INDEX

Explanations

references to websites or email addresses ending in ".wh"

oai_token-act-pair · gpt-3.5-turbo

instances of the word "who."

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

²¾

-0.72

 Ankara

-0.65

 Lazarus

-0.65

 Mali

-0.63

 Medina

-0.59

 rein

-0.59

 improvised

-0.59

 Rein

-0.58

 Universe

-0.58

ITION

-0.57

POSITIVE LOGITS

wh

4.19

Wh

2.37

WH

1.74

Wh

1.65

wh

1.45

white

1.32

WH

1.30

who

1.29

Whit

1.24

whe

1.23

Activations Density 0.008%

references to websites or email addresses ending in ".wh"

instances of the word "who."

No Comments

No Known Activations