INDEX

Explanations

mentions of someone being elected or president-elect

oai_token-act-pair · gpt-3.5-turbo

references to the president-elect

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

 Mama

-0.76

Wra

-0.74

Tus

-0.70

 wrapper

-0.68

ra

-0.66

 monster

-0.66

 bloss

-0.63

 Cache

-0.62

 siblings

-0.62

gor

-0.62

POSITIVE LOGITS

elect

4.44

Elect

3.06

 ELECT

2.14

 Elect

2.00

 elect

1.97

election

1.91

electric

1.70

elected

1.49

LECT

1.38

 Electro

1.36

Activations Density 0.016%

mentions of someone being elected or president-elect

references to the president-elect

No Comments

No Known Activations