Neuron Explorer

Qwen3-0.6B

unclear: weak, articles before nouns

A neuron in layer 10 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron gives only weak responses, mostly on small words like 'the', 'a' or 'some' that come just before a noun in the middle of a sentence. It does not clearly change the writing when turned up or down.

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

    organize and prioritize the goals to be accomplished.⏎Popular methods of time management practiced in the 21

  2. #2 peak

    some extra time to make a decision. It now hopes to make a launch decision on Monday morning.⏎“We have

  3. #3 peak

    One of the most important requirements is to sample where irradiation and oxidation are minimized, either by

  4. #4 peak

    One of the items is coffee, but you can’t think of the other one to save your

  5. #5 peak

    . Here are some great examples in a musical format.⏎We will be looking at numbers next time then

  6. #6 peak

    first of the series of articles I introduced last week about the nutritional habits of various unique groups

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 first, the second, or, or, or, or, or, or, or, or, or, or, or, or, or, or, or, or, or, or, or, or, or,

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

    Academy of Sciences, enlarges a growing body of work linking triclosan to human and environmental health issues. In “test tube” experiments, triclosan impairs the ability of isolated heart muscle cells and skeletal muscle fibers to contract

  2. #2 peak

    use in Europe until silk mesh was more available for trade from the east and a profitable outlet for the medium discovered.⏎Screen printing was first patented in England by Samuel Simon in 1907. It was originally used as a popular method to

  3. #3 peak

    the best available top probiotic foods.⏎Probiotic Drinks List⏎Probiotic drinks include:-⏎– kefir (fermented milk)⏎– water ke

  4. #4 peak

    tumors in the brain, including:⏎Herniation of the brain can also be caused by other factors that lead to increased pressure inside the skull, including:⏎Brain herniation can occur:⏎Patients with a brain herniation have:⏎A brain and nervous

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. L9:15 strongly suppresses
  2. L5:25 strongly suppresses
  3. L5:36 strongly excites
  4. L8:7 suppresses
  5. L6:7 excites
  6. L0:18 excites
  7. L8:10 suppresses
  8. L0:2331 suppresses
  9. L8:3 suppresses
  10. L7:122 suppresses
  11. L8:4 suppresses
  12. L8:13 excites
  13. L4:23 suppresses
  14. L8:16 excites
  15. L6:16 suppresses
  16. L5:4 suppresses
  17. L7:63 suppresses
  18. L6:13 suppresses
  19. L9:5 suppresses
  20. L8:14 suppresses
  21. L5:50 suppresses
  22. L3:59 suppresses
  23. L6:18 suppresses
  24. L7:62 suppresses

Feeds later layers

  1. L12:2159 strongly suppresses
  2. L11:1353 strongly suppresses end-of-document token after a web snippet
  3. L11:48 strongly suppresses
  4. L12:1215 strongly suppresses
  5. L11:957 strongly excites
  6. L12:1499 strongly suppresses
  7. L12:396 strongly excites
  8. L11:34 strongly excites
  9. L12:640 strongly suppresses
  10. L16:192 strongly suppresses
  11. L11:9 strongly excites
  12. L19:1247 suppresses
  13. L15:2763 suppresses
  14. L12:1079 suppresses
  15. L13:654 suppresses
  16. L11:70 suppresses
  17. L18:596 suppresses
  18. L13:105 excites
  19. L15:514 excites
  20. L13:1044 suppresses
  21. L19:1299 suppresses
  22. L15:1278 suppresses
  23. L12:623 excites
  24. L14:68 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.

4 values between -3.04 and -2.926 values between -2.92 and -2.813 values between -2.81 and -2.6911 values between -2.69 and -2.5819 values between -2.58 and -2.4625 values between -2.46 and -2.3434 values between -2.34 and -2.2343 values between -2.23 and -2.1151 values between -2.11 and -2.0085 values between -2.00 and -1.88112 values between -1.88 and -1.76156 values between -1.76 and -1.65197 values between -1.65 and -1.53203 values between -1.53 and -1.42255 values between -1.42 and -1.30274 values between -1.30 and -1.19268 values between -1.19 and -1.07270 values between -1.07 and -0.95303 values between -0.95 and -0.84262 values between -0.84 and -0.72221 values between -0.72 and -0.61182 values between -0.61 and -0.49164 values between -0.49 and -0.37136 values between -0.37 and -0.26104 values between -0.26 and -0.1493 values between -0.14 and -0.03123 values between -0.03 and 0.0958 values between 0.09 and 0.2147 values between 0.21 and 0.3253 values between 0.32 and 0.4443 values between 0.44 and 0.5532 values between 0.55 and 0.6717 values between 0.67 and 0.7921 values between 0.79 and 0.9010 values between 0.90 and 1.026 values between 1.02 and 1.138 values between 1.13 and 1.255 values between 1.25 and 1.372 values between 1.37 and 1.482 values between 1.48 and 1.60offnegativepositiveoften

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. utely
  2. лист
  3. 大卫
  4. �件
  5. atted
  6. пат
  7. cars
  8. atra
  9. attle
  10. departamento

Suppresses

  1. …⏎
  2. ^\
  3. …⏎⏎
  4. eros
  5. Slo
  6. OOM
  7. ypo
  8. 错
  9. …"⏎⏎
  10. GO