INDEX

Explanations

years in the 1990s, with a particularly high activation for the year 1990

oai_token-act-pair · gpt-3.5-turbo

references to the year 1990 and its context within historical narratives

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 5-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.5.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.5.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

hire

-0.78

bow

-0.74

holder

-0.70

 lightsaber

-0.69

 clipboard

-0.69

edge

-0.69

Edge

-0.68

llers

-0.67

cloth

-0.67

semble

-0.67

POSITIVE LOGITS

ĸļ

1.02

ãĤ¦ãĤ¹

0.74

 GOODMAN

0.70

ãĥŁ

0.69

cture

0.68

ãĥ¤

0.65

emonium

0.65

aji

0.64

zsche

0.64

å¹

0.63

Activations Density 0.017%

years in the 1990s, with a particularly high activation for the year 1990

references to the year 1990 and its context within historical narratives

No Comments

No Known Activations