INDEX

Explanations

attends to numbers from token representations appearing later

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

0:0.25

1:0.20

2:0.12

3:0.06

4:0.09

5:0.10

6:0.05

7:0.08

Negative Logits

 tirs

-0.36

 élev

-0.32

 Дата

-0.31

 tirage

-0.31

 Traveling

-0.31

 خارجية

-0.31

Blair

-0.30

USDA

-0.30

 переписи

-0.29

POSITIVE LOGITS

########.

0.42

 AssemblyTitle

0.38

flakes

0.36

/>\

0.36

cleros

0.36

̉

0.35

harusnya

0.34

 Colonna

0.34

 Bourgoin

0.33

hésite

0.33

Activations Density 0.453%

attends to numbers from token representations appearing later

No Comments

No Known Activations