INDEX

Explanations

people's names from interviews or news reports

oai_token-act-pair · gpt-3.5-turbo

proper nouns, particularly names and titles

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 11-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.11.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.11.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

edition

-0.57

stro

-0.56

Introduced

-0.56

due

-0.56

ushima

-0.55

astered

-0.54

Torrent

-0.53

orest

-0.53

ords

-0.52

 earthqu

-0.52

POSITIVE LOGITS

 whether

1.48

why

1.42

how

1.27

 what

1.19

why

1.14

whether

1.13

if

1.12

WHY

1.05

what

1.04

 questions

1.00

Activations Density 0.118%

people's names from interviews or news reports

proper nouns, particularly names and titles

No Comments

No Known Activations