INDEX

Explanations

terms related to online harassment and threats

oai_token-act-pair · gpt-3.5-turbo

instances of hate speech and threats of violence

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

natureconservancy

-0.83

 Alive

-0.77

ITNESS

-0.76

 Accounting

-0.71

 miracle

-0.71

 stabilize

-0.68

etermin

-0.67

addons

-0.66

 transitional

-0.66

illac

-0.65

POSITIVE LOGITS

 hurled

1.58

 directed

1.37

 levied

1.33

 leveled

1.29

 against

1.18

against

1.16

 spew

1.06

 leve

1.06

 slurs

1.03

 emanating

1.03

Activations Density 0.445%

terms related to online harassment and threats

instances of hate speech and threats of violence

No Comments

No Known Activations