INDEX

Explanations

phrases related to debut or first appearance

oai_token-act-pair · gpt-3.5-turbo

terms related to the release or introduction of new content or subjects

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

 Sovere

-0.78

 Hume

-0.73

athi

-0.72

 Trave

-0.71

 Dublin

-0.71

EVE

-0.70

ccess

-0.69

IRE

-0.69

 Corpus

-0.69

esse

-0.67

POSITIVE LOGITS

 debuted

1.83

 Cuban

1.39

 gimmick

1.23

 premiered

1.15

ollah

1.11

 villain

1.10

 pediatric

1.05

Cub

0.99

 villains

0.95

usr

0.94

Activations Density 0.039%

phrases related to debut or first appearance

terms related to the release or introduction of new content or subjects

No Comments

No Known Activations