Neuron Explorer

Qwen3-0.6B

file formats and technical jargon

A neuron in layer 7 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 lights up on technical words, mostly about file formats (AVI, MP4, headers), archives and disk structure. It is not clear what ties them together. Turning it up made the model drift into lists of technical terms and file descriptions.

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

    . Other file formats can be read by creating an NMRView parameter file that contains a description of the layout of the file.⏎|Program||Format||File Header||Block Header||Parameters|⏎|Varian||Serial||32/2

  2. #2 peak

    of MPEG-4.⏎||Moving Picture Experts Group⏎||Flexible, works with a wide range of editing and compression programs; have support for multiple video or audio streams⏎||MP4 is are usually compressed; compatible with more devices like iPad,

  3. #3 peak

    ||MPEG-4 Video File⏎||AVI is a video wrapper format created by Microsoft, storing video data that might be encoded with various codecs. AVI files can be played by various video players, but the player must support the codec

  4. #4 peak

    you highlighted a title with the data, you would use this option to prevent the program from sorting your title into the list. (In other words, the first row stays in place, regardless of its first letter).⏎When you have everything set the

  5. #5 peak

    the player must support the codec used to encode the video data.⏎||MP4 files are an open standard that everyone can use.⏎|Programs that open AVI/MP4 files<|endoftext|>Pioneer Papers connects you with professional substantive editors to build

  6. #6 peak

    .tar utility.⏎- x – Flag is used to extract the zipped files⏎- v – Flag is for verbose, or creating a list of extracted

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 blog. I am a bit of a fan of the blog, and I have been thinking about this. The blog is popular, and I have been thinking about this. Let's say I am a big fan of the blog, and

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

    with a little water and use the sidewalk, cardboard or wood fence. Compare and contrast the textures and patterns the natural bristles make.⏎British artist Andy Goldsworthy creates art using natural materials. He creates his sculptures outside and leaves them in his

  2. #2 peak

    data on on sicklefins, tagged them, and collected sperm and eggs for captive rearing. For more photos, visit: https://www.flickr.com/photos/usfwssoutheast/albums/72157667

  3. #3 peak

    concentrations below 450 parts per million by 2100. Alternative energies will only get humankind so far, and a vast removal of CO2 will be an important tool in the race to remove the gas from the atmosphere.⏎

  4. #4 peak

    hard coal and 276 Gt of lignite. These reserves would result in ~2,900 Gt of CO2 if combusted unabated, with approximately two thirds of this coming from the hard

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. L6:1933 excites
  2. L6:2095 weakly suppresses
  3. L6:1758 weakly excites
  4. L5:513 weakly excites
  5. L5:2640 weakly suppresses
  6. L6:2793 weakly excites
  7. L6:1573 weakly suppresses
  8. L6:1760 weakly excites data quality and data formatting
  9. L6:1131 weakly suppresses
  10. L1:1971 weakly excites
  11. L3:2896 weakly excites
  12. L6:2409 weakly excites
  13. L3:2140 weakly suppresses
  14. L6:2033 weakly suppresses
  15. L3:2551 weakly suppresses
  16. L0:1731 weakly suppresses
  17. L2:2577 weakly suppresses
  18. L3:1744 weakly excites
  19. L2:2843 weakly excites
  20. L6:1158 weakly suppresses
  21. L2:2026 weakly excites
  22. L0:2953 weakly excites
  23. L1:2663 weakly suppresses
  24. L6:2784 weakly excites

Feeds later layers

  1. L20:1674 weakly excites
  2. L20:967 weakly excites
  3. L9:3027 weakly suppresses
  4. L23:1087 weakly suppresses
  5. L20:732 weakly excites
  6. L9:1481 weakly excites
  7. L10:1629 weakly suppresses
  8. L10:1864 weakly excites
  9. L12:2585 weakly excites
  10. L10:1446 weakly suppresses
  11. L18:3061 weakly suppresses
  12. L11:2203 weakly excites
  13. L10:3006 weakly excites
  14. L17:2670 weakly excites
  15. L10:2564 weakly excites
  16. L9:1636 weakly suppresses
  17. L20:678 weakly excites
  18. L12:2472 weakly excites
  19. L24:1899 weakly excites
  20. L17:2352 weakly suppresses
  21. L23:836 weakly suppresses
  22. L10:1724 weakly excites
  23. L11:640 weakly suppresses
  24. L10:1121 weakly suppresses 'to' or verb after 'what' or 'whether'

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 -2.75 and -2.661 values between -2.39 and -2.311 values between -1.59 and -1.502 values between -1.06 and -0.974 values between -0.70 and -0.616 values between -0.61 and -0.524 values between -0.52 and -0.4319 values between -0.43 and -0.3454 values between -0.34 and -0.26154 values between -0.26 and -0.17488 values between -0.17 and -0.081144 values between -0.08 and 0.011085 values between 0.01 and 0.10567 values between 0.10 and 0.19264 values between 0.19 and 0.2878 values between 0.28 and 0.3724 values between 0.37 and 0.467 values between 0.46 and 0.553 values between 0.55 and 0.641 values between 0.64 and 0.731 values between 0.73 and 0.81offnegativepositiveoften

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. renderer
  2. oser
  3. 内容
  4. .header
  5. ako
  6. فس
  7. brief
  8. tokenizer
  9. unpack
  10. utenberg

Suppresses

  1. mma
  2. ью
  3. ��
  4. imity
  5. -append
  6. ところ
  7. ERRUPT
  8. ALLY
  9. 海滨
  10. embro