unclear: 'if' opening a statement, 'local'
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 fairly consistent: several of them 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 is hard to pin down. It fires most on the word 'if' when it starts a line of code or a condition, and on the word 'local' in ordinary sentences. When it was turned up, the model's writing became repetitive and stuck on the word 'sunny', which does not clearly relate to what the neuron responds to.
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 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.
-
headed things you can do - but look at the C++-tagged posts in this blog for examples of what the compiler thinks isn't a bad idea.)⏎if (x=0) init_subsystem();⏎So lots of books have been written
-
be achieved with the following code:⏎TYPE List = REF RECORD next: List; value: INTEGER; END;⏎VAR x := NEW(List, value:=1); BEGIN x.next :=
-
popular⏎foldr. This package provides that same function for the⏎Bool data type; a function called⏎bool function is a replacement for the build-in⏎if then else-syntax. However, since it is a function instead of syntax, it
-
different things I need to return. If I ask for something that isn't there, I get an exception, which is life as usual for a Python programmer.⏎if not server_response['successful']:⏎etag = server_response['file_handle'].info().
-
local traditions and folklore with just a touch of modern influence<|endoftext|>In the last post,
-
local taxes to make them more comparable to gasoline taxes in other countries. Others want these
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 the history of the two main things that are about the history of the two main things that are about the history of the two main things that are about the history of the two main things that are about the history of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
⏎Updated: Sep 24, 2018⏎Detoxification – you’ve heard this word many times over the last few years, but unless you have actually participated in a detoxification program, you probably don’t really understand
-
frequently if they had met the dog before.⏎Control tests were done to ensure that the dogs
-
an emergency.⏎Rest until you are fully recovered⏎after you leave the hospital. This may take 4 to 6 weeks. Your healthcare provider will tell you if you need to be on bedrest. This means you stay in bed and do
-
55 file extension⏎Recommended software programs are sorted by OS platform (Windows, Mac OS X etc
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
- L25:2536 weakly suppresses
- L3:363 weakly excites
- L1:1572 weakly suppresses
- L8:1715 weakly suppresses
- L21:303 weakly excites
- L20:1321 weakly suppresses
- L27:1814 weakly suppresses
- L27:2317 weakly suppresses
- L1:2368 weakly suppresses
- L1:2080 weakly excites
- L5:2960 weakly suppresses
- L4:362 weakly suppresses
- L6:724 weakly suppresses
- L17:2787 weakly excites comma after 'that' or a verb before an aside
- L4:1859 weakly suppresses
- L25:2756 weakly suppresses
- L23:13 weakly excites
- L19:2586 weakly suppresses
- L13:3063 weakly excites
- L27:920 weakly suppresses
- L2:458 weakly excites
- L20:1031 weakly suppresses
- L5:164 weakly excites
- L17:77 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
- nick
- found
- termin
- lix
- planning
- ElapsedTime
- affected
- related
- like
- Termin
Suppresses
- 婢
- 因为她
- sense
- 主义者
- auer
- 男篮
- 剃
- ciné
- <textarea
- ело