Neuron Explorer

Qwen3-0.6B

mixed, digits in URLs and comparisons

A neuron in layer 15 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

It is a loose mix. It fires on the month digit inside web addresses and dates, and on words like 'than' or 'of' in comparison phrases such as 'greater than' or 'previous decades'. No single theme is clear, and turning it up made the model drift into numbers and calculations.

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

    laminar flow cabinets, and other biological safety cabinet components protect.⏎Read More @ https://cmibloginsight.blogspot.com/2023/05/biological-safety-cabinets-ensuring.html<|endoftext|>World

  2. #2 peak

    .⏎This week’s article by Jeremy Rosen on Rav Lichtenstein is available from http://jeremyrosen.blogspot.co.uk/2015/08/rav-lichtenstein-zl.html⏎Your comment will be posted after it

  3. #3 peak

    communities of Mbeliling, western Flores, are highly vulnerable to crop failure. With higher temperatures and decreased precipitation predicted as a result of climate change, the frequency of such events is likely to increase. Through agroforestry and organic farming,

  4. #4 peak

    <|endoftext|>Pharmacists are caring for more individuals of diverse age, gender, race, ethnicity, socioeconomic status, religion, sexual orientation, and health beliefs than in previous decades. Not all residents of the United States equally experience long life spans and good

  5. #5 peak

    s⏎With recent introduction of new APIs and tools for the Java Server Pages (JSP) environment, marrying XML data to Java-based services and applications is easier than ever before. The September 17 release of the JavaServer Pages 1

  6. #6 peak

    trips, which are the kind that form the most contrails. So by 2050, contrails will have a much greater impact on global warming than CO2 emissions from aircraft engines. This means it may be worthwhile for planes to

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: I don. The following is a part of this. Please explain this sentence. The following is a part of this paragraph, and I will explain. I think it. for example, the first paragraph. So,

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

    to produce and market a pound of butter now than it did in 1940 and somewhere between 5

  2. #2 peak

    in any rush to get there.⏎Meet future woman: shorter, plumper, more fertile [New Scientist]<|endoftext|>MERCIER, JEAN ° (Joannes Mercerus ; d. 1570), French Hebraist.

  3. #3 peak

    during their detention, and the impact that it had in their lives after the recovery of democracy in 1983.⏎The Navy Mechanical School (Escuela de Mecánica de la Armada

  4. #4 peak

    a Confederate army hospital during the Civil War. She became something of a legend by way of the ironic and insightful memoir she published of her experiences.⏎- This Day in Jewish History / A ship carrying 152 Jews who had fled Spain sails

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. L14:126 strongly suppresses
  2. L14:1519 excites
  3. L7:8 excites
  4. L14:830 suppresses
  5. L14:659 suppresses
  6. L14:727 suppresses
  7. L8:892 excites
  8. L14:312 excites
  9. L10:272 suppresses
  10. L13:700 excites
  11. L14:2235 suppresses
  12. L13:68 excites
  13. L14:2075 suppresses
  14. L14:313 suppresses
  15. L13:1729 excites
  16. L9:330 suppresses gap where a number was removed
  17. L14:280 excites
  18. L13:1100 excites
  19. L13:287 suppresses
  20. L9:459 excites
  21. L14:655 excites
  22. L13:3 excites
  23. L12:1531 excites
  24. L14:417 suppresses

Feeds later layers

  1. L18:606 excites
  2. L20:60 suppresses
  3. L17:31 suppresses
  4. L16:786 excites
  5. L16:1705 excites
  6. L17:1783 excites
  7. L18:132 suppresses
  8. L25:98 suppresses
  9. L16:2063 excites
  10. L19:152 suppresses
  11. L18:134 weakly suppresses
  12. L20:17 weakly suppresses
  13. L20:113 weakly suppresses
  14. L16:512 weakly suppresses
  15. L16:1911 weakly excites
  16. L19:1336 weakly excites
  17. L16:2147 weakly excites
  18. L20:170 weakly suppresses
  19. L16:875 weakly suppresses
  20. L19:761 weakly excites
  21. L16:143 weakly suppresses
  22. L18:376 weakly excites
  23. L16:1607 weakly excites
  24. L16:1380 weakly 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.

1 values between -3.83 and -3.641 values between -3.64 and -3.461 values between -3.28 and -3.092 values between -3.09 and -2.911 values between -2.91 and -2.731 values between -2.54 and -2.361 values between -2.36 and -2.188 values between -2.18 and -1.995 values between -1.99 and -1.8114 values between -1.81 and -1.6316 values between -1.63 and -1.4418 values between -1.44 and -1.2636 values between -1.26 and -1.0844 values between -1.08 and -0.8963 values between -0.89 and -0.7187 values between -0.71 and -0.53175 values between -0.53 and -0.34318 values between -0.34 and -0.16671 values between -0.16 and 0.02701 values between 0.02 and 0.21536 values between 0.21 and 0.39390 values between 0.39 and 0.57279 values between 0.57 and 0.76186 values between 0.76 and 0.94142 values between 0.94 and 1.1277 values between 1.12 and 1.3156 values between 1.31 and 1.4935 values between 1.49 and 1.6715 values between 1.67 and 1.869 values between 1.86 and 2.048 values between 2.04 and 2.224 values between 2.22 and 2.414 values between 2.41 and 2.591 values between 2.77 and 2.961 values between 2.96 and 3.141 values between 3.32 and 3.51offnegativepositiveoften

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. eds
  2. 乎
  3. iesz
  4. ॰
  5. theless
  6. .asp
  7. ests
  8. sts
  9. ester
  10. )__

Suppresses

  1. ⏎
  2. 持
  3. ta
  4. FromNib
  5. file
  6. ⏎⏎
  7. 感
  8. .mean
  9. 在校
  10. licht