INDEX

Explanations

attends to both the same token from arbitrary earlier tokens

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

0:0.07

1:0.16

2:0.14

3:0.18

4:0.13

5:0.11

6:0.09

7:0.08

Negative Logits

});

-0.41

])));

-0.36

tanleria

-0.35

fromnode

-0.35

enterOuterAlt

-0.35

);

-0.34

 sẻ

-0.33

 незавершена

-0.33

IsContent

-0.32

Doh

-0.32

POSITIVE LOGITS

 canes

0.33

xter

0.33

acia

0.30

 plá

0.30

arean

0.29

LLocation

0.29

liday

0.29

 Kant

0.29

дата

0.28

="%

0.28

Activations Density 0.006%

attends to both the same token from arbitrary earlier tokens

No Comments

No Known Activations