INDEX

Explanations

references and external links in the text

oai_token-act-pair · gpt-4o-mini Triggered by @bot

Looking at the activations, this neuron activates strongly (values 4-5) on tokens like "os", "i", "e", "ne" that appear to be parts of external link sections in Wikipedia articles across multiple languages (Spanish "Enlaces externos", Italian "Collegamenti esterni",

oai_token-act-pair · claude-4-5-sonnet Triggered by @jyhe0408

New Auto-Interp

Top Features by Cosine Similarity

Comparing With LLAMA3.1-8B @ 15-llamascope-res-32k

Configuration

fnlp/Llama3_1-8B-Base-LXR-8x/Llama3_1-8B-Base-L15R-8x

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

cerebras/SlimPajama-627B

Features

32,768

Data Type

bfloat16

Hook Name

blocks.15.hook_resid_post

Hook Layer

Architecture

jumprelu

Context Size

1,024

Dataset

cerebras/SlimPajama-627B

Activation Function

relu

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

blr

-0.15

ariant

-0.15

azor

-0.14

½

-0.14

eof

-0.14

hol

-0.14

iron

-0.14

Hol

-0.13

ob

-0.13

Bon

-0.13

POSITIVE LOGITS

agn

0.18

ï»¿#

0.15

ouri

0.15

ah

0.14

idis

0.14

hex

0.13

\OptionsResolver

0.13

Î¼Î²

0.13

ãĥ¼ãĥł

0.13

thing

0.13

Activations Density 0.009%

references and external links in the text

Looking at the activations, this neuron activates strongly (values 4-5) on tokens like "os", "i", "e", "ne" that appear to be parts of external link sections in Wikipedia articles across multiple languages (Spanish "Enlaces externos", Italian "Collegamenti esterni",

No Comments

No Known Activations