INDEX

Explanations

references to deceptive behavior and questioning the intent behind actions in a narrative context

New Auto-Interp

Top Features by Cosine Similarity

Configuration

ckkissane/attn-saes-gpt2-small-all-layers/gpt2-small_L9_Hcat_z_lr1.20e-03_l11.20e+00_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.9.attn.hook_z

Hook Layer

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Activation Function

relu

Embeds

IFrame

Link

Head Attr Weights

0:0.12

1:0.02

2:0.21

3:0.10

4:0.07

5:0.09

6:0.06

7:0.07

8:0.05

9:0.06

10:0.05

11:0.06

Negative Logits

rentices

-3.02

 Corsair

-3.01

 apprentices

-2.93

MSI

-2.92

 workshop

-2.77

 Yamaha

-2.77

 Mortgage

-2.70

HEL

-2.69

 Helm

-2.64

 militia

-2.62

POSITIVE LOGITS

 bowel

4.51

 digestive

4.47

 feces

4.42

 swallowing

4.38

 digestion

4.29

 gastro

4.19

ur

4.18

 vomit

4.15

 diarrhea

4.13

 intestine

4.06

Activations Density 0.015%

references to deceptive behavior and questioning the intent behind actions in a narrative context

No Known Activations