INDEX

Explanations

phrases related to misleading or deception

oai_token-act-pair · gpt-3.5-turbo

instances of the word "led" in various contexts

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

 separating

-0.67

 diss

-0.64

 disgu

-0.61

 disguise

-0.59

 recip

-0.58

 separ

-0.58

 amen

-0.58

 obtaining

-0.57

ect

-0.57

Ext

-0.56

POSITIVE LOGITS

led

4.23

ling

2.44

les

2.19

ledged

1.92

Led

1.88

LED

1.76

lled

1.74

lings

1.62

lers

1.61

lead

1.44

Activations Density 0.017%

phrases related to misleading or deception

instances of the word "led" in various contexts

No Comments

No Known Activations