INDEX

Explanations

phrases signaling attention towards important news updates

New Auto-Interp

Top Features by Cosine Similarity

Configuration

ckkissane/attn-saes-gpt2-small-all-layers/gpt2-small_L3_Hcat_z_lr1.20e-03_l19.00e-01_ds24576_bs4096_dc1.00e-06_rsanthropic_rie25000_nr4_v9.pt

Prompts (Dashboard)

36,864 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

24,576

Data Type

float32

Hook Name

blocks.3.attn.hook_z

Hook Layer

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Activation Function

relu

Embeds

IFrame

Link

Head Attr Weights

0:0.02

1:0.07

2:0.09

3:0.07

4:0.03

5:0.05

6:0.07

7:0.10

8:0.06

9:0.19

10:0.05

11:0.15

Negative Logits

soDeliveryDate

-1.62

 envelop

-1.33

 metaphor

-1.25

lik

-1.22

 mont

-1.21

artney

-1.17

isSpecialOrderable

-1.16

ē

-1.14

retty

-1.14

 embracing

-1.13

POSITIVE LOGITS

 Bucc

1.20

 Attention

1.20

ANCE

1.10

 Memory

1.09

 Paras

1.09

 Chemical

1.08

 Cellular

1.08

Pir

1.07

 ASAP

1.07

 Berman

1.06

Activations Density 0.001%

phrases signaling attention towards important news updates

No Known Activations