Neuron Explorer

Qwen3-0.6B

unclear: mid-phrase words, no common theme

A neuron in layer 25 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 a grab bag of words in the middle of phrases, such as 'than' in a comparison or 'into' after 'divided'. The examples do not share a clear topic, and turning it up or down gave no consistent change in the writing.

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

    it very clear how much faster the heat pulse solution will be.⏎The heat pulse technique is still at the prototype stage, but has been demonstrated to work using magneto-optical microscopy on thin films. Although it is expected to scale to ter

  2. #2 peak

    pain are as follows-⏎- Central and peripheral nervous systems: Nerve damage of serious nature may occur due to anomalies in visceral functions and perceptions (visceral hyperalgesia

  3. #3 peak

    identification⏎More: Simple Safe Practices Promoted by "Safe Hospital"⏎11 - Measures to prevent catheter and tubing misconnections⏎12 - Evaluation of risk of development of pressure ulcers⏎1

  4. #4 peak

    for bacteria which in turn produce methane gas.⏎Methane gas is a greenhouse gas that traps heat much better than carbon dioxide does. Methane gas (CH4) traps about 34

  5. #5 peak

    of the main concerns about PET bottles is the potential for chemicals to leach from the plastic into the contents of the bottle. PET is a type of polymer, which means it is made up of long chains of molecules called monomers

  6. #6 peak

    current or under power relays, directional or reverse current relays differential relays, and distance relays.⏎According to timing characteristics the relays can be divided into instantaneous, definite time lag, inverse time lag and inverse definite minimum time (ID

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: I think that the integration of an individual into the system of the micro-loan system may be an indi- pride system into the system. The micro-loaners who integrate into the system could be thought of as the micro-loan

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

    ces, New Mexico.⏎The ORNL DAAC announces the release of two new data sets and two revised data sets from Phase 2 of the Vegetation/Ecosystem Modeling and Analysis Project (VEMAP).⏎The ORNL DAAC has

  2. #2 peak

    produce on desalination by 2050.⏎The dominant desalination technology at present, with around 60% of global capacity, is Reverse Osmosis (

  3. #3 peak

    connected with copper refrigeration lines, which basically, transfer heat to the outside, via freon.⏎The efficiency of an air conditioner is rated in SEER and EER. The higher the SEER rating and EER rating, the higher efficient

  4. #4 peak

    ; Basílio da Gama; eighteenth century jesuits.<|endoftext|>Texas Kidneywood Flower⏎image by Jim Gober⏎Texas kidneywood (Eysenhardtia texana) grows up to 10 feet tall with

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. L18:1287 weakly suppresses
  2. L8:2141 weakly excites
  3. L10:126 weakly suppresses
  4. L8:1417 weakly excites
  5. L15:1404 weakly excites
  6. L4:840 weakly suppresses
  7. L17:1774 weakly excites
  8. L18:336 weakly excites
  9. L22:509 weakly excites
  10. L22:2082 weakly excites
  11. L12:2158 weakly excites
  12. L5:2263 weakly excites
  13. L9:1452 weakly excites
  14. L20:461 weakly suppresses
  15. L4:3065 weakly excites
  16. L6:2533 weakly excites
  17. L10:378 weakly suppresses
  18. L11:833 weakly suppresses
  19. L21:2364 weakly suppresses
  20. L1:1389 weakly suppresses
  21. L18:1170 weakly suppresses
  22. L8:1250 weakly excites
  23. L20:414 weakly excites
  24. L13:1469 weakly suppresses

Feeds later layers

  1. L26:1537 weakly excites word fragments like ev and v
  2. L27:285 weakly excites
  3. L27:2127 weakly suppresses
  4. L26:2989 weakly excites
  5. L26:759 weakly excites
  6. L27:2 weakly excites
  7. L26:2801 weakly excites
  8. L26:2305 weakly excites
  9. L27:1772 weakly excites
  10. L26:1881 weakly excites
  11. L27:1236 weakly excites
  12. L27:2276 weakly excites
  13. L27:3007 weakly excites
  14. L26:545 weakly excites
  15. L27:1004 weakly excites
  16. L27:1439 weakly excites
  17. L26:2136 weakly suppresses
  18. L27:1365 weakly suppresses
  19. L26:1969 weakly suppresses
  20. L26:726 weakly excites
  21. L27:125 weakly excites
  22. L26:1529 weakly suppresses
  23. L27:2464 weakly excites
  24. L27:557 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 -16.67 and -15.791 values between -14.91 and -14.021 values between -14.02 and -13.142 values between -13.14 and -12.262 values between -12.26 and -11.378 values between -11.37 and -10.493 values between -10.49 and -9.616 values between -9.61 and -8.728 values between -8.72 and -7.849 values between -7.84 and -6.9618 values between -6.96 and -6.0721 values between -6.07 and -5.1935 values between -5.19 and -4.3148 values between -4.31 and -3.4262 values between -3.42 and -2.54104 values between -2.54 and -1.66206 values between -1.66 and -0.771686 values between -0.77 and 0.11913 values between 0.11 and 0.99335 values between 0.99 and 1.88161 values between 1.88 and 2.7697 values between 2.76 and 3.6463 values between 3.64 and 4.5332 values between 4.53 and 5.4131 values between 5.41 and 6.2911 values between 6.29 and 7.1811 values between 7.18 and 8.065 values between 8.06 and 8.9411 values between 8.94 and 9.826 values between 9.82 and 10.716 values between 10.71 and 11.591 values between 12.47 and 13.361 values between 13.36 and 14.241 values between 16.89 and 17.772 values between 17.77 and 18.66offnegativepositiveoften

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. reign
  2. aclass
  3. (start
  4. rzy
  5. (char
  6. phas
  7. (Class
  8. ecl
  9. IGN
  10. starting

Suppresses

  1. izzazione
  2. egrate
  3. 埇
  4. >}</
  5. popcorn
  6. Panc
  7. ."</
  8. lesen
  9. 氿
  10. 俚