INDEX

Explanations

references to official statements and reports

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

1:0.10

2:0.15

3:0.03

4:0.02

5:0.07

6:0.12

7:0.07

8:0.10

9:0.04

10:0.11

11:0.08

Negative Logits

usterity

-1.20

 wherever

-1.06

 shenan

-1.04

 unlike

-1.02

 interfering

-1.00

 whichever

-0.99

 nicer

-0.98

 gimm

-0.98

 crappy

-0.96

 guiActiveUnfocused

-0.96

POSITIVE LOGITS

ibrarian

1.08

uid

1.07

rb

1.02

ivan

1.01

itud

1.01

encia

1.00

enf

0.99

utive

0.98

 quoted

0.97

bara

0.96

Activations Density 0.100%

references to official statements and reports

No Known Activations