INDEX

Explanations

negatively charged or pejorative terms related to failure or inadequacy

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

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

Not in Any Lists

No Comments

Head Attr Weights

0:0.08

1:0.06

2:0.19

3:0.15

4:0.03

5:0.02

6:0.05

7:0.13

8:0.07

9:0.05

10:0.06

11:0.05

Negative Logits

 Guards

-1.35

 Olympia

-1.23

inus

-1.21

 Marina

-1.13

 Mirage

-1.10

 Swordsman

-1.10

Plex

-1.07

Socket

-1.04

 Maiden

-1.02

Parser

-1.02

POSITIVE LOGITS

 sidx

1.50

 anymore

1.29

 istg

1.20

udden

1.19

estine

1.16

aunder

1.13

externalActionCode

1.12

 guiActiveUn

1.11

arate

1.10

ibaba

1.10

Activations Density 0.088%

negatively charged or pejorative terms related to failure or inadequacy

No Comments

No Known Activations