INDEX

Explanations

words related to names and identities

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

New Auto-Interp

Configuration

ctigges/pythia-70m-deduped__res-sm_processed/2-res-sm

Prompts (Dashboard)

32,768 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Features

32,768

Data Type

torch.float32

Hook Name

blocks.2.hook_resid_post

Hook Layer

Architecture

standard

Context Size

128

Dataset

EleutherAI/the_pile_deduplicated

Activation Function

relu

Embeds

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

cknowled

-1.85

 systematic

-1.60

 explanatory

-1.60

ncia

-1.59

 jeopardy

-1.51

 recess

-1.44

 hearsay

-1.40

 Lisbon

-1.37

 residual

-1.36

vdots

-1.35

POSITIVE LOGITS

§

2.69

Īĺ

2.38

Ĵ

2.33

£

2.32

¬

2.31

ŀ

2.31

®

2.25

Ł

2.22

ļ

2.16

ĭ

2.14

Activations Density 0.383%

words related to names and identities

No Comments

No Known Activations