Neuron Explorer

Qwen3-0.6B

"to" after a noun saying what to do

A neuron in layer 3 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 lights up on the word "to" when it comes right after a noun and introduces an action, as in "things to remember" or "questions to ask". It gets quieter on words like "microwave" and "waves". Turning it up made the model's writing drift into rambling or repeated phrases, not anything about this pattern.

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

    your gut can affect how well you age,1 and this, of course, ties in directly with the latest research on calorie restriction and longevity.⏎One important thing to remember about the microbes in your gut is that they are not static. They can

  2. #2 peak

    we pay attention to it as well.⏎2. This Report Will Be “Revised”⏎So assuming GDP is important to follow, the next thing to understand about this Q3 report is “a”. “a” is the letter

  3. #3 peak

    out what their plans on negotiations are. You will likewise have to exercise a circulation handle them<|endoftext|>Written by Gennifer Albin⏎One of the most important things to me as a parent is encouraging my children to be creative, because within the creative

  4. #4 peak

    just highlight action or represent the subtext; it can also play a large role in clarifying or even creating structure.⏎To that end, here are some questions to ask during the spotting session (or whenever you think of it!)⏎- What is

  5. #5 peak

    , it can also be modified manually in the top menu.⏎Now, the formula to get the percentages is basically: the letter and the cell number of the general amount

  6. #6 peak

    5 networking protocols and port numbers video calls or teleconferencing are made possible with the use of some widely used networking protocols. One of the many fundamental things to know as a network engineer is the function and port number used by a number of

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 way I was thinking about this, and the way I was thinking about that. So I have a lot of them. Maybe. Or a little. And that's a certain type. So I have a lot of them. I

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

    Those costs are ultimately borne by American consumers and act as a tax on everyday goods made from steel, such as lawn mowers, washing machines, and microwaves.⏎Available tools to combat alleged unfair trade practices under domestic trade rules are permissible and

  2. #2 peak

    RF) waves, a form of electromagnetic radiation between FM radio and microwaves. Unlike x-rays or ultraviolet (UV) light, this form of

  3. #3 peak

    improperly acts as one of the best breeding grounds for bacteria and foodborne viruses. Reheating rice through rice cooker or microwave will only increase the chance of bacteria spreading.⏎If they’re left to cool down

  4. #4 peak

    and kitchen:⏎• Limit the time the stove/oven is used. Never light the oven or turn on a burner when a small appliance will do the job. Microwave ovens, toaster ovens, electric griddles, panini makers and,

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. L2:2510 weakly suppresses
  2. L1:373 weakly excites
  3. L0:1344 weakly excites
  4. L2:2692 weakly excites
  5. L0:2839 weakly suppresses
  6. L2:2829 weakly suppresses
  7. L1:24 weakly excites
  8. L2:2638 weakly excites
  9. L2:2261 weakly excites
  10. L1:2414 weakly suppresses
  11. L2:2659 weakly excites
  12. L2:2664 weakly suppresses
  13. L1:416 weakly suppresses
  14. L0:1907 weakly excites
  15. L2:940 weakly excites
  16. L1:2518 weakly excites
  17. L0:3059 weakly suppresses
  18. L2:235 weakly suppresses
  19. L1:759 weakly excites
  20. L2:2495 weakly excites
  21. L0:1471 weakly excites
  22. L0:468 weakly suppresses
  23. L1:926 weakly suppresses
  24. L0:2765 weakly suppresses

Feeds later layers

  1. L10:562 excites
  2. L8:1616 weakly excites
  3. L6:2297 weakly excites
  4. L8:984 weakly excites
  5. L13:2417 weakly suppresses
  6. L16:1016 weakly suppresses
  7. L17:2498 weakly excites
  8. L6:2061 weakly excites
  9. L15:1927 weakly excites
  10. L4:783 weakly suppresses
  11. L22:999 weakly suppresses
  12. L10:941 weakly excites
  13. L24:1300 weakly excites
  14. L21:587 weakly suppresses
  15. L5:634 weakly suppresses
  16. L23:1640 weakly excites
  17. L10:2693 weakly excites
  18. L19:1255 weakly excites
  19. L12:2001 weakly excites
  20. L18:584 weakly suppresses
  21. L7:182 weakly excites
  22. L4:252 weakly suppresses
  23. L9:2348 weakly excites dictionary and citation boilerplate
  24. L4:1266 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.

6 values between -0.89 and -0.7410 values between -0.74 and -0.6020 values between -0.60 and -0.4632 values between -0.46 and -0.31179 values between -0.31 and -0.17942 values between -0.17 and -0.031901 values between -0.03 and 0.11675 values between 0.11 and 0.2695 values between 0.26 and 0.4013 values between 0.40 and 0.5410 values between 0.54 and 0.688 values between 0.68 and 0.832 values between 0.83 and 0.975 values between 0.97 and 1.112 values between 1.11 and 1.263 values between 1.26 and 1.401 values between 1.40 and 1.541 values between 1.97 and 2.111 values between 3.25 and 3.391 values between 4.54 and 4.681 values between 4.68 and 4.82offnegativepositiveoften

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. TZ
  2. evaluate
  3. 剁
  4. değerlendirme
  5. vant
  6. .sav
  7. PJ
  8. aterangepicker
  9. opinión
  10. ITV

Suppresses

  1. ..⏎
  2. inters
  3. eland
  4. config
  5. cest
  6. ilib
  7. eman
  8. gest
  9. mons
  10. البن