Neuron Explorer

Qwen3-0.6B

"addition" in "In addition"

A neuron in layer 1 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 responds to the word "addition" in the phrase "In addition", which starts a new point that adds to the last one. The steered replies did not show a clear link to this idea.

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

    plant stress maps match, and greatly extend into surrounding regions, the on-ground CO2 emission mapping done by the USGS in Menlo Park, California.⏎In addition, vegetation health mapping along with altered mineralization mapping at Mammoth Mountain does reveal

  2. #2 peak

    11-percent MTBE by volume. MTBE has been detected about 4 to 6 times more frequently in the high-use areas than elsewhere.⏎In addition to its studies of ambient ground water, USGS is<|endoftext|>

  3. #3 peak

    , and ponds. Emerging adults are fast, strong fliers. Keeping doors and window screens in good repair can help keep deer flies out of the house. In addition to using insect repellents, wearing long sleeve shirts, long pants,

  4. #4 peak

    development of nozzle technology, the water spray posture has changed a lot. With these materials, through professional art design, you can draw a beautiful aquatic landscape.⏎In addition, different creative fountain forms are suitable for different application scenarios. For example, music

  5. #5 peak

    (including mosquitoes, ticks and some flies) can spread a number of diseases. Use insect repellent.⏎• Shower with soap before and after you swim.⏎In addition, ODH officials report that Ohio’s continuing measles outbreak is the largest in the

  6. #6 peak

    ’t prevent him from being lumped in with the incompetents that sailed on most coffin ships, and he was personally attacked in legal actions and the press.⏎In addition to the dangerous ocean crossing, emigrants faced

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 1 is a number, and the number 2 is a number. The number 3 is a number... and so on. So, I have been thinking about lately. So, I have been thinking about lately.

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

    band for random values⏎Default Value: uniformNoise.y⏎Default Value: truncatedNormalNoise.y⏎|June 22, 2015||<|endoftext|>Sadie Hawkins poems are pieces of poetry written to commemorate Sadie Hawkins day or

  2. #2 peak

    Sign up today for free teaching ideas, lesson plans, online activities, tips for your classroom, and much more.⏎Choose your grade range:⏎See a sample ><|endoftext|>In any hospital, clinic

  3. #3 peak

    still and strong formula to keep the stuff together and you can also visit https://www.customconcretetampa.com/ to get more assistance.<|endoftext|>In this chapter we introduce the other dominant form of interconnection, called Internet Peering. Internet

  4. #4 peak

    grows much faster than hevea, but the latter still has certain advantages.⏎Hevea is less prone to cracking, and doesn't allow as much heat<|endoftext|>

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. L0:1249 weakly excites
  2. L0:1301 weakly suppresses
  3. L0:149 weakly excites
  4. L0:2039 weakly excites
  5. L0:178 weakly suppresses
  6. L0:365 weakly excites
  7. L0:1224 weakly suppresses
  8. L0:1416 weakly excites
  9. L0:2403 weakly excites
  10. L0:454 weakly suppresses
  11. L0:523 weakly excites
  12. L0:259 weakly suppresses
  13. L0:1072 weakly excites
  14. L0:107 weakly suppresses newline starting a new block
  15. L0:1883 weakly suppresses
  16. L0:999 weakly excites
  17. L0:148 weakly suppresses
  18. L0:129 weakly suppresses
  19. L0:1687 weakly excites
  20. L0:927 weakly suppresses
  21. L0:1307 weakly excites
  22. L0:2849 weakly suppresses
  23. L0:2668 weakly excites
  24. L0:225 weakly excites

Feeds later layers

  1. L4:100 excites
  2. L7:260 weakly excites
  3. L6:487 weakly suppresses
  4. L22:599 weakly excites
  5. L10:181 weakly suppresses
  6. L6:866 weakly suppresses
  7. L15:180 weakly suppresses
  8. L13:2670 weakly suppresses
  9. L7:503 weakly excites
  10. L8:814 weakly excites
  11. L14:932 weakly excites
  12. L21:2438 weakly suppresses
  13. L17:933 weakly excites
  14. L10:490 weakly suppresses
  15. L12:1053 weakly suppresses
  16. L6:2744 weakly suppresses
  17. L13:699 weakly suppresses
  18. L9:727 weakly excites
  19. L3:376 weakly excites
  20. L18:341 weakly excites
  21. L5:571 weakly suppresses
  22. L17:1079 weakly suppresses
  23. L6:2819 weakly excites
  24. L19:2232 weakly excites end-of-document boundary token

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 -0.47 and -0.441 values between -0.42 and -0.401 values between -0.37 and -0.357 values between -0.35 and -0.327 values between -0.32 and -0.3013 values between -0.30 and -0.2821 values between -0.28 and -0.2532 values between -0.25 and -0.2332 values between -0.23 and -0.2059 values between -0.20 and -0.1890 values between -0.18 and -0.15129 values between -0.15 and -0.13159 values between -0.13 and -0.11237 values between -0.11 and -0.08336 values between -0.08 and -0.06394 values between -0.06 and -0.03550 values between -0.03 and -0.01666 values between -0.01 and 0.02390 values between 0.02 and 0.04263 values between 0.04 and 0.06194 values between 0.06 and 0.09127 values between 0.09 and 0.1171 values between 0.11 and 0.1449 values between 0.14 and 0.1622 values between 0.16 and 0.1819 values between 0.18 and 0.2113 values between 0.21 and 0.2311 values between 0.23 and 0.267 values between 0.26 and 0.282 values between 0.28 and 0.313 values between 0.33 and 0.351 values between 0.43 and 0.451 values between 0.47 and 0.50offnegativepositiveoften

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. rack
  2. PELL
  3. otics
  4. ampton
  5. 创造性
  6. iversal
  7. istry
  8. andscape
  9. visual
  10. tti

Suppresses

  1. />\
  2. 寥
  3. ac
  4. ờ
  5. 咨
  6. ";"
  7. ("<?
  8. ';'
  9. ;;
  10. arpa