INDEX

Explanations

news captions or headlines ending with '/'

oai_token-act-pair · gpt-3.5-turbo

occurrences of references to news article indicators or metadata

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

racuse

-0.80

 disgu

-0.71

 fool

-0.71

 idol

-0.70

 imperson

-0.70

atche

-0.69

 frogs

-0.66

 lull

-0.65

 chop

-0.65

 advert

-0.64

POSITIVE LOGITS

AFP

1.08

Getty

1.06

EVA

1.04

READ

0.89

Library

0.87

WAR

0.86

WARN

0.82

BUS

0.82

DATA

0.81

European

0.80

Activations Density 0.051%

news captions or headlines ending with '/'

occurrences of references to news article indicators or metadata

No Comments

No Known Activations