INDEX

Explanations

phrases or terms related to specific names or locations, particularly ending with "arat"

oai_token-act-pair · gpt-3.5-turbo

references to educational institutions, specifically those with "arat" and related terms

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 1-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.1.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.1.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

served

-0.84

bed

-0.71

pecially

-0.65

hesion

-0.63

ilst

-0.63

 ignition

-0.61

 RELE

-0.60

THR

-0.60

 DRAG

-0.59

RESULTS

-0.59

POSITIVE LOGITS

arat

0.98

nian

0.88

alore

0.79

nam

0.76

kees

0.76

nia

0.76

itious

0.75

nea

0.73

ket

0.73

inth

0.73

Activations Density 0.025%

phrases or terms related to specific names or locations, particularly ending with "arat"

references to educational institutions, specifically those with "arat" and related terms

No Comments

No Known Activations