Neuron Explorer

Qwen3-0.6B

period ending a claim before elaboration

A neuron in layer 13 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 full stop that closes a sentence which introduces a point, right before the writer explains it, such as the end of a sentence saying there are disadvantages. Turning it up made the text loop on repeated phrases, which is not really about the trigger.

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

    hard work can often lead to greater mastery of skills, as repetition and practice can lead to improvement over time.⏎However, there are also several disadvantages to hard work. First, it can be time-consuming, leading to burnout and exhaustion. Second

  2. #2 peak

    components aids in pinpointing the issue quickly, ensuring timely and accurate repairs or replacements.⏎The Importance of Regular Maintenance⏎Maintenance is vital for the longevity of HVAC systems. This includes routine tasks such as replacing air filters, cleaning coils, inspecting duct

  3. #3 peak

    coding skills have become invaluable for developers looking to make a mark in the industry.⏎The Challenges of Coding for Apps⏎Coding for apps presents its fair share of challenges. Developers must understand the unique requirements of the

  4. #4 peak

    , saving energy and reducing utility costs.⏎Choosing the Right Replacement Parts⏎When it comes to repairing or upgrading an HVAC system, choosing the correct replacement parts is critical. Quality parts ensure compatibility with the system and extend its lifespan. For example, using

  5. #5 peak

    any health risks.⏎Is PET Plastic Safe for the Environment?⏎There have also been concerns about the potential for PET bottles to release harmful chemicals when they are disposed of. Like all plastics, PET bottles are made from non-renewable resources and can

  6. #6 peak

    benefits of good nutrition are many. Not only will it help you maintain a healthy weight and help you build better immunity but it will do a lot more for you. Let’s understand what is that.⏎Good Nutrition Improves Overall Health⏎Taking a

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: of the the, where the, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

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

    a matrix⏎Find nth maximum⏎Matlab Basics II - Create a vector with a repeated entry⏎Create sequnce 1

  2. #2 peak

    void draw_model(void *, float, float, float);⏎void deallocate_model(void *);⏎And that is why I am a header Nazi. Here's how I do<|endoftext|>When you work with people, conflict is bound to happen. The dictionary defines conflict

  3. #3 peak

    damage to the bird, sometimes even leading to death. Here, the bird is carefully manipulating the fish so that it enters its buccal cavity head first.<|endoftext|>In a previous lesson, you used the % (

  4. #4 peak

    target values⏎removes empty columns⏎imputes missing values with mean values⏎To remove default preprocessing, connect an empty Preprocess widget to the learner.<|endoftext|>Start Date: September 2012⏎Robert Waymouth and Christopher Ch

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. L11:132 excites period ending an opening topic sentence
  2. L9:2302 excites
  3. L9:2447 excites
  4. L9:16 suppresses unclear: periods inside numbers and names
  5. L5:155 suppresses
  6. L3:854 weakly suppresses
  7. L8:541 weakly excites
  8. L4:2964 weakly excites
  9. L8:2520 weakly suppresses
  10. L6:542 weakly suppresses
  11. L2:741 weakly excites
  12. L8:1848 weakly excites
  13. L12:1950 weakly excites line break after a section heading
  14. L1:610 weakly suppresses
  15. L3:816 weakly excites
  16. L5:665 weakly suppresses
  17. L3:466 weakly suppresses
  18. L9:257 weakly suppresses
  19. L8:361 weakly excites
  20. L2:1445 weakly excites
  21. L12:1062 weakly excites
  22. L4:848 weakly excites
  23. L6:305 weakly suppresses
  24. L3:454 weakly suppresses

Feeds later layers

  1. L16:1545 excites
  2. L25:198 excites
  3. L19:384 excites
  4. L17:200 excites
  5. L17:347 excites full stops ending informative sentences
  6. L17:1155 excites opening parenthesis after a sentence
  7. L20:296 excites
  8. L27:451 excites
  9. L20:473 excites
  10. L19:2697 excites
  11. L19:2097 excites sentence-end punctuation in speech
  12. L17:1105 excites
  13. L19:344 excites periods ending explanations
  14. L15:1207 excites
  15. L27:157 excites
  16. L15:2235 suppresses
  17. L19:1072 excites
  18. L27:818 excites inside names of journals and institutions
  19. L19:527 excites
  20. L27:133 excites
  21. L27:406 excites
  22. L27:293 excites
  23. L27:371 excites
  24. L27:640 excites 'of' inside fixed titles and names

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 -5.60 and -5.381 values between -4.93 and -4.711 values between -4.71 and -4.492 values between -4.04 and -3.821 values between -3.82 and -3.601 values between -3.15 and -2.932 values between -2.93 and -2.712 values between -2.71 and -2.494 values between -2.49 and -2.271 values between -2.27 and -2.047 values between -2.04 and -1.824 values between -1.82 and -1.6010 values between -1.60 and -1.389 values between -1.38 and -1.1511 values between -1.15 and -0.9318 values between -0.93 and -0.7156 values between -0.71 and -0.49407 values between -0.49 and -0.271575 values between -0.27 and -0.041566 values between -0.04 and 0.18179 values between 0.18 and 0.4019 values between 0.40 and 0.6211 values between 0.62 and 0.857 values between 0.85 and 1.073 values between 1.07 and 1.295 values between 1.29 and 1.511 values between 1.51 and 1.732 values between 1.96 and 2.181 values between 2.18 and 2.401 values between 3.07 and 3.29offnegativepositiveoften

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. preter
  2. lator
  3. itud
  4. weit
  5. 的情
  6. erti
  7. imed
  8. ’S
  9. çi
  10. Knowing

Suppresses

  1. 收获
  2. <?,
  3. {};⏎
  4. ❯
  5. ��
  6. mi
  7. .#
  8. 抟
  9. 灼
  10. ++;⏎