blank after a dash in number ranges
A neuronOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in layerOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. 0 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featuredA neuron picked for the full treatment: steering outputs and a written explanation. About 3,000 of the 86,000 neurons are featured., so it has steering outputs.
What this neuron does
This neuron reacts to the spot right after a dash or slash in a range or rating, such as '45 minutes - 1 hour' or '5 / 5'. It is most active where a second number is about to follow. Turning it up did not produce a clear, related change in the text.
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 activationThe number a neuron produces for a token. Large positive means it fired hard; negative means it pushed the other way. was highest. Amber marks positive activation; the outlined tokenA piece of text the model reads at once: often a word, sometimes part of a word or a punctuation mark. is the peak. Hover a token for its strength.
-
⏎45 minutes - 1 hour⏎Suggested Number of Participants:⏎Up to 20⏎-
-
kerchief in the same pocket. Frequently wash your hands, especially when you are in public areas.⏎- Readers Rating⏎- Rated 5 stars⏎5 / 5 (Reviewers)⏎- Your Rating<|endoftext|>|ARM Technical Support Knowledge Articles|⏎
-
(6-8) Show all Grade Level⏎Resource Type: Activities⏎Results 1 - 7 of 7 matches⏎From Grid to Home⏎Lisa Gardiner, National
-
- 1.⏎- The while loop keeps running while the value of i is equal to
-
- 4 hours⏎Introduce the session with stories about leadership in group activity. You
-
- 8th|⏎|9th - 10th||11th -
SteeringAdding a push along a neuron while the model writes, to see how its output changes. Turning it up or down.
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.
Here is something I have been thinking about lately: I have been thinking about something I have been thinking about lately. So I need to write down my thoughts in order to make sense. But I can't get it. I'm not sure to write down my thoughts. So I need to
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
atmosphere and workplace, HEPA filters, UV lights, laminar flow cabinets, and other biological safety cabinet components protect.⏎Read More @ https://cmibloginsight.blogspot.com/2023/05/biological-safety-c
-
Copyright © 1997, by Random House, Inc., on Infoplease
-
850 at 10 p.m. During that day, Sears kept the same price, Best Buy changed twice and Amazon, 9 times.⏎Call it dynamic pricing.⏎The story of dynamic pricing begins with airline deregulation in 1
-
instance, "hok" warns of serious aerial threats -- usually eagles -- whereas
Connections
Show on mapWhich neuronsOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in earlier layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.
Fed by earlier layers
None: this is the first layer.
Feeds later layers
- L24:1734 weakly suppresses
- L6:109 weakly suppresses
- L4:255 weakly suppresses
- L25:1752 weakly suppresses
- L11:800 weakly suppresses
- L8:127 weakly excites
- L18:677 weakly suppresses
- L9:1900 weakly suppresses
- L4:53 weakly excites blank slot before a number
- L26:30 weakly suppresses
- L25:2515 weakly suppresses
- L17:2353 weakly suppresses
- L18:674 weakly suppresses
- L3:349 weakly suppresses
- L4:1413 weakly excites
- L26:2602 weakly suppresses
- L27:170 weakly excites unclear: spaces in mixed-language text
- L7:835 weakly excites
- L1:1104 weakly suppresses
- L22:2855 weakly suppresses digits inside citation and DOI strings
- L16:139 weakly excites
- L16:1115 weakly excites
- L4:1873 weakly suppresses
- L22:2421 weakly suppresses
Activation histogramA chart of how often a neuron takes each activation value.
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.
Logit lensReading which output tokens a neuron pushes toward or away from, straight off its output weights, before the later layers have their say.
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 layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. instead.
Promotes
- Wars
- ляр
- ucks
- -span
- azine
- 孩子的
- aisal
- stood
- iaux
- IDADE
Suppresses
- If
- tal
- severe
- 若
- omic
- jamin
- If
- ök
- のに
- 弁