INDEX

Explanations

names of individuals, places, or organizations relevant to historical or political contexts

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

Head Attr Weights

0:0.03

1:0.07

2:0.03

3:0.05

4:0.03

5:0.43

6:0.01

7:0.01

8:0.05

9:0.09

10:0.11

11:0.03

Negative Logits

ーク

-1.94

エル

-1.86

ucket

-1.86

erion

-1.84

�

-1.80

 Pars

-1.76

 Smartstocks

-1.75

 Pesh

-1.74

anguages

-1.73

pse

-1.70

POSITIVE LOGITS

pret

1.73

 governs

1.64

 gimm

1.61

 airs

1.61

 remod

1.55

 govern

1.53

kered

1.50

adem

1.50

 dictated

1.44

 imperialist

1.41

Activations Density 0.663%

names of individuals, places, or organizations relevant to historical or political contexts

No Known Activations