Neuron Explorer

Qwen3-0.6B

drug and chemical name fragments

A neuron in layer 24 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

It fires inside the spelling of medical, chemical and drug names, such as the middle of 'hypoxia', 'zidovudine' or 'imidacloprid'. Turning it down made the model get stuck repeating 'contam' or 'leach'.

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

    /or under the skin⏎- Inflammation of the pancreas (pancreatitis)⏎- Shock and low levels of oxygen in the blood and tissues (hypoxia)⏎- Thrombocytopenia – Low platelet or throm

  2. #2 peak

    -antiretroviral-drug regimen beginning as early as 14 weeks of gestation, or giving them the antiretroviral drug zidovudine beginning as early as 14<|endoftext|>Date of this Version⏎

  3. #3 peak

    the amount of salt and high-sodium condiments (e.g. soy and fish sauces). Avoid snacks that are high in salt and sugars; limit intake of sodas and other drinks high in sugars (e.g. fruit juices, flavored drinks

  4. #4 peak

    . Integrated pest management specialists with the University of California, Davis recommend avoiding the use of pesticides containing diazinon, carbaryl, pyrethroids and imidacloprid when caring for the gardenia.⏎A heavy sooty

  5. #5 peak

    ers triclosan resistance to the host.⏎References[change | change source]⏎- Callmigration

  6. #6 peak

    e. In addition, it tones and strengthens all the muscles that support the back, while maintaining a good alignment of the body. Excellent for preventing back pain of sedentary or some athletes. Regular practice helps to relieve tension in the upper back

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: Contexuality and Contexualcontexualcontexualcontexualcontexualcontexualcontexualcontexualcontexualcontexualcontexualcontexualcontexualcontexualcont

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

    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

  2. #2 peak

    cramps) and norovirus (often called "stomach flu" which also causes nausea, vomiting, stomach cramps, fever, and fatigue). Cross-contamination

  3. #3 peak

    For example, if you are baking six breast pieces, each one might cook differently based on size. Don’t forget to clean your thermometer probe between readings to avoid cross-contamination.⏎Food or meat thermometers are fairly inexpensive, and will save you

  4. #4 peak

    shots protects against Hepatitis B, a virus transmitted through bodily fluids. Long-term infection can destroy the liver. People at risk include those whose occupations expose them to blood, such as health-care workers. The vaccine is now commonly given to all newborn

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. L21:1435 suppresses
  2. L10:2202 excites
  3. L21:3020 weakly excites
  4. L6:1853 weakly suppresses
  5. L4:1852 weakly suppresses
  6. L22:845 weakly suppresses
  7. L6:2420 weakly suppresses
  8. L21:2351 weakly suppresses
  9. L8:2117 weakly excites
  10. L15:2045 weakly excites
  11. L15:2782 weakly suppresses
  12. L4:2035 weakly excites
  13. L21:1516 weakly suppresses
  14. L7:1811 weakly excites
  15. L22:2613 weakly excites
  16. L14:2306 weakly excites word after a motion or location verb
  17. L23:633 weakly suppresses
  18. L10:2333 weakly suppresses
  19. L6:2517 weakly suppresses
  20. L18:311 weakly suppresses
  21. L9:2117 weakly excites
  22. L18:1276 weakly excites
  23. L18:1324 weakly suppresses
  24. L3:465 weakly suppresses

Feeds later layers

  1. L26:2865 weakly suppresses
  2. L25:510 weakly suppresses
  3. L25:2216 weakly suppresses
  4. L25:1069 weakly suppresses
  5. L26:1549 weakly suppresses
  6. L27:1754 weakly excites
  7. L26:580 weakly suppresses
  8. L27:221 weakly suppresses
  9. L25:618 weakly excites
  10. L25:2298 weakly excites
  11. L26:739 weakly suppresses
  12. L25:1906 weakly suppresses
  13. L26:2866 weakly excites
  14. L25:1162 weakly suppresses
  15. L27:45 weakly excites
  16. L27:2324 weakly excites
  17. L27:391 weakly suppresses
  18. L25:2108 weakly excites
  19. L25:753 weakly suppresses
  20. L25:1297 weakly suppresses
  21. L25:1040 weakly excites digit 1 inside dates and times
  22. L27:2000 weakly suppresses
  23. L27:1694 weakly excites
  24. L25:1317 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 -26.02 and -24.561 values between -23.10 and -21.641 values between -18.72 and -17.262 values between -17.26 and -15.801 values between -15.80 and -14.344 values between -14.34 and -12.894 values between -11.43 and -9.978 values between -9.97 and -8.5110 values between -8.51 and -7.0510 values between -7.05 and -5.5926 values between -5.59 and -4.1327 values between -4.13 and -2.6789 values between -2.67 and -1.212794 values between -1.21 and 0.25806 values between 0.25 and 1.7143 values between 1.71 and 3.1629 values between 3.16 and 4.6215 values between 4.62 and 6.0813 values between 6.08 and 7.547 values between 7.54 and 9.003 values between 9.00 and 10.465 values between 10.46 and 11.921 values between 11.92 and 13.381 values between 14.84 and 16.304 values between 17.75 and 19.211 values between 22.13 and 23.591 values between 25.05 and 26.511 values between 30.89 and 32.34offnegativepositiveoften

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. ButtonItem
  2. replicate
  3. capital
  4. 樘
  5. 开户
  6. recre
  7. _INITIALIZER
  8. 琢
  9. CallCheck
  10. erect

Suppresses

  1. cont
  2. cont
  3. →cont
  4. Cont
  5. Cont
  6. le
  7. kont
  8. -cont
  9. 污染
  10. le