of list.

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

    , students can compare trends that occur based⏎upon certain variables. Even the death of a fish is an opportunity to⏎address and discuss possible causes, preventions and ways to improve⏎Biology, chemistry, ecology, and physics are just a

  2. #2 peak

    continue to grow, such as writing, tying shoes, typing and so much more⏎2)Fine Motor Coordination– In order to draw shapes, cut patterns, and hand writing, your child is required to use their fine motor coordination. These

  3. #3 peak

    Electronic Devices⏎3 – Alternating Current (AC) Fundamentals⏎4 – DC and AC Power⏎5 – Power Factor⏎6 – Demand, Load Factor, Service Factor and Electrical Power Bill Computation⏎7 – Electric Motors and Generators

  4. #4 peak

    ers, along with a Universal Primer that binds to the universal tag sequence will amplify specific<|endoftext|>Create a free profile to get unlimited access to exclusive videos, sweepstakes, and more!⏎It’s difficult to describe the immensity of galaxy clusters; they

  5. #5 peak

    they meet the required specifications.⏎<|endoftext|>RASPBERRY Pi - LEARNING HOW TO MAKE YOUR ROBOT WORK⏎ANTICS - ARDUINO - ARMOUR - ARTWORK - BIOLOGY - BLACK BOX - COMPUTERS - ELECTRONICS -

  6. #6 peak

    the product, check<|endoftext|>Help support New Advent and get the full contents of this website as an instant download or CD-ROM. Includes the Catholic Encyclopedia, Church Fathers, Summa, Bible and more all for only $19.99...⏎

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: Can this argument have? An example? A very good one? Such? A very good argument? A very good argument? A very good argument? A very good argument? A very good argument? A very good argument? A very good

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

    and maintaining copyrights, netiquette, privacy, and user responsibility.⏎Books & Booklets; Internet Resources⏎Administration & Leadership⏎Administrator, Teacher, Board Member / Trustee<|endoftext|>This post is Day 3 in a series of studies about “

  2. #2 peak

    ⏎Stop breadboarding and soldering – start making immediately! Adafruit’s Circuit Playground is jam-packed with LEDs, sensors, buttons, alligator clip pads and more. Build projects with Circuit Playground in a few minutes with the drag-and-drop Make

  3. #3 peak

    evidence that vitamin D reduces the risk of viral infections. Influenza is more common in winter for two reasons: cold temperatures and low solar UVB doses and vitamin D production. A randomized controlled trial on Japanese school children found that those taking

  4. #4 peak

    1) The dominator⏎Establish procedure to ensure equal participation.⏎Interrupt and ask to summarise⏎Thank the person and hear from others in the group⏎(2) The shy person⏎-use ice-breakers to make them comfortable. Get them

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. L9:705 excites
  2. L10:1953 suppresses
  3. L7:163 suppresses
  4. L10:2518 suppresses
  5. L10:597 suppresses
  6. L8:512 excites
  7. L7:333 excites
  8. L10:1090 weakly suppresses
  9. L8:558 weakly excites
  10. L8:2576 weakly suppresses
  11. L8:1645 weakly excites
  12. L7:821 weakly excites
  13. L10:3020 weakly excites
  14. L8:719 weakly excites
  15. L8:195 weakly suppresses
  16. L4:276 weakly excites
  17. L7:406 weakly suppresses
  18. L9:1656 weakly excites
  19. L7:323 weakly suppresses
  20. L3:377 weakly suppresses
  21. L10:1171 weakly excites
  22. L1:1559 weakly excites
  23. L9:1374 weakly suppresses
  24. L7:303 weakly suppresses

Feeds later layers

  1. L12:1585 suppresses
  2. L12:1937 suppresses
  3. L13:758 excites 'and' joining a pair of options
  4. L14:1335 excites
  5. L13:517 excites comma closing a list item
  6. L12:959 excites
  7. L15:1708 excites
  8. L16:492 suppresses
  9. L14:1159 suppresses
  10. L15:217 weakly suppresses
  11. L12:2726 weakly suppresses
  12. L16:2733 weakly excites
  13. L15:545 weakly excites
  14. L12:749 weakly suppresses
  15. L16:1388 weakly excites
  16. L13:977 weakly excites
  17. L13:125 weakly excites
  18. L14:345 weakly suppresses
  19. L13:1821 weakly excites
  20. L12:1209 weakly suppresses
  21. L13:1533 weakly excites
  22. L16:1294 weakly excites
  23. L12:690 weakly excites
  24. L13:1462 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.

2 values between -3.21 and -2.971 values between -2.49 and -2.253 values between -2.25 and -2.011 values between -1.76 and -1.522 values between -1.52 and -1.285 values between -1.28 and -1.047 values between -1.04 and -0.8027 values between -0.80 and -0.5681 values between -0.56 and -0.32955 values between -0.32 and -0.082518 values between -0.08 and 0.17223 values between 0.17 and 0.4131 values between 0.41 and 0.6513 values between 0.65 and 0.8910 values between 0.89 and 1.136 values between 1.13 and 1.372 values between 1.37 and 1.611 values between 1.61 and 1.852 values between 1.85 and 2.103 values between 2.10 and 2.341 values between 2.34 and 2.582 values between 2.58 and 2.823 values between 2.82 and 3.061 values between 3.06 and 3.303 values between 3.30 and 3.541 values between 3.78 and 4.032 values between 4.03 and 4.271 values between 4.99 and 5.231 values between 6.20 and 6.44offnegativepositiveoften

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. icans
  2. orous
  3. iated
  4. )'),⏎
  5. Near
  6. )||(
  7. tık
  8. ktop
  9. agi
  10. edelta

Suppresses

  1. 水平
  2. プリ
  3. ích
  4. StringUtil
  5. _FRAGMENT
  6. Serialization
  7. SRC
  8. level
  9. 动生成
  10. 车上