Neuron Explorer

Qwen3-0.6B

word pieces before 'trans' or 'transfer'

A neuron in layer 25 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

This neuron fires on the start of words that are about to become 'trans', 'translation', 'transfer' or 'transfusion', such as 'TRAN' in 'TRANSLATED' or 'transf' in 'transfusion'. When it was turned up, the model kept producing the word 'transfer' over and over.

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

    .1 Laminated core⏎- 1.2 Toroidal⏎- 1.3 Autotransformer⏎- 1.4 Variable autotransformer⏎- 1.5 Induction regulator⏎- 1.

  2. #2 peak

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

  3. #3 peak

    ED TOPIC: HERE⏎E-GO EDITED BY: Mooloo⏎DATE ADDED: Jan 20, 2011⏎TRANSLATED BY: まだ⏎J-GO EDITED BY: まだ

  4. #4 peak

    are severe. Families may feel they do not have an opportunity to bid farewell. In areas where Ebola spread, community members often came to mistrust medical workers due to the lack of contact with dying family members. Here is

  5. #5 peak

    types, and the Rh group is either positive or negative. You blood groups help doctors know many things, like what kind of blood you’ll need in a blood transfusion or an organ donation, and if you or your baby will be in danger

  6. #6 peak

    to be that of legal interpretation with the help of the Final Record of the Conference and the preliminary texts (1).⏎PARAGRAPH 2. -- OFFICIAL TRANSLATIONS (2)⏎This provision too is an innovation so far as the Geneva

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: I have a 666, 666, 666, 666, 666, 666, 666, 666, 666,

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

    copyright?⏎You can transfer the economic rights of a unique and original work. You can do it under a license or a protocol.⏎A license transfer⏎- You can still profit financially from your work.⏎- You transfer the rights for a period of

  2. #2 peak

    robot to the appropriate human,” Woods said.⏎“The bottom line is, robots need to be responsive and resilient. They have to be able to protect themselves and also smoothly transfer control to humans when necessary.”⏎Woods admits that one thing is missing from

  3. #3 peak

    . Transfer hydrogenation catalysis (Figure 1) is a promising strategy for hydrocarbon reduction that occurs rapidly<|endoftext|>Raul Garcia-Patron Sanchez and colleagues find that quantum advantage is still some way off⏎Quantum advantage – the moment when quantum

  4. #4 peak

    the processing delay function of the warehouse has been mentioned. Generally, the warehouse with the product processing capability is become the processing warehouse.⏎(4) Transfer warehouse⏎The transit warehouse is in the middle of the cargo transportation system, and stores the goods

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. L23:2059 suppresses
  2. L23:2516 excites
  3. L21:1820 suppresses
  4. L23:2769 excites
  5. L23:2565 suppresses
  6. L18:3057 excites
  7. L23:1722 suppresses
  8. L24:1216 excites
  9. L22:1652 excites
  10. L24:1554 excites
  11. L22:947 excites
  12. L24:1086 excites
  13. L6:1726 weakly excites
  14. L22:408 weakly suppresses
  15. L20:2559 weakly excites
  16. L8:2018 weakly excites
  17. L24:1063 weakly suppresses
  18. L21:1535 weakly suppresses 'Old' as part of a name or phrase
  19. L20:2929 weakly suppresses
  20. L23:1213 weakly suppresses
  21. L23:693 weakly suppresses
  22. L22:1386 weakly excites
  23. L16:2231 weakly excites
  24. L15:911 weakly excites

Feeds later layers

  1. L27:1348 strongly excites
  2. L27:2771 excites
  3. L26:3003 excites
  4. L26:1977 weakly excites
  5. L26:1164 weakly excites
  6. L26:1958 weakly excites
  7. L27:2057 weakly excites
  8. L26:933 weakly excites
  9. L27:1209 weakly excites
  10. L27:844 weakly excites
  11. L27:857 weakly suppresses
  12. L26:2828 weakly excites
  13. L26:3071 weakly excites
  14. L26:1668 weakly suppresses
  15. L27:3046 weakly excites
  16. L27:2134 weakly suppresses
  17. L26:1992 weakly excites
  18. L26:1396 weakly suppresses
  19. L26:2697 weakly suppresses word-start 'le' and 'log' pieces
  20. L26:2755 weakly suppresses
  21. L26:1315 weakly excites
  22. L27:470 weakly suppresses
  23. L27:2603 weakly excites
  24. L26:1686 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.

2 values between -4.34 and -3.982 values between -3.98 and -3.621 values between -3.62 and -3.253 values between -3.25 and -2.892 values between -2.89 and -2.538 values between -2.53 and -2.1710 values between -2.17 and -1.8010 values between -1.80 and -1.4435 values between -1.44 and -1.08112 values between -1.08 and -0.71478 values between -0.71 and -0.351622 values between -0.35 and 0.011350 values between 0.01 and 0.38206 values between 0.38 and 0.7424 values between 0.74 and 1.1015 values between 1.10 and 1.476 values between 1.47 and 1.834 values between 1.83 and 2.195 values between 2.19 and 2.561 values between 2.56 and 2.924 values between 3.28 and 3.642 values between 3.64 and 4.011 values between 4.01 and 4.371 values between 4.37 and 4.731 values between 6.55 and 6.911 values between 7.64 and 8.001 values between 8.36 and 8.731 values between 9.82 and 10.18offnegativepositiveoften

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. trans
  2. Trans
  3. trans
  4. TRANS
  5. transfer
  6. (trans
  7. Trans
  8. -trans
  9. transfer
  10. →trans

Suppresses

  1. ội
  2. тив
  3. RICS
  4. voke
  5. �력
  6. ibili
  7. 息
  8. pillows
  9. 了一口气
  10. icut