Neuron Explorer

Qwen3-0.6B

weak end of document signal

A neuron in layer 12 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 gives a small response at the marker where one document ends and another begins, and a small negative response on single digits inside years. It is weak, so it is hard to say much more. Turning it up gave repetitive or off-topic 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

    manage operations in your warehouse.⏎- Lay out a proper strategy for handling damaged products and returns from customers.⏎- Implement pest control measures to protect your products.⏎-<|endoftext|>Nucleation properties of respirable coal dust (terminal progress report).⏎Physics Department

  2. #2 peak

    will discuss a basic MySQL installation on a Linux virtual server, and some of the basic MySQL commands.⏎Full documentation can be found on the MySQL.com website here.<|endoftext|>The law applicable to military strategic use of outer space⏎New technologies and the law

  3. #3 peak

    the other hand, are designed to handle complex logic functions that require high performance and low latency.⏎Developing software for microcontrollers is relatively easy compared to FPGA development<|endoftext|>The deepest human need is to engage with the world: confidently, productively,

  4. #4 peak

    transfer of title by purchase and sale.⏎- It involves mutual benefit both to the buyer and the seller.⏎- It must be directed for earning one’s livelihood.<|endoftext|>Exactly 100 years ago, on 6 January 191

  5. #5 peak

    00 training examples, you can choose to pick 100 random samples (batch size=100) for each EPOCH (Training cycle)<|endoftext|>|Origin:||co

  6. #6 peak

    be the most effective approach. Smart work, on the other hand, can save time and increase productivity, but may lead to a lack of focus on the details<|endoftext|>Rajan Vempati of ChK Group, Inc. in Plano,

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: "You can't just ask for a specific question to be answered with the information you need to answer the question properly and provide the user with a clear and well-structured response." "Please follow the instructions for the entire process step by step

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

    phoid. Reports of the day state he was frequently asked to describe his techniques before the Ohio Valley Medical Association<|endoftext|>Posted on 09 Aug 2023⏎The arrest of two macaque smugglers in June 202

  2. #2 peak

    for acyclic graphs of pointers.<|endoftext|>The Death And Life Of The Great Lakes Pdf⏎- and pdf⏎- Tuesday, June 1, 2021 2:11:00 AM⏎- 1 comment⏎

  3. #3 peak

    5 - November 1, 2020)⏎This introductory course covers the basic history, core principles and challenges of current

  4. #4 peak

    a potential lunar lava tube entrance – could offer insightful data concerning the moon’s geological history and could hint<|endoftext|>(Updated on June 22, 2020)⏎The mass air flow (MAF) sensor is an important component which

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. L10:58 strongly excites
  2. L9:475 strongly excites
  3. L10:553 suppresses list-style essay phrasing, weak
  4. L10:80 excites
  5. L8:642 suppresses
  6. L9:23 suppresses
  7. L9:302 excites
  8. L3:17 suppresses
  9. L10:1019 suppresses unclear: clause starters in old texts
  10. L7:1005 excites
  11. L8:609 suppresses
  12. L6:2487 suppresses
  13. L2:604 suppresses the word 'some' before a noun
  14. L10:1523 suppresses
  15. L9:1781 suppresses
  16. L4:1760 excites
  17. L9:1826 excites
  18. L4:121 excites
  19. L8:663 suppresses
  20. L8:2902 excites
  21. L3:157 excites
  22. L9:1544 excites
  23. L9:614 weakly suppresses
  24. L8:1462 weakly excites

Feeds later layers

  1. L13:2100 suppresses
  2. L17:349 suppresses
  3. L16:704 weakly suppresses
  4. L14:71 weakly suppresses
  5. L16:1957 weakly suppresses
  6. L26:962 weakly excites
  7. L27:454 weakly excites
  8. L25:257 weakly suppresses
  9. L22:3021 weakly excites
  10. L23:1613 weakly excites
  11. L25:2240 weakly excites
  12. L15:1612 weakly excites
  13. L17:1587 weakly excites
  14. L27:1206 weakly excites
  15. L19:303 weakly suppresses
  16. L15:1111 weakly suppresses
  17. L13:626 weakly suppresses
  18. L17:551 weakly suppresses
  19. L14:1121 weakly suppresses
  20. L17:3007 weakly excites
  21. L27:357 weakly suppresses
  22. L24:587 weakly excites
  23. L25:630 weakly excites
  24. L18:337 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 -0.65 and -0.612 values between -0.61 and -0.571 values between -0.53 and -0.482 values between -0.48 and -0.446 values between -0.44 and -0.4022 values between -0.40 and -0.3631 values between -0.36 and -0.3136 values between -0.31 and -0.2798 values between -0.27 and -0.23138 values between -0.23 and -0.19236 values between -0.19 and -0.14331 values between -0.14 and -0.10464 values between -0.10 and -0.06498 values between -0.06 and -0.02578 values between -0.02 and 0.03441 values between 0.03 and 0.07396 values between 0.07 and 0.11210 values between 0.11 and 0.15171 values between 0.15 and 0.2099 values between 0.20 and 0.2470 values between 0.24 and 0.2836 values between 0.28 and 0.3227 values between 0.32 and 0.376 values between 0.37 and 0.414 values between 0.41 and 0.452 values between 0.45 and 0.491 values between 0.62 and 0.661 values between 1.00 and 1.05offnegativepositiveoften

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. ัก
  2. loth
  3. ��
  4. PLIC
  5. 斋
  6. uide
  7. [${
  8. 化合
  9. DAYS
  10. ль

Suppresses

  1. your
  2. 您的
  3. asn
  4. 你的
  5. ifton
  6. anford
  7. conditional
  8. */⏎⏎⏎⏎
  9. >⏎⏎⏎⏎
  10. innacle