INDEX

Explanations

the word "which"

New Auto-Interp

Top Features by Cosine Similarity

Configuration

ckkissane/attn-saes-gpt2-small-all-layers/gpt2-small_L5_Hcat_z_lr1.20e-03_l11.00e+00_ds49152_bs4096_dc1.00e-06_rsanthropic_rie25000_nr4_v9.pt

Prompts (Dashboard)

36,864 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

49,152

Data Type

float32

Hook Name

blocks.5.attn.hook_z

Hook Layer

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Activation Function

relu

Embeds

IFrame

Link

Head Attr Weights

0:0.07

1:0.05

2:0.09

3:0.09

4:0.09

5:0.06

6:0.07

7:0.09

8:0.06

9:0.08

10:0.09

11:0.09

Negative Logits

']

-1.58

し

-1.49

 Paran

-1.48

FAQ

-1.48

SourceFile

-1.48

 Associates

-1.48

)=(

-1.47

 antiv

-1.44

 Favorite

-1.43

NULL

-1.43

POSITIVE LOGITS

uesday

1.79

‎

1.75

zee

1.74

 culminated

1.58

 Latvia

1.56

 poppy

1.56

sembly

1.53

chwitz

1.50

eport

1.50

ylan

1.50

Activations Density 0.000%

the word "which"

No Known Activations