INDEX

Explanations

attends to specific tokens related to negative numerical values from tokens that are single characters

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

1:0.12

2:0.10

3:0.12

4:0.11

5:0.09

6:0.15

7:0.17

Negative Logits

erle

-0.32

 testGet

-0.31

izarse

-0.31

INCREF

-0.30

 împreună

-0.29

évaluateur

-0.29

 dAtA

-0.29

GLint

-0.29

اریخ

-0.29

 quæ

-0.28

POSITIVE LOGITS

 initComponents

0.33

 numerus

0.32

epresidente

0.32

 المعيارى

0.30

 Rump

0.29

uresti

0.29

payee

0.28

imanapun

0.28

 Clément

0.28

adpleegd

0.28

Activations Density 0.010%

attends to specific tokens related to negative numerical values from tokens that are single characters

No Comments

No Known Activations