INDEX

Explanations

phrases related to crime or suspicious activity

oai_token-act-pair · gpt-3.5-turbo

references to a specific brand or company, particularly "Dod."

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

 compute

-0.74

 Afghans

-0.71

 grade

-0.68

LEASE

-0.68

 computing

-0.67

 shoulder

-0.67

Ħ¢

-0.66

achelor

-0.64

Austin

-0.64

 metast

-0.63

POSITIVE LOGITS

Dod

3.87

dod

2.21

 Brid

1.30

hod

1.27

anim

1.22

 Phill

1.14

Rak

1.06

iod

1.06

 Kers

1.01

odo

0.98

Activations Density 0.028%

phrases related to crime or suspicious activity

references to a specific brand or company, particularly "Dod."

No Comments

No Known Activations