INDEX

Explanations

attends to informational tokens from computationally related 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

0:0.39

1:0.20

2:0.15

3:0.07

4:0.03

5:0.01

6:0.03

7:0.07

Negative Logits

aarrggbb

-0.51

NameInMap

-0.42

#+#

-0.42

featureID

-0.41

principalColumn

-0.39

 JSTOR

-0.39

Tikang

-0.38

__':

-0.38

hoea

-0.38

 Administrativna

-0.38

POSITIVE LOGITS

iarkan

0.20

Bo

0.20

 badan

0.19

ṁ

0.19

</td>

0.19

</h1>

0.19

regel

0.18

 görün

0.18

prin

0.18

Bo

0.18

Activations Density 0.663%

attends to informational tokens from computationally related tokens

No Comments

No Known Activations