INDEX

Explanations

attends to the token "decoder" from tokens related to message decoding structures

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

0:0.10

1:0.18

2:0.12

3:0.10

4:0.17

5:0.10

6:0.09

7:0.10

Negative Logits

 inútil

-0.23

 noses

-0.23

不利

-0.22

 sbag

-0.22

 förr

-0.22

 enfermera

-0.22

 cognac

-0.21

 tecnica

-0.21

 Tecn

-0.21

HIDE

-0.21

POSITIVE LOGITS

Autoritní

0.37

scalatest

0.35

 يتيمه

0.33

حياته

0.33

LookAnd

0.32

Jeografia

0.31

hasMoreElements

0.31

yakarta

0.30

VersionUID

0.30

Personensuche

0.30

Activations Density 0.222%

attends to the token "decoder" from tokens related to message decoding structures

No Comments

No Known Activations