INDEX

Explanations

attends to specific informative tokens from arbitrary later tokens indicating their 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

0:0.17

1:0.13

2:0.12

3:0.12

4:0.12

5:0.03

6:0.13

7:0.14

Negative Logits

 ་་

-0.36

 Vols

-0.36

 Heal

-0.36

BOP

-0.35

 Fram

-0.35

 članak

-0.34

 broder

-0.34

 ―――――

-0.34

 Raum

-0.33

PAD

-0.33

POSITIVE LOGITS

 '\\;'

0.44

UnusedPrivate

0.35

SharedDtor

0.34

HasAnnotation

0.34

verwijspagina

0.33

Cuánt

0.32

RunAsync

0.31

tonode

0.31

httphttps

0.30

HtmlAttribute

0.30

Activations Density 0.441%

attends to specific informative tokens from arbitrary later tokens indicating their context

No Comments

No Known Activations