INDEX

Explanations

instances of phrases referring to being the top or best in a list or ranking

oai_token-act-pair · gpt-3.5-turbo

references to "number one" or top rankings in various contexts

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 5-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.5.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.5.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

ages

-0.80

 Balt

-0.69

 Spaces

-0.68

onics

-0.68

antic

-0.66

 Ages

-0.66

apers

-0.65

uras

-0.64

areth

-0.64

 Distance

-0.63

POSITIVE LOGITS

 priority

0.74

 contender

0.73

eree

0.72

 offender

0.69

 loader

0.69

 trending

0.68

ecause

0.68

 hitter

0.68

maker

0.67

 spot

0.67

Activations Density 0.031%

instances of phrases referring to being the top or best in a list or ranking

references to "number one" or top rankings in various contexts

No Comments

No Known Activations