INDEX

Explanations

acronyms consisting of 'PR' followed by a single digit number

oai_token-act-pair · gpt-3.5-turbo

references to public relations or public relations-related terms

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

 aver

-0.78

 withdrawn

-0.71

 caves

-0.69

wagen

-0.68

 lone

-0.68

ga

-0.68

 solo

-0.66

 stricken

-0.65

 Collins

-0.65

 beetle

-0.64

POSITIVE LOGITS

PR

4.00

PR

1.72

PRO

1.64

FR

1.51

pr

1.50

PL

1.48

PU

1.44

CR

1.40

BR

1.33

FP

1.32

Activations Density 0.006%

acronyms consisting of 'PR' followed by a single digit number

references to public relations or public relations-related terms

No Comments

No Known Activations