INDEX

Explanations

attends to tokens marked with underscores and from tokens marked with brackets in a technical context

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

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Head Attr Weights

0:0.12

1:0.14

2:0.23

3:0.04

4:0.08

5:0.24

6:0.04

7:0.07

Negative Logits

kloped

-0.30

ГЛА

-0.27

ผ

-0.27

 recensement

-0.26

 hauteur

-0.25

urg

-0.24

ובות

-0.23

tri

-0.23

бю

-0.23

 Heimat

-0.22

POSITIVE LOGITS

MemoryWarning

0.36

Autoritní

0.36

 estekak

0.34

AlterField

0.33

disposing

0.33

")));

0.33

 Мексичка

0.33

"])

0.33

 disambiguazione

0.32

ConstraintMaker

0.32

Activations Density 0.895%

attends to tokens marked with underscores and from tokens marked with brackets in a technical context

No Comments

No Known Activations