INDEX

Explanations

lines containing news headlines or text related to news articles from Reuters

oai_token-act-pair · gpt-3.5-turbo

negative sentiment or criticism

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 10-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.10.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.10.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

haus

-0.80

halla

-0.76

 rele

-0.76

 respons

-0.74

ichick

-0.73

bryce

-0.68

 locality

-0.68

 ourselves

-0.68

sie

-0.68

 ignor

-0.66

POSITIVE LOGITS

 Thousands

1.36

 Hundreds

1.26

 Dozens

1.24

 Nearly

1.22

 Despite

1.14

 Protesters

1.12

 Several

1.11

 Authorities

1.11

 Former

1.10

 Scientists

1.10

Activations Density 0.027%

lines containing news headlines or text related to news articles from Reuters

negative sentiment or criticism

No Comments

No Known Activations