INDEX

Explanations

attends to error-related tokens from associated correction-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

0:0.09

1:0.10

2:0.12

3:0.06

4:0.07

5:0.03

6:0.12

7:0.37

Negative Logits

Portail

-0.43

 كومونز

-0.32

Autoritní

-0.25

kuuta

-0.25

 geslacht

-0.24

twimg

-0.23

 Torah

-0.23

 nahilalakip

-0.23

 CreateTagHelper

-0.23

 Loaf

-0.23

POSITIVE LOGITS

fjspx

0.40

IUrlHelper

0.35

AndEndTag

0.34

InstrumentedTest

0.33

UnsafeEnabled

0.33

MergeFrom

0.32

 صوتيه

0.32

IndentedString

0.32

 Miquel

0.32

 bianchi

0.32

Activations Density 0.646%

attends to error-related tokens from associated correction-related tokens

No Comments

No Known Activations