INDEX

Explanations

numbers with additional context like years, counts of people, and measurements

New Auto-Interp

Top Features by Cosine Similarity

Configuration

ckkissane/attn-saes-gpt2-small-all-layers/gpt2-small_L2_Hcat_z_lr1.20e-03_l11.00e+00_ds24576_bs4096_dc1.00e-06_rsanthropic_rie25000_nr4_v4.pt

Prompts (Dashboard)

36,864 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

24,576

Data Type

float32

Hook Name

blocks.2.attn.hook_z

Hook Layer

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Activation Function

relu

Embeds

IFrame

Link

Head Attr Weights

0:0.15

1:0.02

2:0.08

3:0.06

4:0.02

5:0.07

6:0.02

7:0.08

8:0.04

9:0.16

10:0.16

11:0.08

Negative Logits

 adaptations

-1.10

eus

-1.03

 prag

-1.00

ESE

-1.00

 parad

-0.98

 precedence

-0.94

 fate

-0.92

 invari

-0.91

 Selection

-0.90

 protocols

-0.89

POSITIVE LOGITS

pees

1.22

votes

1.19

abytes

1.11

lbs

1.08

renheit

1.03

 acres

1.03

illion

1.02

usercontent

1.02

akh

1.01

 pounds

1.01

Activations Density 0.239%

numbers with additional context like years, counts of people, and measurements

No Known Activations