Neuron Explorer

Qwen3-0.6B

unclear: sentence-ending periods

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

Fires on the full stop that closes a sentence in ordinary prose, such as the end of a factual claim or a short explanation. It does not seem to care much about the topic. Turning it up made the model drift into repetitive, unrelated text, so the steering results say little about its role.

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

    II), men were conscripted to battle, leaving fields and livestock unattended. Who stepped in? Women, of course (cause that's what we do). However, we couldn't do it all,

  2. #2 peak

    who prefer to read about a concept to understand it.⏎- Kinesthetic leaners, who like to use movement or manipulate objects in a "hand on" approach.⏎[NOTE: The word kinesthesia means "muscle sense," which is your

  3. #3 peak

    to establish a federal holiday in 1894. President Grover Cleveland signed the bill soon afterward, designating the first Monday in September as Labor Day.⏎Learners are Leaders, Inc.’s mission includes uplifting future leaders through learning

  4. #4 peak

    initiated a new and comprehensive status review which was finalized as the 2018 SSA. Based on the SSA findings, the Proposed Listing was officially withdrawn in 2020.⏎Idaho has a vested interest in wolverine conservation

  5. #5 peak

    gets filled with water, the air pressure build. As the air pressure builds it pushes a turbine to start spinning. As this turbine starts spinning, electricity is created. Whats also cool, is as the water leaves the compartment, the same thing happens

  6. #6 peak

    is converted into electric signals through the Optic Nerve behind our eyes and sent to our brain. Our brain then creates the beautiful pictures that we see before us. Our eyes are well-oiled machines (metaphorically speaking, of course

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 question of how to define the function f(x) in the context of the problem, and the function of the idea of defining, and the problem of the function, and the need for the problem, and the idea of the problem,

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

    are often called Paste tomatoes, but not all plum shapes have the meaty interior of a paste tomato. Some varieties can grow quite long like a horned pepper. Fleshy ones, like the D.O.P protected San Marzano

  2. #2 peak

    fact, their bowel movements are regular. For example, some people believe they are constipated, or irregular, if they do not have a bowel movement every day. However, there is no right number of daily

  3. #3 peak

    as a victim of identity theft when you make purchases on websites that aren't secure. There are lots of small online retailers that don't have adequately secure payment systems. The best way to make sure that your information doesn't get intercepted is by simply

  4. #4 peak

    rosettes often have a bit more white streaking along veins than nodding thistle, though nodding thistle can have whitish leaf margins at times, as shown in the image below. Variegated thistle has much more white

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. L9:456 suppresses
  2. L12:2677 excites
  3. L11:2220 suppresses
  4. L12:2436 excites
  5. L8:149 excites
  6. L8:430 excites
  7. L12:1824 weakly suppresses
  8. L12:1011 weakly suppresses
  9. L8:319 weakly excites
  10. L9:12 weakly suppresses
  11. L9:682 weakly suppresses
  12. L7:676 weakly excites
  13. L11:46 weakly suppresses
  14. L3:182 weakly excites
  15. L2:92 weakly suppresses
  16. L11:1025 weakly suppresses
  17. L7:333 weakly excites
  18. L9:694 weakly suppresses
  19. L10:404 weakly excites
  20. L9:1802 weakly excites
  21. L9:16 weakly suppresses unclear: periods inside numbers and names
  22. L7:205 weakly excites
  23. L7:1197 weakly excites
  24. L0:933 weakly suppresses

Feeds later layers

  1. L24:1280 weakly suppresses
  2. L24:2748 weakly suppresses
  3. L14:2014 weakly suppresses
  4. L14:46 weakly suppresses
  5. L19:1126 weakly excites
  6. L24:1685 weakly excites
  7. L17:2164 weakly suppresses
  8. L20:873 weakly suppresses
  9. L25:1255 weakly suppresses
  10. L19:278 weakly suppresses
  11. L14:1931 weakly suppresses
  12. L15:2749 weakly suppresses decimal point or digit inside a number
  13. L15:2314 weakly suppresses
  14. L15:527 weakly excites
  15. L23:906 weakly suppresses
  16. L17:74 weakly excites
  17. L23:3030 weakly suppresses 'American' in organization names
  18. L25:2162 weakly suppresses
  19. L15:1291 weakly excites
  20. L22:2230 weakly suppresses
  21. L22:1612 weakly excites
  22. L14:241 weakly excites
  23. L24:2515 weakly suppresses
  24. L15:2597 weakly excites end of a web document

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 -2.20 and -2.083 values between -1.57 and -1.453 values between -1.45 and -1.324 values between -1.32 and -1.195 values between -1.19 and -1.074 values between -1.07 and -0.9413 values between -0.94 and -0.8213 values between -0.82 and -0.6924 values between -0.69 and -0.5636 values between -0.56 and -0.4474 values between -0.44 and -0.31154 values between -0.31 and -0.19613 values between -0.19 and -0.062140 values between -0.06 and 0.07564 values between 0.07 and 0.19155 values between 0.19 and 0.3238 values between 0.32 and 0.4518 values between 0.45 and 0.5718 values between 0.57 and 0.705 values between 0.70 and 0.827 values between 0.82 and 0.954 values between 0.95 and 1.085 values between 1.08 and 1.203 values between 1.45 and 1.583 values between 1.83 and 1.961 values between 2.72 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. anas
  2. ark
  3. 一
  4. 표
  5. uzz
  6. ира
  7. 真
  8. inter
  9. 訪
  10. ana

Suppresses

  1. oleon
  2. adına
  3. ulent
  4. nonetheless
  5. ForObject
  6. 獒
  7. ityEngine
  8. nage
  9. 感觉自己
  10. .Designer