INDEX

Explanations

attends to the last closing double asterisk from the previous token marked as a link on the page

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

IFrame

Link

Not in Any Lists

No Comments

Head Attr Weights

0:0.08

1:0.11

2:0.09

3:0.10

4:0.09

5:0.04

6:0.19

7:0.25

Negative Logits

AndEndTag

-0.43

 Houſe

-0.35

 houſe

-0.33

 createContext

-0.31

extAlignment

-0.31

 fubject

-0.30

 themſelves

-0.29

 purpoſe

-0.29

 ActionResult

-0.29

 RSSSF

-0.28

POSITIVE LOGITS

basicConfig

0.28

amo

0.25

orin

0.24

stam

0.24

Ranks

0.24

requireNonNull

0.23

 Gifford

0.23

ban

0.23

am

0.23

Externí

0.22

Activations Density 0.025%

attends to the last closing double asterisk from the previous token marked as a link on the page

No Comments

No Known Activations