INDEX

Explanations

phrases related to online usernames or identifiers that contain the word "ans"

oai_token-act-pair · gpt-3.5-turbo

references to individuals or groups identified by the term "ans."

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

kefeller

-0.86

bable

-0.82

UTERS

-0.72

displayText

-0.70

 buds

-0.66

è¯

-0.66

 WATCHED

-0.66

 Logged

-0.64

ãĥ¼ãĥ³

-0.64

itude

-0.62

POSITIVE LOGITS

ans

1.19

alis

1.06

igans

0.98

hip

0.89

haw

0.86

omal

0.85

hee

0.84

omen

0.83

laughter

0.83

odon

0.81

Activations Density 0.013%

phrases related to online usernames or identifiers that contain the word "ans"

references to individuals or groups identified by the term "ans."

No Comments

No Known Activations