Neuron Explorer

Qwen3-0.6B

closing quote marks after a quoted term

A neuron in layer 1 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron responds to the punctuation that closes a quotation, like a full stop or comma placed inside a closing quote mark, and it reacts strongly against quote marks around a term such as 'Standard' or 'DOSBox'. The writing it produced when turned up did not clearly show this theme.

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

    uguese translation of 'discontent'⏎Copyright © 2014 by HarperCollins Publishers. All rights reserved.⏎Example Sentences Including 'discontent'⏎Marriage, discontent with home, desire for pastures new, departure.Time of

  2. #2 peak

    th Circuit Court<|endoftext|>Restoring the past – York Minster⏎Tags: 'The Cathedral and Metropolitan Church of St. Peter', 'York Minster Revealed', 1230 - 1472,

  3. #3 peak

    just one of an infinite number of universes that make up the "multiverse." In this show, Brian Greene takes us on a tour of this brave new theory

  4. #4 peak

    ta⏎called a jubba. Both of these clothes are called "kurtas." In the state⏎of Andhra Pradesh, both of these options are common.

  5. #5 peak

    is a blog post on how to hook up or connect your battery. This article will be broken into two parts, the "How-to" and the "Why." The How-to section features detailed instructions with illustrations, including what tools you'll need

  6. #6 peak

    put forth a host of initiatives aimed at improving the region’s defense and security. These proposals are wide-ranging, covering areas from defense spending to a “European Army.” While efforts to create improvements in defense capabilities among European states are commendable, these

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: " " " and so on. " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " "

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

    and Configuration of Drupal⏎Open Browser and type “”http://Server-IP/drupal”” which will open below screen to configure Drupal. Select ‘Standard‘, click on Save and Continue.⏎Choose language, click on Save and Continue⏎

  2. #2 peak

    it and paste it at the following path⏎Now open Run (ctrl+R) and type your command name(Here my command name is “DOSBox“) and hit enter, your application will open. Every time you can do that

  3. #3 peak

    it. The German term for this type of game adds the word for “table”, “Tisch” (pronounce tish) to “Fußball“ – Tischfußball. Back in the day, you’d find those

  4. #4 peak

    what you perceive as a German term – “Foosball” – is actually a misnomer and misspelled, at that. “Fußball“ (pronounce ‘foos-bull rhyming with “hull”)

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. L0:178 suppresses
  2. L0:1224 suppresses
  3. L0:1301 suppresses
  4. L0:1690 weakly suppresses
  5. L0:1 weakly suppresses
  6. L0:2913 weakly suppresses
  7. L0:856 weakly excites
  8. L0:97 weakly excites
  9. L0:2818 weakly suppresses
  10. L0:749 weakly suppresses
  11. L0:517 weakly suppresses
  12. L0:952 weakly excites
  13. L0:37 weakly excites
  14. L0:318 weakly excites
  15. L0:2701 weakly suppresses
  16. L0:868 weakly suppresses
  17. L0:1023 weakly excites
  18. L0:1379 weakly excites
  19. L0:958 weakly excites
  20. L0:192 weakly excites
  21. L0:1814 weakly excites
  22. L0:1944 weakly suppresses
  23. L0:1674 weakly suppresses
  24. L0:2424 weakly excites

Feeds later layers

  1. L12:2977 weakly suppresses year in a written date
  2. L14:2625 weakly excites
  3. L20:2935 weakly excites
  4. L5:1367 weakly excites
  5. L14:2506 weakly suppresses
  6. L8:1996 weakly excites
  7. L2:1669 weakly suppresses
  8. L14:1288 weakly suppresses
  9. L10:818 weakly suppresses
  10. L5:471 weakly excites
  11. L19:1552 weakly excites
  12. L14:693 weakly suppresses
  13. L11:703 weakly suppresses
  14. L8:261 weakly suppresses
  15. L16:693 weakly suppresses
  16. L20:580 weakly suppresses
  17. L3:1056 weakly excites
  18. L12:767 weakly excites
  19. L12:2689 weakly suppresses
  20. L12:170 weakly excites
  21. L10:1816 weakly suppresses
  22. L18:219 weakly suppresses
  23. L20:1692 weakly excites
  24. L13:460 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 -7.58 and -7.371 values between -2.37 and -2.161 values between -1.95 and -1.741 values between -1.74 and -1.531 values between -1.33 and -1.122 values between -1.12 and -0.913 values between -0.91 and -0.702 values between -0.70 and -0.492 values between -0.49 and -0.28129 values between -0.28 and -0.073678 values between -0.07 and 0.1361 values between 0.13 and 0.3410 values between 0.34 and 0.5516 values between 0.55 and 0.76offnegativepositiveoften

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. ’am
  2. sing
  3. $")⏎
  4. ”)
  5. leştir
  6. 'am
  7. "`⏎
  8. Leap
  9. 花生
  10. ) ⏎

Suppresses

  1. icial
  2. inclusive
  3. ศาส
  4. 레이
  5. куп
  6. 皮肤
  7. ↑
  8. engr
  9. ори
  10. GFX