INDEX

Explanations

instances of selection and categorization in various contexts

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

Head Attr Weights

0:0.04

1:0.04

2:0.11

3:0.05

4:0.03

5:0.04

6:0.08

7:0.10

8:0.08

9:0.07

10:0.07

11:0.25

Negative Logits

uca

-1.19

 Clown

-1.11

 misinformation

-1.10

 underest

-1.04

 tremend

-1.04

 Lisbon

-1.04

 mistaken

-1.03

 Abdullah

-1.00

 scrimmage

-1.00

 Hiroshima

-0.99

POSITIVE LOGITS

selected

1.43

actions

1.22

 chosen

1.20

 category

1.18

omever

1.17

selection

1.17

 selected

1.14

 categories

1.13

 destination

1.12

 Choose

1.09

Activations Density 0.040%

instances of selection and categorization in various contexts

No Known Activations