Neuron Explorer

Qwen3-0.6B

unclear: verbs and particles mid-phrase

A neuron in layer 13 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 words in the middle of a common phrase, like 'come and go', 'break apart' or 'into space'. The pattern is hard to see, and turning it up only made the text repeat words, so treat it as unexplained.

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

    ’s Department of Earth Sciences, that reveals how “ghost glaciers” have protected Greenland’s ancient bedrock landscapes from the island’s ice sheet.⏎“These ghost glaciers come and go, and leave very little evidence of their presence,” Corbett tells LiveScience

  2. #2 peak

    But when we calculate the forces available from plate tectonics, we find that they are not large enough to do the job. We know that continents break apart and have done so repeatedly in the geological past. So, how

  3. #3 peak

    03 and repeated it a few times since then.⏎But there were two new dimensions of the recent flight. One, it carried a Chinese woman astronaut into space for the first time. Secondly and far more significantly, it demonstrated China's capability

  4. #4 peak

    ances⏎· Species movements and life stages⏎· Ecosystem processes⏎· Climate change<|endoftext|>What is Encryption?⏎Encryption is a process by which we use software to scramble sensitive information while it is in transit to Dell.⏎Please take a moment to read

  5. #5 peak

    is a new obj. instance var vh.i = 10 and HENCE v.i=10. LOCAL VARIABLE i is assigned value 0 on entry to method another() so printing these values gives 10 and 0 respectively.

  6. #6 peak

    individual school and on a system-wide basis.⏎Some of the challenges to working with parents:⏎1. By nature, they are a moving target: each year parents enter and depart the school system as they sequence from elementary, to middle, to high

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: what if it was possible to make a supercomputer? I have been thinking about this, so I know that it would be impossible, but I have been thinking about this, so I know that it would be impossible. I have been thinking

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

    you click on your city. The satellite or spacecraft name, local date, and time fields are self-explanatory.⏎Duration is the length of time the satellites will be visible in your sky on a clear night. The maximum elevation is how high

  2. #2 peak

    to respect others. Begin with the basics (saying "please" and "thank you"), then try these six other strategies.<|endoftext|>This archived Web page remains online for reference, research or recordkeeping purposes. This page will not be altered or

  3. #3 peak

    creating your Spark projects and collaborating with other data scientists. When you sign up, you get free access to Watson Studio and all other IBM services for 30 days. Start now and take advantage of this offer.⏎- Learn how it performs at

  4. #4 peak

    then instructed to close his eyes and maintain balance for up to 30 seconds.1⏎The number of seconds that the patient/client is able to maintain this position is recorded. Termination or a fail test is recorded if 1) the

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. L10:1499 excites
  2. L12:2377 excites
  3. L11:2317 excites
  4. L12:2221 excites
  5. L11:1141 suppresses
  6. L12:136 suppresses
  7. L12:2571 excites
  8. L12:2901 excites
  9. L9:2152 excites
  10. L11:2050 suppresses
  11. L11:3049 suppresses
  12. L7:2923 weakly suppresses
  13. L12:882 weakly suppresses
  14. L12:2183 weakly suppresses
  15. L12:2111 weakly excites
  16. L11:118 weakly excites
  17. L7:1440 weakly excites
  18. L11:186 weakly excites
  19. L10:24 weakly suppresses
  20. L3:251 weakly suppresses
  21. L12:1078 weakly suppresses
  22. L9:1928 weakly excites
  23. L11:1215 weakly excites
  24. L12:2530 weakly suppresses

Feeds later layers

  1. L15:1280 strongly excites
  2. L17:872 strongly excites
  3. L14:2033 strongly excites
  4. L14:1299 strongly excites
  5. L14:2969 excites
  6. L17:1078 excites
  7. L14:2924 excites
  8. L14:1882 excites
  9. L14:1239 excites
  10. L16:1920 excites
  11. L14:2544 excites
  12. L15:2832 excites
  13. L15:454 suppresses
  14. L16:2253 excites
  15. L14:2225 excites
  16. L14:843 excites
  17. L14:2354 excites
  18. L16:2902 excites
  19. L15:485 excites
  20. L14:862 excites
  21. L14:2236 excites
  22. L15:3005 excites
  23. L15:1675 excites
  24. L15:2367 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.

2 values between -1.02 and -0.831 values between -0.83 and -0.6520 values between -0.65 and -0.46181 values between -0.46 and -0.27978 values between -0.27 and -0.081993 values between -0.08 and 0.11637 values between 0.11 and 0.2982 values between 0.29 and 0.4810 values between 0.48 and 0.672 values between 0.67 and 0.861 values between 0.86 and 1.041 values between 6.29 and 6.48offnegativepositiveoften

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. LinkId
  2. :href
  3. 固定的
  4. 小编一起
  5. @Id
  6. /frontend
  7. /end
  8. /disable
  9. /archive
  10. Tran

Suppresses

  1. v
  2. 跃
  3. alom
  4. k
  5. owell
  6. oval
  7. Kelley
  8. ind
  9. j
  10. ample