Neuron Explorer

Qwen3-0.6B

periods ending explanations

A neuron in layer 19 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

The neuron fires on the full stop that ends an explanatory sentence, the kind that wraps up a point before the next one begins, as in describing how a method works. Turning it up produced repeated filler like 'The Most Of The Most', so the steering does not clearly confirm the pattern.

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

    aspects of AI is its ability to learn and adapt. Machine learning algorithms can analyze vast amounts of data to identify patterns and make predictions, and they improve with experience. This capability has applications

  2. #2 peak

    the maximum possible mark. Once you’ve determined the maximum, you can also divide it by the total marks. This will give you the percentage of a given number. In addition to finding the actual value of a number, you should<|endoftext|>Early Men

  3. #3 peak

    x, or %. Then, multiply the result by 100 to get the percent. This method is also useful for finding the exact value of the number. Once you have the answer, you can move onto the next step in interpreting the

  4. #4 peak

    servlets constructed to invoke such libraries. In other words, a tag library defines a set of Java classes for the custom JSP actions that the tag library supports. Invoking the tag library imports custom processing and actions into the Java environment so a

  5. #5 peak

    9. These are the "stems". For each value in the dataset, we pair its right-most digit to the corresponding stem that matches it's left digit. For example, if we had a⏎76 in our dataset, we would

  6. #6 peak

    The calculator will do the rest, including converting the decimal to a percentage.⏎To find the percentage value of a mark, you’ll need to know its total mark. In order to find this, use basic arithmetic. Divide the mark by the highest

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:. P.............................................

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

    women are to the manufacturing sector and how companies are working towards inclusivity. As a woman myself, doing research on dissolving the gender gap is incredibly rewarding. Over the past few decades, the United States has made remarkable improvements in empowering women of

  2. #2 peak

    using a compass, just position the point at the center of the cross, and change the radius ever so slightly to arrive at the desired image ahead of time.⏎Now for the lines!⏎Second Step - Very Simple! Very Easy!⏎Our outline for

  3. #3 peak

    .⏎Still, despite the fact that buckwheat flour is highly popular, it’s important to make sure that its potential side effects aren’t being overlooked.⏎Out of all the healthy grains out there, buckwheat has the highest levels of amino

  4. #4 peak

    Be Used for Armor, More.")⏎The discovery may inspire performance engineers to create better and more robust vehicles—such as bicycles, boats, and race cars—in the not-so-distant future.⏎"It's about translating design principles found in limpets to

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. L16:432 suppresses period after month or initial
  2. L11:132 excites period ending an opening topic sentence
  3. L5:168 suppresses
  4. L9:16 suppresses unclear: periods inside numbers and names
  5. L17:347 excites full stops ending informative sentences
  6. L15:403 suppresses period inside an abbreviation
  7. L6:2302 suppresses 'and' joining two parallel items
  8. L13:53 excites period ending a claim before elaboration
  9. L18:791 suppresses
  10. L4:7 excites
  11. L17:2414 weakly suppresses
  12. L3:259 weakly suppresses
  13. L10:612 weakly excites full stop ending boilerplate or a title
  14. L18:227 weakly suppresses
  15. L8:541 weakly excites
  16. L18:2434 weakly excites
  17. L5:155 weakly suppresses
  18. L16:489 weakly suppresses
  19. L2:156 weakly excites
  20. L14:155 weakly excites sentence-ending periods in prose
  21. L18:366 weakly suppresses comma inside numbers and dates
  22. L8:2262 weakly suppresses
  23. L17:1054 weakly suppresses
  24. L16:781 weakly excites

Feeds later layers

  1. L21:1086 excites end of sentence before a claim or statistic
  2. L23:179 excites
  3. L20:296 excites
  4. L25:217 excites
  5. L23:278 excites
  6. L27:451 excites
  7. L26:1200 excites
  8. L24:353 excites decimal point inside a number
  9. L26:448 excites
  10. L23:509 excites
  11. L25:198 excites
  12. L27:133 excites
  13. L23:561 excites
  14. L20:473 excites
  15. L20:343 excites
  16. L27:452 excites
  17. L21:201 excites
  18. L26:299 excites
  19. L24:421 excites
  20. L21:524 excites
  21. L27:854 excites
  22. L27:508 excites
  23. L27:640 excites 'of' inside fixed titles and names
  24. L26:517 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 -10.93 and -9.474 values between -8.01 and -6.5512 values between -6.55 and -5.0932 values between -5.09 and -3.6389 values between -3.63 and -2.17523 values between -2.17 and -0.712956 values between -0.71 and 0.75134 values between 0.75 and 2.2113 values between 2.21 and 3.6711 values between 3.67 and 5.1311 values between 5.13 and 6.598 values between 6.59 and 8.0510 values between 8.05 and 9.515 values between 9.51 and 10.978 values between 10.97 and 12.436 values between 12.43 and 13.891 values between 13.89 and 15.354 values between 15.35 and 16.811 values between 16.81 and 18.271 values between 18.27 and 19.732 values between 19.73 and 21.191 values between 21.19 and 22.653 values between 22.65 and 24.116 values between 24.11 and 25.577 values between 25.57 and 27.035 values between 27.03 and 28.4910 values between 28.49 and 29.954 values between 29.95 and 31.4110 values between 31.41 and 32.8711 values between 32.87 and 34.337 values between 34.33 and 35.794 values between 35.79 and 37.252 values between 37.25 and 38.714 values between 40.17 and 41.631 values between 46.01 and 47.47offnegativepositiveoften

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. 这对于
  2. 但这
  3. 这是
  4. 这是因为
  5. 这
  6. Here
  7. For
  8. These
  9. конечно
  10. 这里面

Suppresses

  1. -hearted
  2. -muted
  3. -way
  4. grave
  5. -price
  6. quisites
  7. /off
  8. quisite
  9. yte
  10. -dependent