INDEX

Explanations

repetitive expressions of approval or commendation

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

pretty + adjective

np_acts-logits-general · gemini-2.5-flash-lite

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GEMMA-2-9B-IT @ 20-gemmascope-res-131k

Configuration

google/gemma-scope-9b-it-res/layer_20/width_131k/average_l0_81

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Features

131,072

Data Type

float32

Hook Name

blocks.20.hook_resid_post

Hook Layer

Architecture

jumprelu

Context Size

1,024

Dataset

monology/pile-uncopyrighted

Activation Function

relu

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 Cardona

-0.37

 return

-0.35

 enrolment

-0.35

 location

-0.34

 clave

-0.33

absence

-0.32

DRS

-0.32

Jonas

-0.31

 Location

-0.30

JoinColumn

-0.30

POSITIVE LOGITS

 pretty

1.23

pretty

1.19

 Pretty

1.13

Pretty

1.11

retty

0.95

 ziemlich

0.85

 ſehr

0.81

 quite

0.79

astanza

0.79

 fairly

0.78

Activations Density 0.004%

repetitive expressions of approval or commendation

pretty + adjective

No Comments

No Known Activations