INDEX

Explanations

attends to empty or non-informative tokens from legal or formal context tokens

oai_attention-head · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GEMMA-2-2B @ 0-gemmascope-att-16k

Configuration

google/gemma-scope-2b-pt-att/layer_0/width_16k/average_l0_104

Prompts (Dashboard)

36,864 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Features

16,384

Data Type

float32

Hook Name

blocks.0.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

0:0.03

1:0.04

2:0.04

3:0.39

4:0.07

5:0.12

6:0.13

7:0.14

Negative Logits

↵

-1.25

<eos>

-1.17

-1.15

↵↵

-1.15

-1.14

-1.05

in

-1.03

-1.02

and

-1.01

POSITIVE LOGITS

Efq

2.63

 myſelf

2.59

 itſelf

2.50

 Monfieur

2.28

 Jefus

2.23

 Theſe

2.22

 ſeveral

2.19

 purpoſe

2.17

^(@)

2.16

 raiſ

2.11

Activations Density 0.001%

attends to empty or non-informative tokens from legal or formal context tokens

No Comments

No Known Activations