Neuron Explorer

Qwen3-0.6B

zeros after a decimal point in small numbers

A neuron in layer 11 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

Fires on a zero that sits inside a small decimal number, like the middle zeros in "0.001" or "p = 0.01". Turning it up made the model repeat phrases, which is not related to numbers.

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

    orientation of a nanometer-sized magnetic grain in just a few picoseconds (1 picosecond is 0.000000000001 seconds). As a comparison, today’s hard drives take a few nan

  2. #2 peak

    eters to millimeters (meter to mm converter) returns numbers of mm giving meters m value. Converting 1 meter to millimeter as 0.001. See also millimeters into meters calculator. See also millimeters into meters

  3. #3 peak

    – this is one thousandth of a meter. In scientific notation, 1 millimeter is 1 – E3 m, which means 0.001 x 1

  4. #4 peak

    10101100.00010000.00000000.00000000 SM: 11111111.11

  5. #5 peak

    higher after 11 years on the 3MS than did those in the lowest quintile of the DASH score (p-values < 0.001).⏎They also found that four of the nine food-group/nutrient components

  6. #6 peak

    difference –0.51 pg/mL, 95% confidence interval, –0.88 to –0.14; P = .008) and when symptoms had resolved (mean difference –0.71 pg

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 problem of the relationship between the word and the matter, and the matter of the words. But the matter is now in the first step, the matter, and the word is. The problem is to be the matter of the words,

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

    0 00:10:22 CEST 2010⏎On 9/9/10 4:39 PM, Baba wrote:⏎> In below code "

  2. #2 peak

    52.5162 / 52°30'58"N⏎Longitude: -3.0604 / 3°3'37"W⏎OS Eastings: 328136

  3. #3 peak

    -photorealistic-house-from-scratch-to-end-in-AutoCAD⏎- Publication date:⏎- September 2019⏎- 2 hours 45 minutes<|endoftext|>Titan's lakes are composed of a wide variety of hydrocarbons

  4. #4 peak

    could be twisted together to form anchor cable.⏎Topics. This historical marker is listed in this topic list: Industry & Commerce.⏎Location. 38° 54.781′ N, 77° 3.6

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. L7:2353 weakly excites
  2. L3:732 weakly suppresses
  3. L10:548 weakly excites zeros inside large numbers
  4. L10:1706 weakly suppresses
  5. L10:3025 weakly excites
  6. L10:1269 weakly suppresses
  7. L5:50 weakly excites
  8. L3:641 weakly suppresses
  9. L7:1504 weakly suppresses
  10. L10:2547 weakly excites
  11. L6:878 weakly suppresses
  12. L5:2220 weakly excites
  13. L9:2017 weakly suppresses
  14. L3:1245 weakly excites
  15. L6:511 weakly suppresses
  16. L3:2453 weakly suppresses
  17. L0:65 weakly suppresses
  18. L10:2983 weakly excites
  19. L10:2337 weakly excites
  20. L8:2484 weakly excites
  21. L6:13 weakly excites
  22. L9:2222 weakly excites
  23. L4:1818 weakly suppresses
  24. L7:2072 weakly suppresses

Feeds later layers

  1. L20:2659 weakly excites
  2. L16:280 weakly excites
  3. L14:2882 weakly excites
  4. L16:2789 weakly excites
  5. L13:2797 weakly excites
  6. L24:2271 weakly suppresses
  7. L16:2382 weakly excites
  8. L21:2268 weakly excites
  9. L15:1884 weakly suppresses
  10. L20:1048 weakly excites word before a quantity or number
  11. L27:781 weakly excites
  12. L17:3005 weakly excites
  13. L16:1450 weakly suppresses
  14. L25:581 weakly suppresses
  15. L24:519 weakly suppresses
  16. L25:1756 weakly suppresses
  17. L13:1921 weakly suppresses
  18. L17:1958 weakly suppresses
  19. L20:993 weakly suppresses
  20. L17:2579 weakly suppresses
  21. L19:1973 weakly excites
  22. L26:1813 weakly suppresses
  23. L17:1464 weakly excites
  24. L23:241 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 -0.67 and -0.641 values between -0.60 and -0.573 values between -0.57 and -0.547 values between -0.51 and -0.476 values between -0.47 and -0.4420 values between -0.44 and -0.4122 values between -0.41 and -0.3844 values between -0.38 and -0.3451 values between -0.34 and -0.3176 values between -0.31 and -0.28106 values between -0.28 and -0.25126 values between -0.25 and -0.21149 values between -0.21 and -0.18201 values between -0.18 and -0.15233 values between -0.15 and -0.12312 values between -0.12 and -0.08354 values between -0.08 and -0.05332 values between -0.05 and -0.02371 values between -0.02 and 0.02287 values between 0.02 and 0.05294 values between 0.05 and 0.08221 values between 0.08 and 0.11174 values between 0.11 and 0.15147 values between 0.15 and 0.1888 values between 0.18 and 0.2192 values between 0.21 and 0.2471 values between 0.24 and 0.2831 values between 0.28 and 0.3124 values between 0.31 and 0.3424 values between 0.34 and 0.3712 values between 0.37 and 0.4113 values between 0.41 and 0.448 values between 0.44 and 0.472 values between 0.47 and 0.502 values between 0.54 and 0.571 values between 0.57 and 0.602 values between 0.60 and 0.63offnegativepositiveoften

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. anded
  2. vincial
  3. masked
  4. [sizeof
  5. ]][
  6. ickerView
  7. WithValue
  8. 仅供
  9. 深层
  10. 扼

Suppresses

  1. (mut
  2. ­tion
  3. gressive
  4. ionate
  5. reaction
  6. redient
  7. ragon
  8. aining
  9. zb
  10. Forum