INDEX

Explanations

attends to the conjunctions and connecting phrases that relate two parts from earlier tokens to later 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

0:0.10

1:0.10

2:0.08

3:0.13

4:0.27

5:0.04

6:0.13

7:0.10

Negative Logits

bootstrapcdn

-0.41

 AssemblyCulture

-0.40

 médec

-0.33

 vectoriales

-0.33

 preuves

-0.33

قایناق‌لار

-0.33

 vérit

-0.33

 câbles

-0.32

 vectorielles

-0.32

 avoient

-0.32

POSITIVE LOGITS

sei

0.29

 AssertionError

0.28

 disemb

0.28

writeFieldEnd

0.28

ργ

0.27

 vulcan

0.27

 بيها

0.27

HasBeenSet

0.27

EndProject

0.26

ilien

0.26

Activations Density 0.198%

attends to the conjunctions and connecting phrases that relate two parts from earlier tokens to later tokens

No Comments

No Known Activations