INDEX

Explanations

references to architectural elements and features

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

wall displays

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

Rüyada

-0.50

Carriera

-0.46

 régulière

-0.43

 équilibr

-0.41

 nég

-0.40

 régulier

-0.40

Diweddarwch

-0.39

 esboço

-0.38

 onts

-0.38

Hochspringen

-0.38

POSITIVE LOGITS

 wall

0.86

 mural

0.70

 Wall

0.62

 murals

0.62

 WALL

0.61

 parete

0.60

 walls

0.58

 panel

0.57

wall

0.56

Wall

0.56

Activations Density 0.021%

references to architectural elements and features

wall displays

No Comments

No Known Activations