INDEX

Explanations

attends to initial tokens marked with "first" from subsequent tokens that follow

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GEMMA-2-2B @ 0-gemmascope-att-16k

Configuration

google/gemma-scope-2b-pt-att/layer_0/width_16k/average_l0_104

Prompts (Dashboard)

36,864 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Features

16,384

Data Type

float32

Hook Name

blocks.0.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

0:0.74

1:0.03

2:0.01

3:0.02

4:0.09

5:0.05

6:0.01

7:0.02

Negative Logits

ora

-0.34

ara

-0.34

aw

-0.33

ade

-0.31

дак

-0.31

ana

-0.31

one

-0.30

ina

-0.29

<sup>

-0.29

ul

-0.29

POSITIVE LOGITS

GraphicsUnit

0.40

 Italijanski

0.38

ècie

0.38

Efq

0.36

 myſelf

0.34

 كومونز

0.34

 تضيفلها

0.34

HideFlags

0.33

ectady

0.33

 snippetHide

0.33

Activations Density 0.023%

attends to initial tokens marked with "first" from subsequent tokens that follow

No Comments

No Known Activations