INDEX

Explanations

attends to highlighted tokens from empty or neutral 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

0:0.12

1:0.15

2:0.13

3:0.11

4:0.14

5:0.14

6:0.07

7:0.09

Negative Logits

 Obrador

-0.45

RTDA

-0.44

 disambiguazione

-0.43

rrggbb

-0.42

 AssemblyTitle

-0.41

Демографія

-0.40

ValueStyle

-0.39

 ujednoznacz

-0.37

addPreferredGap

-0.37

 насељу

-0.37

POSITIVE LOGITS

 recirc

0.27

 Cato

0.26

 Haller

0.25

 اع

0.25

****************

0.25

 recue

0.24

ๆ

0.24

 sign

0.23

formik

0.23

 Einsch

0.23

Activations Density 0.013%

attends to highlighted tokens from empty or neutral tokens

No Comments

No Known Activations