INDEX

Explanations

mentions within the document related to the broadcaster NBC

oai_token-act-pair · gpt-3.5-turbo

references to the NBC network

oai_token-act-pair · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 3-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.3.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.3.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

aunder

-0.83

cific

-0.78

wikipedia

-0.77

piration

-0.74

andr

-0.73

erva

-0.72

ebook

-0.72

rosse

-0.69

 McCartney

-0.68

ende

-0.67

POSITIVE LOGITS

Universal

1.25

 affiliate

0.90

 News

0.81

WS

0.80

SN

0.79

NBC

0.79

 Universal

0.76

 affiliates

0.74

NBC

0.74

 Networks

0.72

Activations Density 0.007%

mentions within the document related to the broadcaster NBC

references to the NBC network

No Comments

No Known Activations