INDEX

Explanations

mentions of individuals working or collaborating on various projects

oai_token-act-pair · gpt-3.5-turbo

instances of the word "worked."

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

iba

-0.65

 initiation

-0.64

 invasion

-0.57

 invading

-0.57

mount

-0.57

 indict

-0.56

rising

-0.55

ogens

-0.54

 Registrar

-0.54

 Deity

-0.54

POSITIVE LOGITS

 worked

3.39

worked

1.99

 WORK

1.82

 works

1.70

 collaborated

1.69

Work

1.63

 Work

1.63

 work

1.62

work

1.54

Working

1.52

Activations Density 0.018%

mentions of individuals working or collaborating on various projects

instances of the word "worked."

No Comments

No Known Activations