INDEX

Explanations

proper nouns, particularly names associated with individuals or teams

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

0:0.03

1:0.04

2:0.08

3:0.28

4:0.03

5:0.02

6:0.07

7:0.12

8:0.04

9:0.04

10:0.05

11:0.13

Negative Logits

hower

-1.17

CSS

-1.15

 lapse

-1.11

 apologies

-1.09

ascript

-1.05

 Became

-1.04

 photographers

-1.03

 Visit

-1.02

minecraft

-1.02

 Centauri

-1.02

POSITIVE LOGITS

UME

1.40

sbm

1.29

ilib

1.25

idated

1.25

omb

1.25

amph

1.23

andi

1.18

andro

1.18

pees

1.17

arters

1.17

Activations Density 0.005%

proper nouns, particularly names associated with individuals or teams

No Known Activations