INDEX

Explanations

attends to hide flag-related tokens from arbitrary subsequent 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

1:0.13

2:0.12

3:0.17

4:0.13

5:0.05

6:0.12

7:0.13

Negative Logits

RenderAtEndOf

-0.46

SourceChecksum

-0.40

 pinulongan

-0.40

PreferredItem

-0.40

AsUp

-0.39

ftagPool

-0.39

ьаж

-0.38

RegressionTest

-0.38

 चीज़ों

-0.38

Personendaten

-0.38

POSITIVE LOGITS

 Britton

0.29

indt

0.28

itano

0.27

itm

0.25

 Centr

0.25

daille

0.25

side

0.25

name

0.25

Corollary

0.24

ً

0.24

Activations Density 0.001%

attends to hide flag-related tokens from arbitrary subsequent tokens

No Comments

No Known Activations