Neuron Explorer

Qwen3-0.6B

'in' after high or low

A neuron in layer 5 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 "in" when it follows "high" or "low", as in "high in fiber" or "low in calories". Turning it up did not produce anything about nutrition, so the steering outputs are unrelated.

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

    aspberries. Raspberries are not only low in calories, sugar and carbohydrates, they are also one of the highest sources of fiber

  2. #2 peak

    of cancer, including lung, stomach, prostate, cervical and breast cancer.⏎2. Supports your Heart Health⏎Because these nuts are high in monounsaturated fats, they are able to reduce the bad cholesterol, and also

  3. #3 peak

    line: Eat more protein, especially plant protein, for your health. [And this means real foods, not supplements!]⏎From Science Daily: Diets high in protein, particularly plant protein, linked to lower risk of death⏎Diets high

  4. #4 peak

    and many have found little help for their bloating, gas, abdominal pain, constipation and diarrhea. Now, more and more research shows that a diet low in some types of FODMAP foods may help bring relief.

  5. #5 peak

    blood pressure [*].⏎Focus on foods high in potassium too. Potassium is an electrolyte needed to balance out too much sodium

  6. #6 peak

    kind of acid called galacturonic acid. This acid is beneficial for elimination of some harmful substances. Thus apples are beneficial in diabetes.⏎An apple is low in calories and contains high fiber and for this reason it fills the stomach without increasing blood

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: but the question: What is the value of the temperature? as a result of the fusion of two different temperatures of matter and physics, but the question of a thermal expansion in physics, but the question of a thermal expansion in physics. -

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

    Centre (PVRC) with both projects being led by Dr Colin Copeland in partnership with industry.⏎The first project, Additive Manufacturing for Cooled High-temperature Automotive Radial Machinery (CHARM), is in collaboration with HiETA Technologies.⏎

  2. #2 peak

    and drinking water to peak at extremely high concentrations.⏎The most recent data confirms that atrazine continues to contaminate waters

  3. #3 peak

    Unfortunately, the “hot spots” worldwide appear to be the same waters that support the most productive fisheries, including upwelling zones, cold high-latitude seas, and<|endoftext|>'Lose my life, my leg or my hand?':

  4. #4 peak

    , we studied the light-induced processes occurring at the photocathode/electrolyte interface by advanced ultrafast spectroscopy." We show that hydroxyl groups formed at the NiO/w

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. L4:787 excites
  2. L4:3059 weakly excites
  3. L3:2637 weakly suppresses
  4. L1:739 weakly excites
  5. L3:1478 weakly suppresses
  6. L1:198 weakly suppresses
  7. L3:2184 weakly suppresses
  8. L2:2983 weakly suppresses
  9. L3:187 weakly excites
  10. L2:1926 weakly suppresses
  11. L2:1514 weakly excites
  12. L4:2476 weakly suppresses
  13. L4:280 weakly suppresses
  14. L1:2918 weakly excites
  15. L4:1402 weakly suppresses
  16. L1:722 weakly excites
  17. L2:1104 weakly excites
  18. L4:1171 weakly suppresses
  19. L3:2505 weakly suppresses
  20. L4:2544 weakly suppresses
  21. L0:2443 weakly suppresses
  22. L2:1219 weakly suppresses
  23. L3:995 weakly suppresses
  24. L4:23 weakly excites

Feeds later layers

  1. L8:1109 weakly suppresses
  2. L12:551 weakly suppresses
  3. L8:233 weakly suppresses
  4. L13:1837 weakly excites
  5. L11:1617 weakly excites
  6. L9:3058 weakly excites
  7. L16:2839 weakly suppresses
  8. L24:1768 weakly excites
  9. L23:1714 weakly excites
  10. L6:453 weakly suppresses
  11. L21:153 weakly suppresses
  12. L8:62 weakly suppresses
  13. L11:1915 weakly excites
  14. L23:1675 weakly suppresses
  15. L19:1785 weakly excites
  16. L8:1882 weakly excites
  17. L6:773 weakly excites
  18. L25:1772 weakly excites
  19. L16:2326 weakly excites
  20. L14:18 weakly suppresses
  21. L8:2314 weakly excites
  22. L8:1458 weakly suppresses
  23. L10:2178 weakly excites 'this' after a verb or preposition
  24. L23:2083 weakly excites 'a' meaning one single

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 -2.04 and -1.821 values between -1.61 and -1.391 values between -1.39 and -1.181 values between -1.18 and -0.964 values between -0.96 and -0.755 values between -0.75 and -0.5330 values between -0.53 and -0.32593 values between -0.32 and -0.102691 values between -0.10 and 0.11537 values between 0.11 and 0.3330 values between 0.33 and 0.557 values between 0.55 and 0.762 values between 0.76 and 0.981 values between 1.19 and 1.412 values between 2.05 and 2.271 values between 4.42 and 4.641 values between 6.36 and 6.57offnegativepositiveoften

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. omnia
  2. 项目的
  3. ói
  4. elman
  5. 并向
  6. пром
  7. juana
  8. poi
  9. 用户的
  10. Pandora

Suppresses

  1. end
  2. enc
  3. ensored
  4. minLength
  5. (Encoding
  6. equip
  7. 抱
  8. 开奖
  9. 骝
  10. ements