INDEX

Explanations

No Explanations Found

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

Not in Any Lists

No Comments

Head Attr Weights

0:0.08

1:0.07

2:0.07

3:0.08

4:0.07

5:0.07

6:0.09

7:0.09

8:0.08

9:0.08

10:0.07

11:0.09

Negative Logits

�

-3.16

apore

-2.87

Redditor

-2.81

wcsstore

-2.77

wcs

-2.71

cuts

-2.68

anooga

-2.68

rimp

-2.59

anka

-2.58

 welf

-2.56

POSITIVE LOGITS

 Entity

2.73

 Painter

2.59

 Motorsport

2.51

 Johann

2.44

 Payne

2.44

 preached

2.42

 Religion

2.42

 Carrier

2.41

 Amend

2.37

 Patriot

2.36

Activations Density 0.000%

No Known Activations

This feature has no known activations.