INDEX

Explanations

words related to editorial content

oai_token-act-pair · gpt-3.5-turbo

suffixes related to characteristics or states of being

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

 Atlas

-0.72

 Olympus

-0.57

 feat

-0.56

 Week

-0.54

Khe

-0.53

feat

-0.53

see

-0.52

Ot

-0.52

pps

-0.51

 Eagles

-0.51

POSITIVE LOGITS

ially

4.55

ials

2.48

IAL

2.33

ial

2.19

atively

1.79

iate

1.61

iated

1.51

iously

1.44

ally

1.38

iates

1.36

Activations Density 0.009%

words related to editorial content

suffixes related to characteristics or states of being

No Comments

No Known Activations