INDEX

Explanations

attends to numeric tokens from their adjacent, possibly related 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

IFrame

Link

Not in Any Lists

No Comments

Head Attr Weights

0:0.11

1:0.14

2:0.06

3:0.07

4:0.39

5:0.05

6:0.05

7:0.09

Negative Logits

AndEndTag

-0.29

AccessorTable

-0.28

 CABLE

-0.28

 charts

-0.27

atro

-0.27

 Galen

-0.26

 Bart

-0.26

xter

-0.26

CardType

-0.26

 CARDS

-0.26

POSITIVE LOGITS

migiano

0.45

 CreateTagHelper

0.42

stantial

0.38

tinyos

0.34

seamnă

0.34

 AssemblyTitle

0.33

égorie

0.32

σουμε

0.32

InternalFrame

0.32

ureux

0.32

Activations Density 0.149%

attends to numeric tokens from their adjacent, possibly related tokens

No Comments

No Known Activations