Neuron Explorer

Qwen3-0.6B

words after 'will be' or 'subject to' in formal text

A neuron in layer 8 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 fires on small connecting words in formal or policy sentences, such as 'subject to', 'diagnosed with', or 'prosecuted for'. Turning it up did not change the topic in a clear way.

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

    a slip twixt cup and lip there. It will be interesting to see what is in the relevant journal article when -- and if -- it has passed peer review⏎THE first hard evidence is in that global action under the Montreal Protocol to mend

  2. #2 peak

    place marked out by the Lord your God for himself;⏎Gold cannot be exchanged for it,and silver cannot be weighed out for its price.⏎Gold and glass do not compare with it,and articles of fine gold cannot be exchanged for it.⏎They exchanged their

  3. #3 peak

    page remains online for reference, research or recordkeeping purposes. This page will not be altered or updated. Web pages that are archived on the Internet are not subject to the Government of Canada Web Standards. As per the Communications Policy of the Government of

  4. #4 peak

    asked to come back for repeat blood pressure checks. If you have three visits with readings over 140/90 mmHG, you will be diagnosed with high blood pressure.⏎Some people’s blood pressure goes up when they are at the

  5. #5 peak

    that the most important aspect of a post-cubist squarist painting is the largest diameter of any shape. He announces that there is a reward for whoever can determine this in his newest artwork. Of course, Leo's genius won't help

  6. #6 peak

    has exactly the same sentence as the reader sits down, along with the reader. This continues until the students left standing do indeed have sentences that are unique and win a point.⏎I Want – Hand out picture cards or real items that are cognates

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 seasons with the longest periods of time for the longest (most stable) rising tides. The tides when the sun is at its highest point in the sky. What does this include the longest periods of time for the longest (most

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

    local governments. This bill will also establish a federal grant program to fund planning, construction & monitoring of Green Infrastructure projects. Special consideration for these grants will be given to low-income and communities with combined sewer systems. Additionally, this bill will establish a

  2. #2 peak

    .) The virus only infects humans.⏎In older children and adults, RSV usually causes very bad colds

  3. #3 peak

    change that is still being worked out.⏎Readers can also take a more thematic view of American history in the 20th century. Here we will list only the major events and developments.⏎American<|endoftext|>What is an Endodontist McLean

  4. #4 peak

    case of fixed price, you have to factor in costs of production before you come up with your price. With what is variable CPV, you would only pay for the actual item purchased. Cost of good sale is usually included in the pricing.

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. L6:2600 weakly excites
  2. L7:1399 weakly suppresses "to" between a starting and ending value
  3. L6:2191 weakly suppresses
  4. L6:1881 weakly excites
  5. L7:2030 weakly suppresses
  6. L6:2301 weakly suppresses
  7. L7:1493 weakly excites
  8. L3:1303 weakly excites
  9. L1:1753 weakly suppresses
  10. L6:337 weakly excites
  11. L7:2071 weakly suppresses
  12. L2:1870 weakly suppresses
  13. L6:2576 weakly suppresses
  14. L4:1109 weakly suppresses
  15. L7:1921 weakly excites
  16. L5:562 weakly suppresses
  17. L6:1765 weakly excites
  18. L6:2463 weakly suppresses
  19. L6:2122 weakly excites
  20. L5:1806 weakly excites
  21. L6:2270 weakly suppresses 'on' or 'after' before a measured topic
  22. L3:750 weakly excites
  23. L0:731 weakly excites
  24. L5:2694 weakly excites

Feeds later layers

  1. L10:2704 suppresses
  2. L9:3028 suppresses
  3. L12:2303 suppresses
  4. L10:1893 suppresses
  5. L19:2139 suppresses
  6. L20:1737 suppresses
  7. L18:2993 suppresses
  8. L17:2059 suppresses
  9. L17:3002 suppresses
  10. L13:2899 suppresses
  11. L13:2807 excites
  12. L20:1176 suppresses
  13. L15:758 suppresses
  14. L12:1973 weakly suppresses
  15. L15:513 weakly suppresses
  16. L16:1759 weakly suppresses
  17. L17:2931 weakly suppresses
  18. L15:2789 weakly suppresses
  19. L15:1911 weakly suppresses
  20. L20:1533 weakly suppresses
  21. L10:2186 weakly suppresses
  22. L9:2014 weakly suppresses
  23. L18:2757 weakly suppresses
  24. L14:2239 weakly suppresses

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 -2.45 and -2.331 values between -2.21 and -2.101 values between -1.74 and -1.631 values between -1.39 and -1.281 values between -1.28 and -1.161 values between -1.16 and -1.042 values between -1.04 and -0.922 values between -0.92 and -0.814 values between -0.81 and -0.697 values between -0.69 and -0.5715 values between -0.57 and -0.4545 values between -0.45 and -0.34162 values between -0.34 and -0.22499 values between -0.22 and -0.101698 values between -0.10 and 0.021069 values between 0.02 and 0.13277 values between 0.13 and 0.2577 values between 0.25 and 0.3716 values between 0.37 and 0.4812 values between 0.48 and 0.604 values between 0.60 and 0.723 values between 0.72 and 0.842 values between 0.84 and 0.955 values between 0.95 and 1.071 values between 1.66 and 1.771 values between 2.01 and 2.131 values between 2.13 and 2.24offnegativepositiveoften

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. ادر
  2. ático
  3. plex
  4. 寰
  5. PLEX
  6. метр
  7. ibri
  8. arme
  9. mund
  10. plotlib

Suppresses

  1. .TabStop
  2. elight
  3. wherever
  4. tern
  5. those
  6. "/>⏎⏎
  7. opoly
  8. топ
  9. newPos
  10. WHERE