Neuron Explorer

Qwen3-0.6B

the article 'a' before a noun

A neuron in layer 21 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 reacts to the word 'a' (or 'an') in the middle of a sentence, right before the noun it describes. Turning it down made the model get stuck repeating the word 'a' over and over, which fits its focus on that small word.

Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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

    some extent and organize state or national lotteries. Lottery prizes are typically paid in cash, although some states also award goods or services. In many cases, a winner must pay taxes on the prize amount before receiving it.⏎Lottery tickets can

  2. #2 peak

    obtain an “A grade” or perform well in the office. But if their standards aren’t met, they get upset. Perfectionists don’t care about a company’s goal. It doesn’t matter to them. So they’ll always put

  3. #3 peak

    a single winner or a small group of winners, while others offer a large number of smaller prizes. The prizes can be cash or goods. In some cases, a percentage of the total pool is reserved for costs related to organizing and promoting the lottery

  4. #4 peak

    ucky' and 'Red Globe'.⏎Photographed in author's garden.<|endoftext|>- Health Library⏎- PetMD U⏎|8 Common Snacks That Will Prompt a Portly Pet||Overweight Pets: Addressing the Epidemic||How to

  5. #5 peak

    the outcome, I don't think we would have had such a comprehensive service from any other law firm, and you took the worry away...I do not regret a single second of the whole process, apart from the bit before you got involved.⏎

  6. #6 peak

    created to an industrial model. They were built to ensure that kids could competently deal with the routine tasks that were necessary in factories. Some kids would succeed in a different way; but in the main, we were educating kids to be employed in

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: A a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a

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

    considerations to be taken into account when writing a technical report are highlighted. This section will briefly address the following points:⏎What is a technical report?⏎In simple terms, a report is a document that presents or communicates to a target audience the findings (

  2. #2 peak

    of the trust. The Houston living trust lawyers at Mendel Law Firm, L.P. explain when and how a living trust terminates.⏎What Is a Living Trust?⏎The person who creates a trust goes by many names. They can be referred to

  3. #3 peak

    a central eye. Depending on their location and strength, the storms are called different things. What gets dubbed a hurricane in the Atlantic, for example, would be called

  4. #4 peak

    In this article, we will delve into the main differences between microcontrollers and FPGAs, as well as their advantages and disadvantages.⏎What is a Microcontroller?⏎A microcontroller is a small computer on a single integrated circuit that is designed to

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. L19:684 strongly suppresses
  2. L19:654 excites
  3. L19:2808 suppresses
  4. L18:873 excites
  5. L19:2269 suppresses
  6. L19:626 excites
  7. L19:1030 suppresses
  8. L20:2609 suppresses
  9. L18:2590 suppresses
  10. L20:2092 suppresses
  11. L7:2593 excites
  12. L20:1857 suppresses
  13. L19:1743 suppresses
  14. L19:1695 suppresses
  15. L19:2522 suppresses
  16. L19:859 excites
  17. L19:959 suppresses
  18. L18:2451 suppresses
  19. L20:1863 excites 'a' before an object noun
  20. L18:140 suppresses
  21. L19:1817 excites
  22. L20:1972 suppresses 'of' in 'of course'
  23. L19:639 suppresses
  24. L17:1278 suppresses

Feeds later layers

  1. L25:1586 strongly suppresses
  2. L27:797 strongly suppresses
  3. L25:45 strongly excites
  4. L24:2538 strongly suppresses
  5. L27:663 strongly suppresses
  6. L22:103 strongly excites
  7. L26:484 strongly suppresses
  8. L24:823 suppresses
  9. L27:231 excites
  10. L27:709 suppresses
  11. L25:1132 excites
  12. L27:173 excites
  13. L22:2889 suppresses
  14. L22:2302 suppresses
  15. L26:213 excites
  16. L22:171 suppresses
  17. L27:1044 suppresses
  18. L27:854 suppresses
  19. L26:411 excites
  20. L27:2396 suppresses
  21. L22:303 suppresses
  22. L22:247 suppresses
  23. L25:549 suppresses
  24. L25:568 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 -23.03 and -21.861 values between -14.80 and -13.621 values between -12.45 and -11.271 values between -8.92 and -7.744 values between -6.57 and -5.398 values between -5.39 and -4.2112 values between -4.21 and -3.0426 values between -3.04 and -1.86153 values between -1.86 and -0.683070 values between -0.68 and 0.49488 values between 0.49 and 1.6751 values between 1.67 and 2.8523 values between 2.85 and 4.0212 values between 4.02 and 5.2011 values between 5.20 and 6.377 values between 6.37 and 7.552 values between 7.55 and 8.735 values between 8.73 and 9.908 values between 9.90 and 11.083 values between 11.08 and 12.254 values between 12.25 and 13.438 values between 13.43 and 14.614 values between 14.61 and 15.781 values between 15.78 and 16.962 values between 16.96 and 18.141 values between 18.14 and 19.311 values between 22.84 and 24.02offnegativepositiveoften

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. わけです
  2. qrstuvwxyz
  3. pelo
  4. ép
  5. ро
  6. 蹙
  7. 黍
  8. ulous
  9. 럴
  10. ê

Suppresses

  1. a
  2. an
  3. 一个
  4. →an
  5. →a
  6. -a
  7. 是一个
  8. )a
  9. 一个小
  10. 也是一个