Neuron Explorer

Qwen3-0.6B

percent-encoded url characters

A neuron in layer 1 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 reacts to the leftover characters of web addresses and encoded text, such as the '0' in '%20' or the 'F' in '%2F'. Turning it up made the model write about unrelated things, so the steering gives no clear picture.

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

    017, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5707683/#:~:text=It%20is%20a%20potent%20antioxidant

  2. #2 peak

    and%20adaptive%20immune%20system.⏎- How Vitamin C Supports a

  3. #3 peak

    7075. . . http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0057075.<|endoftext|>Racism

  4. #4 peak

    archive.jpr.org.uk/10.1177%2F1359104505053754

  5. #5 peak

    - Understanding a patient's values and behaviors is seen as one way to reduce health disparities⏎- Cultural competency acknowledges that provider and patient may have different goals%2C experiences⏎- Race and Health Disparities workgroup of FR%3DE

  6. #6 peak

    ) is a proto-oncogene that encodes a protein known as hepatocyte growth factor receptor (HGFR). The hepatocyte growth factor receptor protein possesses tyrosine-kinase activity

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 had a lot of time to think about it" and that it was something I had to think about it. So when I think about it, I had to think about it. But it was something I had to think about it

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

    from a blog posting entitled Disk Partition Alignment Best Practices for SQL Server by Jimmy May and Denny Lee at Microsoft.⏎You have a server with a RAID volume – For the example we will say RAID 1

  2. #2 peak

    6 bytes, exactly 1024 KB or 1 MB. This is a much better solution because it scales well across the most common stripe sizes ââ‚<|endoftext|>Many developing countries rely on traditional medicine as an accessible and affordable treatment

  3. #3 peak

    (core) city ; Greek ἄστυ asty ) with the associated surrounding area ( χώρ

  4. #4 peak

    ν ἄνθρωπον ἐγένετο καὶ οὐχ ὁ ἄνθρωπος διὰ τὸ σάββ

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. L0:1301 suppresses
  2. L0:1224 suppresses
  3. L0:178 suppresses
  4. L0:1 weakly suppresses
  5. L0:1849 weakly suppresses
  6. L0:2809 weakly excites
  7. L0:961 weakly suppresses
  8. L0:2082 weakly suppresses
  9. L0:19 weakly excites
  10. L0:29 weakly suppresses
  11. L0:1229 weakly suppresses
  12. L0:1591 weakly suppresses
  13. L0:2628 weakly suppresses
  14. L0:2907 weakly excites
  15. L0:652 weakly suppresses
  16. L0:365 weakly excites
  17. L0:1021 weakly excites
  18. L0:2433 weakly suppresses
  19. L0:202 weakly excites start of a new document
  20. L0:2299 weakly suppresses
  21. L0:237 weakly suppresses
  22. L0:61 weakly excites
  23. L0:920 weakly suppresses digit '9' inside years and numbers
  24. L0:2422 weakly excites

Feeds later layers

  1. L9:2150 excites
  2. L22:550 weakly excites
  3. L12:2858 weakly excites comma or bracket after a name in a byline
  4. L15:1415 weakly excites
  5. L9:13 weakly excites
  6. L8:2221 weakly excites
  7. L9:2572 weakly excites
  8. L20:2851 weakly excites
  9. L23:1050 weakly excites
  10. L2:2688 weakly suppresses
  11. L22:620 weakly suppresses the word 'if'
  12. L18:976 weakly suppresses
  13. L19:1893 weakly suppresses
  14. L23:575 weakly excites
  15. L4:1064 weakly suppresses
  16. L22:488 weakly excites
  17. L9:2616 weakly excites
  18. L13:2005 weakly excites
  19. L9:2258 weakly suppresses
  20. L16:1944 weakly suppresses
  21. L12:2810 weakly excites
  22. L22:1577 weakly excites
  23. L25:349 weakly excites
  24. L11:2721 weakly 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.

15 values between -0.48 and -0.3341 values between -0.33 and -0.19624 values between -0.19 and -0.043163 values between -0.04 and 0.1157 values between 0.11 and 0.265 values between 0.26 and 0.412 values between 0.41 and 0.561 values between 5.31 and 5.46offnegativepositiveoften

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. ansom
  2. ÃO
  3. <const
  4. Resolved
  5. ��
  6. Requires
  7. balls
  8. ancia
  9. ônica
  10. oner

Suppresses

  1. 事
  2. ạt
  3. center
  4. hid
  5. 事迹
  6. 事后
  7. Mouth
  8. /Internal
  9. mouth
  10. -cache