INDEX

Explanations

terms related to low-quality or undesirable items

oai_token-act-pair · gpt-3.5-turbo

references to "junk" 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 @ 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

Æ

-0.80

cing

-0.79

oral

-0.71

ndum

-0.70

mberg

-0.70

gypt

-0.68

idency

-0.67

sis

-0.65

ignty

-0.65

utic

-0.64

POSITIVE LOGITS

 heap

0.98

 junk

0.97

 drawer

0.89

 pile

0.85

ies

0.80

busters

0.79

 mail

0.78

unks

0.77

ãģı

0.77

etsu

0.77

Activations Density 0.010%

terms related to low-quality or undesirable items

references to "junk" in various contexts

No Comments

No Known Activations