INDEX

Explanations

instances where the audience is directly addressed

oai_token-act-pair · gpt-3.5-turbo

references to the audience or reader

oai_token-act-pair · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 5-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.5.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.5.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

ayne

-0.69

 Paddock

-0.68

 Called

-0.68

Ī

-0.67

 Cance

-0.67

 Marcos

-0.64

 Mecca

-0.63

 Guam

-0.62

Parameters

-0.62

 Chap

-0.61

POSITIVE LOGITS

 guys

1.17

tub

1.10

 know

0.89

 yourselves

0.89

're

0.82

hei

0.82

endi

0.79

RS

0.77

 filthy

0.76

 gentlemen

0.75

Activations Density 0.055%

instances where the audience is directly addressed

references to the audience or reader

No Comments

No Known Activations