INDEX

Explanations

alcohol-related words, particularly the word "liquor," with varying activations based on the context of the word in the document

oai_token-act-pair · gpt-3.5-turbo

references to alcoholic beverages and liquor-related terms

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 4-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.4.hook_resid_pre

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

24,576

Data Type

torch.float32

Hook Point

blocks.4.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

wered

-0.72

 Hawk

-0.71

Aj

-0.69

DIR

-0.69

 Canterbury

-0.65

pta

-0.65

 Steps

-0.64

CFR

-0.63

 Shogun

-0.62

 Macro

-0.62

POSITIVE LOGITS

 liquor

1.10

cohol

1.03

ocaust

0.96

 beverage

0.89

 licence

0.85

 licenses

0.84

 drinkers

0.83

 alcohol

0.83

 Liqu

0.82

 liqu

0.81

Activations Density 0.005%

alcohol-related words, particularly the word "liquor," with varying activations based on the context of the word in the document

references to alcoholic beverages and liquor-related terms

No Comments

No Known Activations