INDEX

Explanations

attends to index-related tokens from other tokens appearing 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

0:0.08

1:0.10

2:0.09

3:0.14

4:0.15

5:0.16

6:0.13

7:0.11

Negative Logits

yser

-0.27

ges

-0.27

照

-0.25

思

-0.25

YR

-0.24

 myth

-0.24

gew

-0.24

 consult

-0.24

 Consult

-0.23

ses

-0.23

POSITIVE LOGITS

RegistryLite

0.44

########.

0.40

Tikang

0.38

AndEndTag

0.37

IsContent

0.37

tvguidetime

0.35

 onCreateView

0.35

 AssemblyProduct

0.35

TagMode

0.34

 تضيفلها

0.33

Activations Density 0.536%

attends to index-related tokens from other tokens appearing in the sequence

No Comments

No Known Activations