INDEX

Explanations

keywords related to tools and kits

oai_token-act-pair · gpt-3.5-turbo

phrases that reference various types of toolkits

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

 Presidents

-0.72

 transsexual

-0.72

 Dickinson

-0.69

 credits

-0.68

 Braves

-0.66

 partic

-0.66

 Perse

-0.65

 Amendments

-0.65

 heel

-0.63

 Whites

-0.63

POSITIVE LOGITS

kit

4.59

Kit

2.39

kit

1.44

Kit

1.21

 kits

1.20

box

1.20

kar

1.19

krit

1.14

kat

1.14

pack

1.11

Activations Density 0.008%

keywords related to tools and kits

phrases that reference various types of toolkits

No Comments

No Known Activations