INDEX

Explanations

attends to tokens related to specific actions or processes from tokens indicating categories or lists

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

0:0.12

1:0.09

2:0.06

3:0.08

4:0.10

5:0.08

6:0.25

7:0.18

Negative Logits

 pleaſure

-0.44

 ſch

-0.44

 purpoſe

-0.43

 ſta

-0.42

 CreateTagHelper

-0.41

 becauſe

-0.41

 myſelf

-0.40

 houſe

-0.40

 iſt

-0.40

تقاوى

-0.39

POSITIVE LOGITS

</em>

0.36

‘

0.34

</u>

0.32

дешь

0.31

EndProject

0.30

vs

0.29

__;

0.29

</h5>

0.29

0.28

Activations Density 0.006%

attends to tokens related to specific actions or processes from tokens indicating categories or lists

No Comments

No Known Activations