INDEX

Explanations

references to political figures and historical 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

0:0.04

1:0.02

2:0.12

3:0.10

4:0.09

5:0.08

6:0.03

7:0.03

8:0.07

9:0.11

10:0.13

11:0.13

Negative Logits

rament

-1.19

adena

-1.18

rir

-1.11

lad

-1.11

ritional

-1.08

arie

-1.07

avour

-1.06

izons

-1.05

ritz

-1.05

ancouver

-1.04

POSITIVE LOGITS

 THEN

1.08

 anyway

1.08

Nazis

1.03

の魔

1.00

Wr

0.99

JV

0.99

 blah

0.98

 inverse

0.98

 somehow

0.96

 deleted

0.95

Activations Density 0.225%

references to political figures and historical events

No Known Activations