INDEX

Explanations

information-related words and phrases

oai_token-act-pair · gpt-3.5-turbo

mentions of being informed or updated

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

strip

-0.78

 weakness

-0.74

 Boom

-0.66

 Chaos

-0.60

 series

-0.60

px

-0.59

row

-0.59

ifix

-0.59

 Layer

-0.58

 stretch

-0.58

POSITIVE LOGITS

 informed

3.74

informed

2.38

 notified

1.91

 informs

1.78

 informing

1.72

 inform

1.62

 alerted

1.61

 educated

1.48

 briefed

1.44

 advised

1.42

Activations Density 0.023%

information-related words and phrases

mentions of being informed or updated

No Comments

No Known Activations