INDEX

Explanations

attends to concept-related tokens from specific phrasing or concepts elaborated upon later in the sequence

oai_attention-head · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GEMMA-2-2B @ 10-gemmascope-att-16k

Configuration

google/gemma-scope-2b-pt-att/layer_10/width_16k/average_l0_70

Prompts (Dashboard)

36,864 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Features

16,384

Data Type

float32

Hook Name

blocks.10.attn.hook_z

Hook Layer

Architecture

jumprelu

Context Size

1,024

Dataset

monology/pile-uncopyrighted

Activation Function

relu

Embeds

IFrame

Link

Not in Any Lists

No Comments

Head Attr Weights

0:0.18

1:0.24

2:0.15

3:0.10

4:0.11

5:0.04

6:0.05

7:0.09

Negative Logits

 kasarigan

-0.41

hdashline

-0.29

 EconPapers

-0.28

richTextPanel

-0.28

}]

-0.27

olesale

-0.26

hless

-0.26

Географија

-0.25

 mxArray

-0.24

resaid

-0.24

POSITIVE LOGITS

 MonoBehaviour

0.26

SimpleDateFormat

0.26

الإنجليزية

0.25

NgModule

0.23

 marchés

0.23

featureID

0.23

 naturali

0.23

schedulers

0.22

stock

0.22

kta

0.22

Activations Density 0.589%

attends to concept-related tokens from specific phrasing or concepts elaborated upon later in the sequence

No Comments

No Known Activations