Neuron Explorer

Qwen3-0.6B

split names and comparative stems

A neuron in layer 27 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

The strongest pulls come from two sides. The positive side fires on the first piece of a split place or drug name, like 'Var' in Varanasi. The negative side fires on word stems right before a comparison ending, like 'lik' in likelier or 'swe' in sweeter. When pushed up, the model wrote 'larger' over and over.

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

    Ganges river below:⏎Step 1 Uttarakhand (state + Start)⏎Step 2 Kanpur, Soron, Kannauj, Allahabad, Varanasi, Patna, Ghazipur,Bhagalpur (towns

  2. #2 peak

    centres.⏎Hundreds of bumboats ferrying goods from warehouses along the river to cargo ships out at sea were moved to a new lighter anchorage at Pasir Panjang by the Port of Singapore Authority.⏎Foul-smelling mud also had to

  3. #3 peak

    treated father. Male mice in the study were exposed to one of three common anticancer drugs—cyclophosphamide, mitomycin C, or procarbazine—and then allowed to mate with untreated females. After sequencing a small

  4. #4 peak

    Madison McDermott⏎The Green New Deal Federal Resolution was introduced in 2019 by Rep. Ocasio-Cortez and Sen. Ed Markey as a plan to tackle a trifecta of crises: the climate crisis

  5. #5 peak

    HG (examples: promethazine or prochlorperazine). However, doctors often recommend that women with HG take certain anti

  6. #6 peak

    printed copy of this publication.⏎- Our on-line publications are scanned and captured using Adobe Acrobat. During the capture process some typographical errors may occur. Please contact Sharon Hobrla, email@example.com if you notice any errors which make this publication

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: life has come in stages thanks to life itself, thanks to life beings having come forth thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks thanks

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

    Stress is an underlying factor in illness as it reduces your immune system’s ability to fight off antigens. This means that the more stressed you are, the likelier you are to pick up infections and fight illness.⏎Therefore, a regular workout

  2. #2 peak

    , and can lead to hospitalization, kidney failure, and even death.⏎Why do people drink raw milk?⏎Some people believe that raw milk tastes sweeter and fresher, and that it’s nutritional properties can help various health conditions. Farmers who

  3. #3 peak

    usually start ripening in February. The best method to determine if the fruit is sweet is your own taste-test. The longer fruit stays on the tree, the sweeter it becomes. It will not sweeten after being picked. If you can

  4. #4 peak

    worse and we know this will impact our nature, wildlife, people and food supplies. We urgently need ambitious action to cut our emissions and to build a cleaner, greener economy to tackle climate change before we pass the point of no return that we

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. L20:944 strongly excites
  2. L26:551 strongly excites
  3. L19:2597 strongly excites
  4. L19:2024 strongly excites
  5. L21:2949 strongly suppresses comma closing 'the more X, the more Y' clauses
  6. L20:2321 strongly excites
  7. L22:1988 strongly suppresses "the" in "the more..., the more" phrases
  8. L20:2491 strongly suppresses
  9. L24:2340 strongly excites
  10. L20:2323 strongly excites
  11. L23:456 strongly suppresses
  12. L24:847 strongly excites
  13. L20:2852 excites
  14. L23:751 excites
  15. L25:3025 excites
  16. L25:657 suppresses
  17. L26:124 excites
  18. L18:2761 excites
  19. L21:2031 excites
  20. L18:2316 excites
  21. L15:1060 suppresses
  22. L19:739 excites
  23. L23:2817 excites the word "no" in "no longer"
  24. L17:2278 excites

Feeds later layers

None: this is the last layer.

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 -22.44 and -21.741 values between -21.74 and -21.051 values between -19.66 and -18.961 values between -16.88 and -16.191 values between -15.49 and -14.801 values between -14.80 and -14.101 values between -12.71 and -12.024 values between -9.93 and -9.242 values between -9.24 and -8.541 values between -8.54 and -7.851 values between -7.85 and -7.154 values between -7.15 and -6.464 values between -5.76 and -5.073 values between -5.07 and -4.382 values between -4.38 and -3.682 values between -3.68 and -2.9913 values between -2.99 and -2.2924 values between -2.29 and -1.60122 values between -1.60 and -0.901882 values between -0.90 and -0.211675 values between -0.21 and 0.4985 values between 0.49 and 1.1842 values between 1.18 and 1.8817 values between 1.88 and 2.578 values between 2.57 and 3.274 values between 3.27 and 3.964 values between 3.96 and 4.662 values between 4.66 and 5.35offnegativepositiveoften

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. larger
  2. 更大
  3. 更大的
  4. longer
  5. 更
  6. wider
  7. bigger
  8. 更高的
  9. 更高
  10. broader

Suppresses

  1. orphism
  2. ibur
  3. ‾
  4. icious
  5. яет
  6. thanks
  7. lobe
  8. CHASE
  9. دة
  10. renched