INDEX

Explanations

references to payment amounts and salary comparisons in various 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

0:0.02

1:0.02

2:0.09

3:0.04

4:0.17

5:0.03

6:0.06

7:0.25

8:0.02

9:0.03

10:0.15

11:0.07

Negative Logits

 causation

-1.61

 fatal

-1.46

 progress

-1.45

 diffusion

-1.45

 rupture

-1.41

 regress

-1.40

 deadliest

-1.40

 countdown

-1.39

 fragmentation

-1.39

crete

-1.37

POSITIVE LOGITS

Laughs

1.57

doctor

1.53

lance

1.53

 babys

1.51

 salaries

1.49

cius

1.47

 Mechdragon

1.47

ESPN

1.46

 handsome

1.43

 Emin

1.40

Activations Density 0.007%

references to payment amounts and salary comparisons in various contexts

No Known Activations