INDEX

Explanations

numerical identifiers, especially those related to locations or events

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

0:0.14

1:0.01

2:0.13

3:0.11

4:0.08

5:0.03

6:0.06

7:0.02

8:0.05

9:0.08

10:0.14

11:0.08

Negative Logits

lyak

-1.45

 Stub

-1.43

avorite

-1.37

 Sneak

-1.36

 Schne

-1.35

 wrist

-1.34

 Adin

-1.32

 pursu

-1.25

 Stamina

-1.22

 securely

-1.22

POSITIVE LOGITS

clinton

1.83

;;;;;;;;;;;;

1.79

abama

1.76

DragonMagazine

1.71

united

1.70

addons

1.69

communications

1.64

roads

1.62

road

1.60

history

1.59

Activations Density 0.000%

numerical identifiers, especially those related to locations or events

No Known Activations