Neuron Explorer

Qwen3-0.6B

opening bracket after an imperial measurement

A neuron in layer 16 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 fires on the opening bracket that appears after a measurement written in miles, feet or pounds, just before the metric version is given, like '250 miles (400 km)'. Turning it up did not produce any clear theme in the model's writing.

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

    the Tunisia Campaign of World War II in February of 1943. It was a series of battles fought around Kasserine Pass, a two-mile (3 km) wide gap in the Grand Dorsal chain of the Atlas Mountains

  2. #2 peak

    to the Upper Danube.⏎Marlborough marched a mixed force of 40,000 British, Dutch and German troops 250 miles (400km) in five weeks. This extraordinary feat of strategy and logistics

  3. #3 peak

    . One thing<|endoftext|>Beira, Mozambique's second largest city, was swept away by a cyclone with a speed of 160 kilometres per hour (km/h), named Eloise. In addition to the human and material damage,

  4. #4 peak

    mph (105 km/h).⏎On September 14, Humberto had reached post-tropical cyclone status, with winds at 40 mph (65 km/h). According to the NHC, that morning the storm maintained

  5. #5 peak

    perianth segments. There can be as many as half a dozen flowers on a stalk. The flowering stalk may stand up to 3 ft (1 m) tall. The sword shaped leaves are mostly basal, in two ranks

  6. #6 peak

    God) and the charming Temple of Baalshamin (god of fertility) — and a monumental arch that opens onto a 1,200-meter (3,937 ft.) long colonnade, the center of this

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: what if I have the following to do to have a "real life for a real life for a " real life for a real: let's do this. Here's what you have to do for a real: Here's what you have

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

    holder (kinara) with seven candles.⏎Hanukkah is the Jewish Festival of Lights. It is observed for eight nights with the kindling of one additional light on each night of the holiday. Families gather on these nights to play dre

  2. #2 peak

    If you miss a dose of Ampicillin and are taking it regularly, take it as soon as possible. If it is almost time for your next dose, skip the missed dose and go back to your regular dosing schedule. Do not take

  3. #3 peak

    in extreme diets, fasting, or exercise; or binge-eating to purging, in which the person eats a large amount of food within a certain time frame then purges it through the use of vomiting or

  4. #4 peak

    and 3 to 6 percent for bulimia. (Binge eating disorder is a newer diagnosis; I've heard 3 percent for BED but don't know if it's accurate.) Which somehow

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. L11:821 strongly excites
  2. L14:896 strongly suppresses
  3. L14:1764 excites
  4. L10:910 suppresses
  5. L14:2813 excites
  6. L14:948 suppresses
  7. L15:3007 suppresses
  8. L11:734 excites
  9. L10:1727 suppresses
  10. L13:1321 excites
  11. L13:1675 suppresses
  12. L12:1873 suppresses
  13. L13:1118 excites
  14. L14:272 excites unclear: quantity and choice words
  15. L12:1590 excites
  16. L13:1731 suppresses
  17. L13:2026 suppresses
  18. L10:972 excites
  19. L12:1849 excites
  20. L13:1905 suppresses
  21. L6:542 excites
  22. L8:1296 suppresses
  23. L12:1480 excites
  24. L15:1740 suppresses

Feeds later layers

  1. L22:739 suppresses 'are' after 'here' introducing a list
  2. L17:2031 suppresses
  3. L19:2149 suppresses
  4. L19:2250 suppresses
  5. L19:2122 suppresses
  6. L18:548 suppresses
  7. L17:1462 suppresses
  8. L17:1510 suppresses
  9. L17:243 excites
  10. L17:2198 suppresses
  11. L20:2676 suppresses
  12. L17:2851 suppresses
  13. L18:2147 suppresses
  14. L22:1525 suppresses
  15. L18:2359 suppresses
  16. L18:2139 suppresses
  17. L18:2837 suppresses
  18. L17:2850 suppresses
  19. L20:686 suppresses
  20. L22:426 suppresses
  21. L18:1076 suppresses
  22. L18:1806 suppresses
  23. L21:1580 suppresses
  24. L17:2648 suppresses number word before a list of reasons

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 -3.97 and -3.811 values between -3.04 and -2.883 values between -2.73 and -2.575 values between -2.42 and -2.264 values between -2.26 and -2.116 values between -2.11 and -1.953 values between -1.95 and -1.809 values between -1.80 and -1.6410 values between -1.64 and -1.4918 values between -1.49 and -1.3321 values between -1.33 and -1.1841 values between -1.18 and -1.0255 values between -1.02 and -0.8760 values between -0.87 and -0.71123 values between -0.71 and -0.56180 values between -0.56 and -0.40259 values between -0.40 and -0.25490 values between -0.25 and -0.091184 values between -0.09 and 0.07573 values between 0.07 and 0.22329 values between 0.22 and 0.38209 values between 0.38 and 0.53131 values between 0.53 and 0.6977 values between 0.69 and 0.8451 values between 0.84 and 1.0025 values between 1.00 and 1.1518 values between 1.15 and 1.309 values between 1.30 and 1.466 values between 1.46 and 1.613 values between 1.61 and 1.771 values between 1.77 and 1.921 values between 1.92 and 2.082 values between 2.08 and 2.23offnegativepositiveoften

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. 霉素
  2. 锁定
  3. nant
  4. 这个词
  5. çekil
  6. 增长率
  7. %B
  8. 上が
  9. Resizable
  10. 時に

Suppresses

  1. 如下
  2. features
  3. Features
  4. features
  5. _attributes
  6. _features
  7. 主要包括
  8. 包括
  9. —including
  10. Attributes