INDEX

Explanations

terms related to controversial or scandalous content

oai_token-act-pair · gpt-3.5-turbo Triggered by @bot

references to struggling or suffering, particularly through painful emotions or experiences

oai_token-act-pair · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 2-att-kk

Configuration

ckkissane/attn-saes-gpt2-small-all-layers/gpt2-small_L2_Hcat_z_lr1.20e-03_l11.00e+00_ds24576_bs4096_dc1.00e-06_rsanthropic_rie25000_nr4_v4.pt

Prompts (Dashboard)

36,864 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

24,576

Data Type

float32

Hook Name

blocks.2.attn.hook_z

Hook Layer

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Activation Function

relu

Embeds

IFrame

Link

Not in Any Lists

No Comments

Head Attr Weights

0:0.07

1:0.04

2:0.23

3:0.06

4:0.15

5:0.06

6:0.04

7:0.03

8:0.09

9:0.09

10:0.06

11:0.04

Negative Logits

 unden

-1.51

 fict

-1.28

 substance

-1.11

ッド

-1.10

 disclaim

-1.10

 uncont

-1.09

 simultane

-1.08

 nonex

-1.08

 believable

-1.07

 disclaimer

-1.07

POSITIVE LOGITS

ousy

1.31

mony

1.28

feat

1.26

interstitial

1.25

irds

1.21

incial

1.21

favorite

1.18

tty

1.18

ingo

1.17

ails

1.15

Activations Density 0.001%

terms related to controversial or scandalous content

references to struggling or suffering, particularly through painful emotions or experiences

No Comments

No Known Activations