INDEX

Explanations

the word "Knight" with high activation values

oai_token-act-pair · gpt-3.5-turbo

references to a specific individual named Knight

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

namese

-0.75

endum

-0.73

ongs

-0.72

 Syrians

-0.72

ipes

-0.69

Reviewer

-0.69

tin

-0.68

anned

-0.66

BILL

-0.66

gin

-0.64

POSITIVE LOGITS

 Knight

4.01

Knight

2.74

 Knights

2.23

 knight

2.08

 knights

1.58

 Paladin

1.45

 Mage

1.33

 Castle

1.22

 Swordsman

1.20

 Monk

1.20

Activations Density 0.023%

the word "Knight" with high activation values

references to a specific individual named Knight

No Comments

No Known Activations