INDEX

Explanations

shortened or altered versions of words

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_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

Not in Any Lists

No Comments

Head Attr Weights

0:0.02

1:0.03

2:0.05

3:0.04

4:0.03

5:0.04

6:0.51

7:0.03

8:0.04

9:0.06

10:0.06

11:0.04

Negative Logits

 Fargo

-1.56

jriwal

-1.43

 omission

-1.38

 Petersen

-1.32

abase

-1.30

 Daly

-1.23

��

-1.23

 Estonia

-1.22

 shape

-1.21

 Dane

-1.20

POSITIVE LOGITS

monary

1.70

umatic

1.54

pload

1.49

earances

1.43

lator

1.40

ogram

1.33

orah

1.29

incial

1.28

accompan

1.25

ograms

1.25

Activations Density 0.119%

shortened or altered versions of words

No Comments

No Known Activations