INDEX

Explanations

instances of access and connection in descriptions of spaces or structures

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

reveal bear-like claws

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

 uſed

-0.69

 ſta

-0.68

 ſte

-0.67

 myſelf

-0.67

 ſtand

-0.67

 ſche

-0.67

ſelf

-0.64

 anſ

-0.60

ſelves

-0.60

 ſtill

-0.59

POSITIVE LOGITS

 חיצוניים

0.42

0.41

the

0.40

 front

0.35

principalTable

0.34

 open

0.33

an

0.33

sp

0.33

open

0.32

front

0.32

Activations Density 0.003%

instances of access and connection in descriptions of spaces or structures

reveal bear-like claws

No Comments

No Known Activations