INDEX

Explanations

attends to terms related to a specific token from related structural tokens

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

0:0.08

1:0.09

2:0.10

3:0.13

4:0.23

5:0.09

6:0.14

7:0.10

Negative Logits

 Мексичка

-0.51

migrationBuilder

-0.49

Personensuche

-0.46

 Wikimedijinoj

-0.46

LookAnd

-0.44

 فريبيس

-0.43

 ویکی‌پدی

-0.42

ustimmung

-0.41

MigrationBuilder

-0.40

parsedMessage

-0.40

POSITIVE LOGITS

 Arend

0.31

ciuta

0.30

 bombs

0.29

 Hinds

0.28

 Koko

0.28

withOpacity

0.27

ziehen

0.26

டுத்த

0.26

minecraft

0.26

Activations Density 98.145%

attends to terms related to a specific token from related structural tokens

No Comments

No Known Activations

attends to terms related to a specific token from related structural tokens

No Comments

No Known Activations