INDEX

Explanations

verbs ending in 'ete', potentially related to control or manipulation

oai_token-act-pair · gpt-3.5-turbo

repeated patterns of specific sequences of letters and syllables

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

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

spring

-0.68

rises

-0.61

 maiden

-0.59

 Atmospheric

-0.58

 traced

-0.57

 routed

-0.57

 fixme

-0.56

matic

-0.56

 Kush

-0.56

birth

-0.56

POSITIVE LOGITS

ete

1.42

eme

1.15

chnology

1.13

uve

0.97

ula

0.91

ulas

0.86

eta

0.85

herer

0.84

opard

0.84

estern

0.83

Activations Density 0.006%

verbs ending in 'ete', potentially related to control or manipulation

repeated patterns of specific sequences of letters and syllables

No Comments

No Known Activations