INDEX

Explanations

proper nouns, particularly names and titles related to individuals

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

0:0.02

1:0.01

2:0.05

3:0.04

4:0.04

5:0.02

6:0.24

7:0.11

8:0.04

9:0.04

10:0.28

11:0.05

Negative Logits

 slang

-2.00

 retailer

-1.76

 mustard

-1.74

 sailor

-1.71

 Pastebin

-1.71

tro

-1.68

spr

-1.65

 Harvest

-1.62

 purch

-1.61

 camouflage

-1.61

POSITIVE LOGITS

 Gorsuch

2.16

OTUS

2.15

 Scalia

2.09

instein

1.97

ultz

1.90

infeld

1.85

attr

1.82

 secretaries

1.78

otom

1.76

ournal

1.69

Activations Density 0.000%

proper nouns, particularly names and titles related to individuals

No Known Activations