INDEX

Explanations

acronyms starting with "AW" at a high activation level

oai_token-act-pair · gpt-3.5-turbo

mentions of the phrase "AW" 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 @ 0-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.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.0.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

rine

-0.71

roid

-0.71

lude

-0.69

onic

-0.66

conom

-0.65

cale

-0.64

 landlords

-0.64

mpeg

-0.64

ram

-0.63

ultural

-0.63

POSITIVE LOGITS

AW

3.97

AH

1.39

aw

1.24

AF

1.22

AW

1.22

WR

1.21

FW

1.16

Aw

1.16

EW

1.10

OUT

1.09

Activations Density 0.019%

acronyms starting with "AW" at a high activation level

mentions of the phrase "AW" in various contexts

No Comments

No Known Activations