INDEX

Explanations

specific phrases and expressions related to subjective opinions and assertions

New Auto-Interp

Top Features by Cosine Similarity

Configuration

ckkissane/attn-saes-gpt2-small-all-layers/gpt2-small_L3_Hcat_z_lr1.20e-03_l19.00e-01_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.3.attn.hook_z

Hook Layer

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Activation Function

relu

Embeds

IFrame

Link

Head Attr Weights

0:0.01

1:0.05

2:0.22

3:0.14

4:0.01

5:0.08

6:0.07

7:0.07

8:0.11

9:0.08

10:0.06

11:0.04

Negative Logits

 Trials

-1.15

ombat

-1.05

slaught

-1.00

 WARRANT

-0.98

©

-0.93

 Challenge

-0.91

 Result

-0.90

earable

-0.87

 Kelvin

-0.87

 Rampage

-0.86

POSITIVE LOGITS

UCT

1.06

 pupp

1.04

alam

1.02

hematic

0.99

rael

0.98

here

0.96

nob

0.94

agog

0.92

esides

0.91

 helicop

0.90

Activations Density 0.122%

specific phrases and expressions related to subjective opinions and assertions

No Known Activations