INDEX

Explanations

references to individuals and their actions or impacts

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

0:0.05

1:0.02

2:0.06

3:0.24

4:0.02

5:0.05

6:0.01

7:0.04

8:0.03

9:0.01

10:0.38

11:0.04

Negative Logits

 enhances

-2.66

 lasts

-2.63

 corresponds

-2.51

 denotes

-2.51

 helps

-2.48

 anytime

-2.47

 ensures

-2.43

 exists

-2.40

 achieves

-2.32

 signifies

-2.30

POSITIVE LOGITS

 reluctant

2.75

 reconsider

2.68

 hesitant

2.51

 rethink

2.49

 furious

2.46

 scrambling

2.38

 hastily

2.35

 unsuccessful

2.32

 resignation

2.28

 scorn

2.23

Activations Density 0.076%

references to individuals and their actions or impacts

No Known Activations