INDEX

Explanations

references to intended or indirect communication

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

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Head Attr Weights

0:0.02

1:0.02

2:0.06

3:0.04

4:0.18

5:0.04

6:0.10

7:0.20

8:0.02

9:0.03

10:0.11

11:0.12

Negative Logits

 resil

-1.35

itol

-1.25

ongevity

-1.22

PG

-1.17

 adolesc

-1.15

EEK

-1.14

wagon

-1.14

alloween

-1.13

 Rebell

-1.13

 mosqu

-1.13

POSITIVE LOGITS

 dict

1.35

 Dictionary

1.28

 nonexistent

1.27

 wrongly

1.25

 "…

1.25

 accus

1.24

 emails

1.24

 "...

1.21

 articles

1.20

 inaccurate

1.20

Activations Density 0.013%

references to intended or indirect communication

No Comments

No Known Activations