INDEX

Explanations

strong emotional expressions, particularly those conveying appreciation or love

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

0:0.13

1:0.14

2:0.04

3:0.03

4:0.02

5:0.29

6:0.11

7:0.01

8:0.07

9:0.05

10:0.03

11:0.02

Negative Logits

 embark

-1.64

anmar

-1.58

cel

-1.55

 infiltr

-1.50

Gy

-1.48

 participating

-1.47

 assisted

-1.46

 superv

-1.43

 Chau

-1.42

ppelin

-1.42

POSITIVE LOGITS

ends

1.96

oqu

1.82

rx

1.80

hex

1.74

Reason

1.73

cum

1.69

sts

1.69

require

1.69

ame

1.69

�

1.68

Activations Density 0.023%

strong emotional expressions, particularly those conveying appreciation or love

No Known Activations