Neuron Explorer

Qwen3-0.6B

'from' after verbs and adjectives of separation or origin

A neuron in layer 5 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 fires on the word 'from' after words like 'fall', 'fell', 'sick', 'tired' or 'wet', as in 'fell from 13.1 to 10.7'. Weak negative activity appears on 'of' after words like 'number' and 'percentage'. Turning it up didn't produce a clear theme.

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

    when you are sick.⏎Currently there are no advisories restricting travel to the Arabian Peninsula or South Korea.<|endoftext|>"The most significant fact is that it will not fall from inside pressure. Only outside, foreign, hostile intervention can destroy it."⏎In fact

  2. #2 peak

    experienced a significantly smaller reduction in their poverty rates than those counties that did not add new Wal-Mart stores.⏎Overall, the portion of families living in poverty nationwide fell from 13.1 to 10.7 percent between 19

  3. #3 peak

    2% of their income on entertainment – plus they would have better access to water, sanitation and electricity. These, then, are the “winners” from globalisation: an expanding group for whom global growth has meant a serious rise in

  4. #4 peak

    ped into the room, ivory-handled pistol on his belt, jodhpurs flaring from his hips to knees. The general, tired and irritable from the long months of planning and training, and the longer months of fighting, was

  5. #5 peak

    to check the soil of the Medusa head before watering. If the soil is dry, it means that the plant requires irrigation. If the soil is still wet from the last watering, it is better to wait for the soil to completely dry to

  6. #6 peak

    time each day. This method appeals to many people because you can tailor the timing to your lifestyle and schedule, says Kelly. For example, some people may fast from 8pm until lunch around midday the next day, or they may eat

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: 10% chance of being in, and if you get any number of times that you have, you will have to stay, and if you get a lot, you should be ready to take action. (See the same for

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

    loss surgery developed an alcohol-abuse problem compared to the 11 percent of those who underwent gastric banding.⏎“We knew there was an increase in the number of people experiencing problems with alcohol within the first two years of surgery, but we didn

  2. #2 peak

    Opinions and Lifestyle Survey.⏎The percentage of current smokers in Great Britain has remained at 19% compared with 2013, but the percentage of male smokers

  3. #3 peak

    adults older than 65 — they accounted for a smaller proportion of deaths in these populations.⏎Among the eldest two age groups, however, the percentage of deaths attributable to opioids remained

  4. #4 peak

    to cardiovascular-related disability and death is also clear. In fact, in a study from Helena, Mont., last year, researchers found an immediate reduction in the number of heart attacks – a 40 percent decline – when a clean indoor air policy

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. L4:658 excites
  2. L0:2085 weakly suppresses
  3. L4:2310 weakly excites
  4. L2:1789 weakly suppresses
  5. L4:2166 weakly suppresses
  6. L4:1409 weakly suppresses
  7. L1:2924 weakly excites
  8. L3:2800 weakly suppresses
  9. L4:192 weakly excites
  10. L1:883 weakly excites
  11. L1:522 weakly excites
  12. L0:2426 weakly excites
  13. L4:827 weakly excites
  14. L4:2735 weakly excites
  15. L2:2322 weakly excites
  16. L2:2570 weakly excites
  17. L0:2423 weakly excites
  18. L4:879 weakly excites
  19. L1:2771 weakly suppresses
  20. L1:338 weakly suppresses
  21. L4:1394 weakly excites
  22. L2:2718 weakly excites
  23. L3:808 weakly excites
  24. L0:330 weakly excites

Feeds later layers

  1. L6:853 weakly excites
  2. L19:1868 weakly suppresses
  3. L6:2143 weakly excites
  4. L20:1643 weakly excites
  5. L22:457 weakly excites
  6. L15:1019 weakly suppresses
  7. L10:1784 weakly excites
  8. L19:1666 weakly suppresses
  9. L18:2751 weakly suppresses
  10. L17:975 weakly excites
  11. L9:1806 weakly suppresses
  12. L18:2300 weakly excites
  13. L17:1776 weakly excites
  14. L11:1384 weakly excites
  15. L17:1792 weakly excites
  16. L7:1798 weakly excites
  17. L26:2580 weakly suppresses
  18. L25:515 weakly suppresses
  19. L20:1440 weakly excites
  20. L7:1882 weakly excites
  21. L14:2462 weakly excites
  22. L13:2249 weakly suppresses
  23. L10:1496 weakly excites
  24. L24:2159 weakly 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 -0.95 and -0.852 values between -0.85 and -0.761 values between -0.76 and -0.666 values between -0.57 and -0.477 values between -0.47 and -0.3845 values between -0.38 and -0.28127 values between -0.28 and -0.19383 values between -0.19 and -0.091214 values between -0.09 and 0.001363 values between 0.00 and 0.10522 values between 0.10 and 0.19153 values between 0.19 and 0.2957 values between 0.29 and 0.388 values between 0.38 and 0.472 values between 0.47 and 0.574 values between 0.57 and 0.661 values between 0.66 and 0.763 values between 0.76 and 0.851 values between 0.85 and 0.952 values between 1.04 and 1.141 values between 1.52 and 1.611 values between 1.80 and 1.901 values between 1.90 and 1.991 values between 1.99 and 2.081 values between 2.27 and 2.371 values between 2.75 and 2.84offnegativepositiveoften

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. quil
  2. -League
  3. HOLDER
  4. booklet
  5. olith
  6. 设计
  7. orna
  8. 的设计
  9. etas
  10. ubo

Suppresses

  1. lichkeit
  2. 望
  3. keit
  4. APPED
  5. minLength
  6. Intermediate
  7. mj
  8. aris
  9. 吻
  10. commodo